| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
Implement the `freshen_object()` callback function for the in-memory
source.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Implement the `count_objects()` callback function for the in-memory
source.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Implement the `find_abbrev_len()` callback function for the in-memory
source.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Implement the `for_each_object()` callback function for the in-memory
source.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The in-memory source stores its objects in a simple array that we grow as
needed. This has a couple of downsides:
- The object lookup is O(n). This doesn't matter in practice because
we only store a small number of objects.
- We don't have an easy way to iterate over all objects in
lexicographic order.
- We don't have an easy way to compute unique object ID prefixes.
Refactor the code to use an oidtree instead. This is the same data
structure used by our loose object source, and thus it means we get a
bunch of functionality for free.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The oidtree data structure is currently only used to store object IDs,
without any associated data. So consequently, it can only really be used
to track which object IDs exist, and we can use the tree structure to
efficiently operate on OID prefixes.
But there are valid use cases where we want to both:
- Store object IDs in a sorted order.
- Associated arbitrary data with them.
Refactor the oidtree interface so that it allows us to store arbitrary
payloads within the respective nodes. This will be used in the next
commit.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The cbtree subsystem allows the user to store arbitrary data in a
prefix-free set of strings. This is used by us to store object IDs in a
way that we can easily iterate through them in lexicograph order, and so
that we can easily perform lookups with shortened object IDs.
In its current form, it is not easily possible to store arbitrary data
with the tree nodes. There are a couple of approaches such a caller
could try to use, but none of them really work:
- One may embed the `struct cb_node` in a custom structure. This does
not work though as `struct cb_node` contains a flex array, and
embedding such a struct in another struct is forbidden.
- One may use a `union` over `struct cb_node` and ones own data type,
which _is_ allowed even if the struct contains a flex array. This
does not work though, as the compiler may align members of the
struct so that the node key would not immediately start where the
flex array starts.
- One may allocate `struct cb_node` such that it has room for both its
key and the custom data. This has the downside though that if the
custom data is itself a pointer to allocated memory, then the leak
checker will not consider the pointer to be alive anymore.
Refactor the cbtree to drop the flex array and instead take in an
explicit offset for where to find the key, which allows the caller to
embed `struct cb_node` is a wrapper struct.
Note that this change has the downside that we now have a bit of padding
in our structure, which grows the size from 60 to 64 bytes on a 64 bit
system. On the other hand though, it allows us to get rid of the memory
copies that we previously had to do to ensure proper alignment. This
seems like a reasonable tradeoff.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Implement the `write_object_stream()` callback function for the in-memory
source.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Implement the `write_object()` callback function for the in-memory
source.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Implement the `read_object_stream()` callback function for the in-memory
source.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Implement the `read_object_info()` callback function for the in-memory
source.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The function `odb_pretend_object()` writes an object into the in-memory
object database source. The effect of this is that the object will now
become readable, but it won't ever be persisted to disk.
Before storing the object, we first verify whether the object already
exists. This is done by calling `odb_has_object()` to check all sources,
followed by `find_cached_object()` to check whether we have already
stored the object in our in-memory source.
This is unnecessary though, as `odb_has_object()` already checks the
in-memory source transitively via:
- `odb_has_object()`
- `odb_read_object_info_extended()`
- `do_oid_object_info_extended()`
- `find_cached_object()`
Drop the explicit call to `find_cached_object()`.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Implement the `free()` callback function for the "in-memory" source.
Note that this requires us to define `struct cached_object_entry` in
"odb/source-inmemory.h", as it is accessed in both "odb.c" and
"odb/source-inmemory.c" now. This will be fixed in subsequent commits
though.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Next to our typical object database sources, each object database also
has an implicit source of "cached" objects. These cached objects only
exist in memory and some use cases:
- They contain evergreen objects that we expect to always exist, like
for example the empty tree.
- They can be used to store temporary objects that we don't want to
persist to disk, which is used by git-blame(1) to create a fake
worktree commit.
Overall, their use is somewhat restricted though. For example, we don't
provide the ability to use it as a temporary object database source that
allows the user to write objects, but discard them after Git exists. So
while these cached objects behave almost like a source, they aren't used
as one.
This is about to change over the following commits, where we will turn
cached objects into a new "in-memory" source. This will allow us to use
it exactly the same as any other source by providing the same common
interface as the "files" source.
For now, the in-memory source only hosts the cached objects and doesn't
provide any logic yet. This will change with subsequent commits, where
we move respective functionality into the source.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
* jt/odb-transaction-write:
odb/transaction: make `write_object_stream()` pluggable
object-file: generalize packfile writes to use odb_write_stream
object-file: avoid fd seekback by checking object size upfront
object-file: remove flags from transaction packfile writes
odb: update `struct odb_write_stream` read() callback
odb/transaction: use pluggable `begin_transaction()`
odb: split `struct odb_transaction` into separate header
|
|
How an ODB transaction handles writing objects is expected to vary
between implementations. Introduce a new `write_object_stream()`
callback in `struct odb_transaction` to make this function pluggable.
Rename `index_blob_packfile_transaction()` to
`odb_transaction_files_write_object_stream()` and wire it up for use
with `struct odb_transaction_files` accordingly.
Signed-off-by: Justin Tobler <jltobler@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The `index_blob_packfile_transaction()` function streams blob data
directly from an fd. This makes it difficult to reuse as part of a
generic transactional object writing interface.
Refactor the packfile write path to operate on a `struct
odb_write_stream`, allowing callers to supply data from arbitrary
sources.
Signed-off-by: Justin Tobler <jltobler@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
In certain scenarios, Git handles writing blobs that exceed
"core.bigFileThreshold" differently by streaming the object directly
into a packfile. When there is an active ODB transaction, these blobs
are streamed to the same packfile instead of using a separate packfile
for each. If "pack.packSizeLimit" is configured and streaming another
object causes the packfile to exceed the configured limit, the packfile
is truncated back to the previous object and the object write is
restarted in a new packfile.
This works fine, but requires the fd being read from to save a
checkpoint so it becomes possible to rewind the input source via seeking
back to a known offset at the beginning. In a subsequent commit, blob
streaming is converted to use `struct odb_write_stream` as a more
generic input source instead of an fd which doesn't provide a mechanism
for rewinding.
For this use case though, rewinding the fd is not strictly necessary
because the inflated size of the object is known and can be used to
approximate whether writing the object would cause the packfile to
exceed the configured limit prior to writing anything. These blobs
written to the packfile are never deltified thus the size difference
between what is written versus the inflated size is due to zlib
compression. While this does prevent packfiles from being filled to the
potential maximum is some cases, it should be good enough and still
prevents the packfile from exceeding any configured limit.
Use the inflated blob size to determine whether writing an object to a
packfile will exceed the configured "pack.packSizeLimit".
Signed-off-by: Justin Tobler <jltobler@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The `index_blob_packfile_transaction()` function handles streaming a
blob from an fd to compute its object ID and conditionally writes the
object directly to a packfile if the INDEX_WRITE_OBJECT flag is set. A
subsequent commit will make these packfile object writes part of the
transaction interface. Consequently, having the object write be
conditional on this flag is a bit awkward.
In preparation for this change, introduce a dedicated
`hash_blob_stream()` helper that only computes the OID from a `struct
odb_write_stream`. This is invoked by `index_fd()` instead when the
INDEX_WRITE_OBJECT is not set. The object write performed via
`index_blob_packfile_transaction()` is made unconditional accordingly.
Signed-off-by: Justin Tobler <jltobler@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The `read()` callback used by `struct odb_write_stream` currently
returns a pointer to an internal buffer along with the number of bytes
read. This makes buffer ownership unclear and provides no way to report
errors.
Update the interface to instead require the caller to provide a buffer,
and have the callback return the number of bytes written to it or a
negative value on error. While at it, also move the `struct
odb_write_stream` definition to "odb/streaming.h". Call sites are
updated accordingly.
Signed-off-by: Justin Tobler <jltobler@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Each ODB source is expected to provide an ODB transaction implementation
that should be used when starting a transaction. With d6fc6fe6f8
(odb/source: make `begin_transaction()` function pluggable, 2026-03-05),
the `struct odb_source` now provides a pluggable callback for beginning
transactions. Use the callback provided by the ODB source accordingly.
Signed-off-by: Justin Tobler <jltobler@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The current ODB transaction interface is colocated with other ODB
interfaces in "odb.{c,h}". Subsequent commits will expand `struct
odb_transaction` to support write operations on the transaction
directly. To keep things organized and prevent "odb.{c,h}" from becoming
more unwieldy, split out `struct odb_transaction` into a separate
header.
Signed-off-by: Justin Tobler <jltobler@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
limit_list() maintains a date-sorted work queue of commits using a
linked list with commit_list_insert_by_date() for insertion. Each
insertion walks the list to find the right position — O(n) per insert.
In repositories with merge-heavy histories, the symmetric difference
can contain thousands of commits, making this O(n) insertion the
dominant cost.
Replace the sorted linked list with a prio_queue (binary heap). This
gives O(log n) insertion and O(log n) extraction instead of O(n)
insertion and O(1) extraction, which is a net win when the queue is
large.
The still_interesting() and everybody_uninteresting() helpers are
updated to scan the prio_queue's contiguous array instead of walking a
linked list. process_parents() already accepts both a commit_list and
a prio_queue parameter, so the change in limit_list() simply switches
which one is passed.
Benchmark: git rev-list --left-right --count HEAD~N...HEAD
Repository: 2.3M commits, merge-heavy DAG (monorepo)
Best of 5 runs, times in seconds:
commits in
symmetric diff baseline patched speedup
-------------- -------- ------- -------
10 0.01 0.01 1.0x
50 0.01 0.01 1.0x
3751 21.23 8.49 2.5x
4524 21.70 8.29 2.6x
10130 20.10 6.65 3.0x
No change for small traversals; 2.5-3.0x faster when the queue grows
to thousands of commits.
Signed-off-by: Kristofer Karlsson <krka@spotify.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
In partial clones, `git grep` fetches necessary blobs on-demand one
at a time, which can be very slow. In partial clones, add an extra
preliminary walk over the tree similar to grep_tree() which collects
the blobs of interest, and then prefetches them.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
In partial clones, `git cherry` fetches necessary blobs on-demand one
at a time, which can be very slow. We would like to prefetch all
necessary blobs upfront. To do so, we need to be able to first figure
out which blobs are needed.
`git cherry` does its work in a two-phase approach: first computing
header-only IDs (based on file paths and modes), then falling back to
full content-based IDs only when header-only IDs collide -- or, more
accurately, whenever the oidhash() of the header-only object_ids
collide.
patch-ids.c handles this by creating an ids->patches hashmap that has
all the data we need, but the problem is that any attempt to query the
hashmap will invoke the patch_id_neq() function on any colliding objects,
which causes the on-demand fetching.
Insert a new prefetch_cherry_blobs() function before checking for
collisions. Use a temporary replacement on the ids->patches.cmpfn
in order to enumerate the blobs that would be needed without yet
fetching them, and then fetch them all at once, then restore the old
ids->patches.cmpfn.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
promisor_remote_get_direct() does not, on its happy path, filter out
OIDs that are already present in the local object store: every OID
the caller supplies is written to the fetch subprocess's stdin and
ends up in the response pack. The only filtering it performs is in
remove_fetched_oids(), and that only runs after a fetch failure when
falling back to a different configured promisor remote.
Almost every existing caller already filters locally-present OIDs out
itself (typically with odb_read_object_info_extended() and
OBJECT_INFO_FOR_PREFETCH, or odb_has_object() with no fetch flag). But
the existing API comment does not state this expectation, so a new
caller is easy to write incorrectly (I missed this originally and wrote
two problematic callers). Omitting the filter still "works" in the
sense that the desired objects end up local, but it silently makes the
fetch request -- and the response pack -- larger than necessary,
defeating part of the point of batching.
Spell the contract out so future callers know to filter (and
deduplicate) themselves, and point them at the helpers they should
use to check local presence without accidentally triggering a lazy
fetch.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
git rev-parse --bisect does not work when alternate bisect terms are
used, simply listing no revisions at all.
This is because a such bisect using e.g. "old" and "new" in place of
"good" and "bad" will name refs "refs/bisect/old" (or new) accordingly
so the hardcoded "refs/bisect/bad" (and good) yields no results in a
bisect using alternate terms.
Use the current bisect_terms to make rev-parse --bisect work in an
alternate term bisect.
Signed-off-by: Jonas Rebmann <kernel@schlaraffenlan.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
As bisect terms can be arbitrarily chosen, they have been quoted in some
status messages, and in even more by translators.
To make the role of bisect terms more clear, including in translations,
and for consistency, 'enquote' all occurrences of bisect terms in status
messages.
Signed-off-by: Jonas Rebmann <kernel@schlaraffenlan.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Alternate bisect terms are helpful when the terms "good" and "bad" are
confusing such as when bisecting for the resolution of an issue (the
first good commit) rather than the introduction of a regression.
These terms must be used when marking a commit (e.g. `git bisect new`),
they will be used in reference names (e.g. refs/bisect/new) and they are
used in parts of git's log output such as "<sha> was both old and new"
in git bisect skip's output.
However, hardcoded "good"/"bad" terms are still used in a few status
messages and can cause confusion about the status of the bisect such as:
$ git bisect old
[sha] is the first new commit
or about the required action such as:
status: waiting for bad commit, 1 good commit known
$ git bisect bad
error: Invalid command: you're currently in a new/old bisect
fatal: unknown command: 'bad'
This commit updates all remaining output messages which use hardcoded
"good" and "bad" terms to use the selected terms consistently across the
bisect output and adds tests.
Signed-off-by: Jonas Rebmann <kernel@schlaraffenlan.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Add a function for adding the full hexadecimal hash value of an object
ID to a strbuf. It's thread-safe and slightly more efficient than using
strbuf_addstr() with oid_to_hex() because it doesn't have to determine
the length of the string or copy it from the intermediate static buffer.
Add and apply a semantic patch to use it throughout the code base.
I get a tiny speedup for git log showing a single hash per commit:
Benchmark 1: ./git_main log --format=%H
Time (mean ± σ): 91.2 ms ± 0.7 ms [User: 51.9 ms, System: 38.6 ms]
Range (min … max): 89.8 ms … 92.6 ms 31 runs
Benchmark 2: ./git log --format=%H
Time (mean ± σ): 90.5 ms ± 0.7 ms [User: 51.0 ms, System: 38.8 ms]
Range (min … max): 89.2 ms … 92.3 ms 32 runs
Summary
./git log --format=%H ran
1.01 ± 0.01 times faster than ./git_main log --format=%H
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The "gc.auto" configuration has traditionally been used to turn off
running git-gc(1) as part of our auto-maintenance. We have eventually
switched over to git-maintenance(1) in a95ce12430 (maintenance: replace
run_auto_gc(), 2020-09-17), and with 1942d48380 (maintenance: optionally
skip --auto process, 2020-08-28) we have introduced "maintenance.auto"
to control whether or not to run auto-maintenance.
At that point though we still shelled out to git-gc(1) internally. So
if "gc.auto=0" was set we would still _execute_ git-maintenance(1), but
the command would have exited fast because git-gc(1) itself knew to
honor the config key.
This has recently changed though, as we have adapted the default
maintenance strategy to not use git-gc(1) anymore. The consequence is
that "gc.auto=0" doesn't have an effect anymore, which is a somewhat
surprising change in behaviour for our users.
Adapt `run_auto_maintenance()` so that it knows to also read "gc.auto",
similar to how it also reads both "maintenance.autoDetach" and
"gc.autoDetach".
Reported-by: Jean-Christophe Manciot <actionmystique@gmail.com>
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
When running git-maintenance(1), we create a lockfile that is supposed
to keep other maintenance processes from running at the same time. This
lockfile is broken though in case the "--detach" flag is passed: the
lockfile is created by the parent process and will be cleaned up either
manually or on exit. But when detaching, the parent will exit before all
of the background maintenance tasks have been run, and consequently the
lock only covers a smaller part of the whole maintenance process.
Fix this bug by reassigning all tempfiles from the parent process to the
child process when daemonizing so that it becomes the responsibility of
the child to clean them up.
Note that this is a broader fix, as we now always reassign tempfiles
when daemonizing. This is a natural consequence of the semantics of
`daemonize()` though, as it essentially promises to continue running the
current process in the background. It is thus sensible to have that
function perform the whole dance of assigning resources to the child
process, including tempfiles.
There's only a single other caller in "daemon.c", but that process
doesn't create any tempfiles before the call to `daemonize()` and is
thus not impacted by this change.
Reported-by: Jean-Christophe Manciot <actionmystique@gmail.com>
Helped-by: Jeff King <peff@peff.net>
Helped-by: Derrick Stolee <stolee@gmail.com>
Co-authored-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Headers from glibc 2.43 when used with clang does not allow
disabling C11 language features, causing build failures..
* ps/clang-w-glibc-2.43-and-_Generic:
build: tolerate use of _Generic from glibc 2.43 with Clang
|
|
gitignore(5) says that the per-repository ignore file is
$GIT_DIR/info/exclude, but in a worktree that is not the case:
git rev-parse --git-path info/exclude
/path/to/main/worktree/.git/info/exclude
git rev-parse --git-common-dir
/path/to/main/worktree/.git
We actually use $GIT_COMMON_DIR/info/exclude. Adjust the documentation
and some code comments to say so.
Signed-off-by: D. Ben Knoble <ben.knoble+github@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The documentation says the flag prevents an option from being
"negated" without specifying what that means. Add a parenthetical
to clarify that it rejects the "--no-<option>" form.
Signed-off-by: Michael Montalbo <mmontalbo@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The xdemitconf_t fields ctxlen and interhunkctxlen are typed as long
(signed), but negative values are not meaningful for context line
counts. Unlike the diff_options fields changed in the previous two
commits, these cannot be converted to unsigned because the xdiff
arithmetic relies on signed subtraction:
s1 = XDL_MAX(xch->i1 - xecfg->ctxlen, 0);
If ctxlen were unsigned long, the signed operand would be implicitly
converted to unsigned, and the subtraction would wrap to a large
positive value when i1 < ctxlen, defeating the XDL_MAX clamp. The
signed type is required for correct context-window calculations.
The previous two commits reject negative values at the parse layer
for --inter-hunk-context and -U/--unified, so negative values should
no longer reach xdiff in normal use. Add BUG() guards at the top of
xdl_get_hunk() as defense in depth to catch programming errors in
current or future callers that bypass option parsing.
xdl_get_hunk() is called by both xdl_emit_diff() and
xdl_call_hunk_func(), so a single guard covers all xdiff consumers.
Signed-off-by: Michael Montalbo <mmontalbo@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Passing a negative value to -U is silently accepted and produces
corrupt unified diff output with malformed hunk headers:
$ git log -1 -p -U-500 -- GIT-VERSION-GEN | grep '^@@'
@@ -503,999- +503,999- @@
Line 503 of a 106-line file, count "999-" is not a valid integer.
The config variable diff.context already rejects negative values, but
the command line callback diff_opt_unified() uses strtol() with no
range check.
Change the type of diff_options.context and its static default from
int to unsigned int, matching the change to interhunkcontext in the
previous commit. The type change requires reworking the callback and
config parsing to validate in a local variable before assigning to
the now-unsigned field.
Unlike --inter-hunk-context which could be converted to OPT_UNSIGNED,
-U needs OPT_CALLBACK_F for PARSE_OPT_OPTARG (bare -U with no value
enables patch output). Add a range check in the callback instead.
Signed-off-by: Michael Montalbo <mmontalbo@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Negative values for --inter-hunk-context produce structurally invalid
diff output with overlapping hunks:
$ git log -1 -p -U3 --inter-hunk-context=-100 791aeddfa2 \
-- git-compat-util.h | grep '^@@'
@@ -110,6 +110,9 @@
@@ -115,6 +118,9 @@
@@ -116,6 +122,7 @@
Hunk 1 covers lines 110-115, hunk 2 starts at 115 (overlap), hunk 3
starts at 116 (overlaps both). The resulting patch cannot be applied.
The config variable diff.interHunkContext already rejects negative
values, but the command line option does not.
Change the type of diff_options.interhunkcontext and its static
default from int to unsigned int, and switch the option parser from
OPT_INTEGER_F to OPT_UNSIGNED. This rejects negative values at parse
time via git_parse_unsigned() and enforces the correct type at compile
time via BARF_UNLESS_UNSIGNED.
Signed-off-by: Michael Montalbo <mmontalbo@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
When a dumb http server reports alternates with an absolute path, we try
to paste that onto the root of the URL we're trying to fetch from. So if
we go to "http://example.com/path/to/child.git" and it tells us about an
alternate at "/parent.git", we'll hit "http://example.com/parent.git".
But there's a bug in computing the base when the URL does not have any
path component at all, like "http://example.com". When looking for the
first slash after the host, strchr() returns NULL, and we compute a
nonsense value for the length of the host portion. And then when we use
that length to copy the base of the URL into a strbuf, we're likely to
fail.
The security implications are minimal here. We store the nonsense length
("serverlen") as an int, so on a 64-bit system it may effectively be
anything (it is zero minus a 64-bit heap pointer, then truncated to
32-bits and stuffed into a signed value). When we feed that length to
strbuf_add(), it is cast into a size_t and one of four things will
happen:
1. If serverlen was negative, it will turn into a very large positive
value and strbuf_add() will fail to allocate, ending the program.
Ditto if serverlen was positive but just very large.
This doesn't really get an attacker anything; the victim will just
fail to clone their evil repo.
2. If serverlen was small enough, we'll successfully extend the target
strbuf, and then copy an arbitrary set of bytes from "base". And
then one of these is true:
a. That set of bytes is much larger than the length of the "base"
string. This is an out-of-bounds read, but there's no
out-of-bounds write, since the strbuf code both allocates and
copies using the same size_t. This is likely to cause a
segfault as we try to read unmapped pages of memory.
b. Like (2a), but if the set of bytes is small enough we might
not segfault. We might read random memory from the process and
copy it into the "target" strbuf.
What happens then? We know that "base" ends with a NUL
terminator, which will be copied into "target" as well. So
even though target.len might be 1000 bytes (or whatever), when
interpreted as a NUL-terminated string, target.buf is still
the exact same string as "base".
And that's all we ever do with target: pass it around as a C
string, and then eventually strbuf_detach() it to become a C
string. So even though there was arbitrary memory copied into
the strbuf, we never access it.
c. The other interesting case is when serverlen is actually
_shorter_ than the length of base. And there we truncate the
string. Probably in a way that makes it totally invalid, but
if you were very unlucky you could turn something like:
http://victim.com.evil.domain:8000
into:
http://victim.com
Which looks like the start of a redirect attack, except that
the attacker could just have written "http://victim.com" in
the first place! Either way we feed it to
is_alternate_allowed(), which is where we check redirect and
protocol rules.
I think we can just treat this like a regular bug.
And it's quite a weird setup in the first place, as it implies that the
root of the web server is serving a repository (i.e., that you can get
something useful from "http://example.com/info/refs"). The bug has been
there since b3661567cf ([PATCH] Add support for alternates in HTTP,
2005-09-14) without anybody noticing.
I kind of doubt anybody really cares about making this work, but it's
easy enough to do so: the host-portion of the URL ends at either the
first slash or the end-of-string. So we can just replace strchr() with
strchrnul().
The test setup is a little gross, as we take over the httpd document
root by shoving our bare-repo components into it. But it demonstrates
the problem and shows that our solution actually allows the alternate to
function, if the server is configured to allow it.
Reported-by: slonkazoid <slonkazoid@slonk.ing>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
At the top of add_rfc2047() we do this:
strbuf_grow(sb, len * 3 + strlen(encoding) + 100);
where "len" is the size of the header (like an author name) we are about
to encode into the buffer. This pre-sizing is purely an optimization; we
use strbuf_addf() and friends to actually write into the buffer, and
they will grow the buffer as necessary.
But there's a problem with the code above: the input can be arbitrarily
large, so we might overflow a size_t while doing that computation,
ending up with a too-small allocation request. Overflowing requires an
impractically large input on a 64-bit system, but is easy to demonstrate
on a 32-bit system with a commit whose author name is ~1.4GB.
Because this pre-sizing is just an optimization, there's no real harm.
We'll start with a smaller buffer and grow it as necessary. But it
_looks_ like a vulnerability, since some other code may pre-size a
strbuf and then write directly into its buffer. So it's worth avoiding
the overflow in the first place.
The obvious way to do that is via checked operations like st_add() and
friends. But taking a step back, is this pre-sizing actually helping
anything?
The computation goes all the way back to 4234a76167 (Extend
--pretty=oneline to cover the first paragraph,, 2007-06-11), but back
then we really were sizing the array to write into directly! In
674d172730 (Rework pretty_print_commit to use strbufs instead of custom
buffers., 2007-09-10) that switched to a strbuf, and at that point it
was a pure optimization.
Is the optimization helping? I don't think so. Even for a gigantic case
like the 1.4GB author name, I couldn't measure any slowdown when
removing it. And most input will be much smaller, and added to a running
strbuf containing the rest of the email-header output. We can just rely
on strbuf's usual amortized-linear growth.
So deleting the line seems like the best way to go. It eliminates the
integer overflow and makes the code a tiny bit simpler.
Reported-by: Luke Martin <lmartin@paramenoeng.com>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Speed up printing of objectsize values by using the specialized function
strbuf_add_uint() as well as strbuf_insert() for padding instead of the
general-purpose function strbuf_addf(). Here are the numbers I get when
listing objects in the Linux kernel repo:
Benchmark 1: ./git_main -C ../linux ls-tree -r --format='%(objectsize)' HEAD
Time (mean ± σ): 294.4 ms ± 0.4 ms [User: 231.5 ms, System: 59.4 ms]
Range (min … max): 293.9 ms … 295.0 ms 10 runs
Benchmark 2: ./git -C ../linux ls-tree -r --format='%(objectsize)' HEAD
Time (mean ± σ): 291.2 ms ± 0.4 ms [User: 227.9 ms, System: 62.1 ms]
Range (min … max): 290.6 ms … 292.0 ms 10 runs
Benchmark 3: ./git_main -C ../linux ls-tree -r --format='%(objectsize:padded)' HEAD
Time (mean ± σ): 295.3 ms ± 0.6 ms [User: 232.0 ms, System: 59.6 ms]
Range (min … max): 294.3 ms … 296.3 ms 10 runs
Benchmark 4: ./git -C ../linux ls-tree -r --format='%(objectsize:padded)' HEAD
Time (mean ± σ): 291.9 ms ± 0.4 ms [User: 228.5 ms, System: 61.5 ms]
Range (min … max): 291.2 ms … 292.3 ms 10 runs
Summary
./git -C ../linux ls-tree -r --format='%(objectsize)' HEAD ran
1.00 ± 0.00 times faster than ./git -C ../linux ls-tree -r --format='%(objectsize:padded)' HEAD
1.01 ± 0.00 times faster than ./git_main -C ../linux ls-tree -r --format='%(objectsize)' HEAD
1.01 ± 0.00 times faster than ./git_main -C ../linux ls-tree -r --format='%(objectsize:padded)' HEAD
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Speed up printing of objectsize values by using the specialized function
strbuf_add_uint() as well as strbuf_insert() for padding instead of the
general-purpose function strbuf_addf(). Here are the numbers I get when
listing files in the Linux kernel repo:
Benchmark 1: ./git_main -C ../linux ls-files --format='%(objectsize)'
Time (mean ± σ): 257.3 ms ± 0.4 ms [User: 197.4 ms, System: 56.7 ms]
Range (min … max): 256.7 ms … 258.1 ms 11 runs
Benchmark 2: ./git -C ../linux ls-files --format='%(objectsize)'
Time (mean ± σ): 253.4 ms ± 0.3 ms [User: 193.6 ms, System: 56.6 ms]
Range (min … max): 253.0 ms … 253.8 ms 11 runs
Benchmark 3: ./git_main -C ../linux ls-files --format='%(objectsize:padded)'
Time (mean ± σ): 257.9 ms ± 0.3 ms [User: 198.0 ms, System: 56.6 ms]
Range (min … max): 257.3 ms … 258.5 ms 11 runs
Benchmark 4: ./git -C ../linux ls-files --format='%(objectsize:padded)'
Time (mean ± σ): 254.6 ms ± 1.0 ms [User: 194.6 ms, System: 56.7 ms]
Range (min … max): 253.7 ms … 256.8 ms 11 runs
Summary
./git -C ../linux ls-files --format='%(objectsize)' ran
1.00 ± 0.00 times faster than ./git -C ../linux ls-files --format='%(objectsize:padded)'
1.02 ± 0.00 times faster than ./git_main -C ../linux ls-files --format='%(objectsize)'
1.02 ± 0.00 times faster than ./git_main -C ../linux ls-files --format='%(objectsize:padded)'
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Speed up printing of objectsize atoms by using the specialized function
strbuf_add_uint() instead of the general-purpose function strbuf_addf():
Benchmark 1: ./git_main cat-file --batch-all-objects --batch-check='%(objectsize)'
Time (mean ± σ): 751.7 ms ± 1.5 ms [User: 733.5 ms, System: 17.1 ms]
Range (min … max): 750.5 ms … 755.0 ms 10 runs
Benchmark 2: ./git cat-file --batch-all-objects --batch-check='%(objectsize)'
Time (mean ± σ): 720.4 ms ± 0.4 ms [User: 701.9 ms, System: 16.7 ms]
Range (min … max): 719.7 ms … 721.2 ms 10 runs
Summary
./git cat-file --batch-all-objects --batch-check='%(objectsize)' ran
1.04 ± 0.00 times faster than ./git_main cat-file --batch-all-objects --batch-check='%(objectsize)'
Benchmark 1: ./git_main cat-file --batch-all-objects --batch-check='%(objectsize:disk)'
Time (mean ± σ): 404.6 ms ± 0.9 ms [User: 397.8 ms, System: 5.7 ms]
Range (min … max): 403.3 ms … 405.9 ms 10 runs
Benchmark 2: ./git cat-file --batch-all-objects --batch-check='%(objectsize:disk)'
Time (mean ± σ): 378.3 ms ± 0.9 ms [User: 371.2 ms, System: 5.9 ms]
Range (min … max): 376.8 ms … 380.2 ms 10 runs
Summary
./git cat-file --batch-all-objects --batch-check='%(objectsize:disk)' ran
1.07 ± 0.00 times faster than ./git_main cat-file --batch-all-objects --batch-check='%(objectsize:disk)'
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
strbuf_addf() calls vsnprintf(3) underneath, which supports a plethora
of formatting options. We can avoid its overhead in basic cases by
providing specialized functions like strbuf_addstr() for strings. Add
another one, strbuf_add_uint(), for unsigned integers.
Prepare the number string in a temporary buffer. Make it big enough for
any unsigned integer value: A decimal digit can represent ln(10)/ln(2) ≈
3.32 bits; dividing the number of bits of uintmax_t by 3.3 and rounding
up gives a sufficiently close conservative size estimate.
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The octopus merge path checks whether each remote head is already
an ancestor of HEAD by computing all merge-bases via
repo_get_merge_bases() and comparing the first result's OID to
the remote head. This is more expensive than necessary:
repo_get_merge_bases() calls paint_down_to_common() with
min_generation=0, performs the full STALE drain, and may run
remove_redundant(), when all we need is a yes/no reachability
answer.
Replace this with repo_in_merge_bases(), which answers the
is-ancestor question directly. When generation numbers are
available, repo_in_merge_bases() uses can_all_from_reach() -- a
DFS bounded by generation number that stops as soon as the target
is found or ruled out, without entering paint_down_to_common() at
all. Without generation numbers, it still benefits from a tighter
min_generation floor.
Signed-off-by: Kristofer Karlsson <krka@spotify.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
To help Windows 10 installations, avoid removing files whose
contents are still mmap()'ed.
* js/maintenance-fix-deadlock-on-win10:
maintenance(geometric): do release the `.idx` files before repacking
mingw: optionally use legacy (non-POSIX) delete semantics
|
|
Test fix.
* jc/t5551-fix-expensive:
t5551: "GIT_TEST_LONG=Yes make test" is broken
|
|
Avoid hitting the pathname limit for socks proxy socket during the
test..
* js/t5564-socks-use-short-path:
t5564: use a short path for the SOCKS proxy socket
|
