When we lock a reference like "foo" we need to handle the case where
"foo" exists, but is an empty directory. That's what this code added
in bc7127ef0f (ref locking: allow 'foo' when 'foo/bar' used to exist
but not anymore., 2006-09-30) seems like it should be dealing with.
Except it doesn't, and we never take this branch. The reason is that
when bc7127ef0f was written this looked like:
ref = resolve_ref([...]);
if (!ref && errno == EISDIR) {
[...]
And in resolve_ref() we had this code:
fd = open(path, O_RDONLY);
if (fd < 0)
return NULL;
I.e. we would attempt to read "foo" with open(), which would fail with
EISDIR and we'd return NULL. We'd then take this branch, call
remove_empty_directories() and continue.
Since a1c1d8170d (refs_resolve_ref_unsafe: handle d/f conflicts for
writes, 2017-10-06) we don't. E.g. in the case of
files_copy_or_rename_ref() our callstack will look something like:
[...] ->
files_copy_or_rename_ref() ->
lock_ref_oid_basic() ->
refs_resolve_ref_unsafe()
At that point the first (now only) refs_resolve_ref_unsafe() call in
lock_ref_oid_basic() would do the equivalent of this in the resulting
call to refs_read_raw_ref() in refs_resolve_ref_unsafe():
/* Via refs_read_raw_ref() */
fd = open(path, O_RDONLY);
if (fd < 0)
/* get errno == EISDIR */
/* later, in refs_resolve_ref_unsafe() */
if ([...] && errno != EISDIR)
return NULL;
[...]
/* returns the refs/heads/foo to the caller, even though it's a directory */
return refname;
I.e. even though we got an "errno == EISDIR" we won't take this
branch, since in cases of EISDIR "resolved" is always
non-NULL. I.e. we pretend at this point as though everything's OK and
there is no "foo" directory.
We then proceed with the entire ref update and don't call
remove_empty_directories() until we call commit_ref_update(). See
5387c0d883 (commit_ref(): if there is an empty dir in the way, delete
it, 2016-05-05) for the addition of that code, and
a1c1d8170d (refs_resolve_ref_unsafe: handle d/f conflicts for writes,
2017-10-06) for the commit that changed the original codepath added in
bc7127ef0f to use this "EISDIR" handling.
Further historical commentary:
Before the two preceding commits the caller in files_reflog_expire()
was the only one out of our 4 callers that would pass non-NULL as an
oid. We would then set a (now gone) "resolve_flags" to
"RESOLVE_REF_READING" and just before that "errno != EISDIR" check do:
if (resolve_flags & RESOLVE_REF_READING)
return NULL;
There may have been some case where this ended up mattering and we
couldn't safely make this change before we removed the "oid"
parameter, but I don't think there was, see [1] for some discussion on
that.
In any case, now that we've removed the "oid" parameter in a preceding
commit we can be sure that this code is redundant, so let's remove it.
1. http://lore.kernel.org/git/871r801yp6.fsf@evledraar.gmail.com
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In the preceding commit the last caller that passed a non-NULL OID was
changed to pass NULL to lock_ref_oid_basic(). As noted in preceding
commits use of this API has been going away (we should use ref
transactions, or lock_raw_ref()), so we're unlikely to gain new
callers that want to pass the "oid".
So let's remove it, doing so means we can remove the "mustexist"
condition, and therefore anything except the "flags =
RESOLVE_REF_NO_RECURSE" case.
Furthermore, since the verify_lock() function we called did most of
its work when the "oid" was passed (as "old_oid") we can inline the
trivial part of it that remains in its only remaining caller. Without
a NULL "oid" passed it was equivalent to calling refs_read_ref_full()
followed by oidclr().
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since the the preceding commit the "oid" parameter to reflog_expire()
is always NULL, but it was not cleaned up to reduce the size of the
diff. Let's do that subsequent API and documentation cleanup now.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
During reflog expiry, the cmd_reflog_expire() function first iterates
over all reflogs in logs/*, and then one-by-one acquires the lock for
each one and expires it. This behavior has been with us since this
command was implemented in 4264dc15e1 ("git reflog expire",
2006-12-19).
Change this to stop calling lock_ref_oid_basic() with the OID we saw
when we looped over the logs, instead have it pass the OID it managed
to lock.
This mostly mitigates a race condition where e.g. "git gc" will fail
in a concurrently updated repository because the branch moved since
"git reflog expire --all" was started. I.e. with:
error: cannot lock ref '<refname>': ref '<refname>' is at <OID-A> but expected <OID-B>
This behavior of passing in an "oid" was needed for an edge-case that
I've untangled in this and preceding commits though, namely that we
needed this OID because we'd:
1. Lookup the reflog name/OID via dwim_log()
2. With that OID, lock the reflog
3. Later in builtin/reflog.c we use the OID we looked as input to
lookup_commit_reference_gently(), assured that it's equal to the
OID we got from dwim_log().
We can be sure that this change is safe to make because between
dwim_log (step #1) and lock_ref_oid_basic (step #2) there was no other
logic relevant to the OID or expiry run in the cmd_reflog_expire()
caller.
We can thus treat that code as a black box, before and after this
change it would get an OID that's been locked, the only difference is
that now we mostly won't be failing to get the lock due to the TOCTOU
race[0]. That failure was purely an implementation detail in how the
"current OID" was looked up, it was divorced from the locking
mechanism.
What do we mean with "mostly"? It mostly mitigates it because we'll
still run into cases where the ref is locked and being updated as we
want to expire it, and other git processes wanting to update the refs
will in turn race with us as we expire the reflog.
That remaining race can in turn be mitigated with the
core.filesRefLockTimeout setting, see 4ff0f01cb7 ("refs: retry
acquiring reference locks for 100ms", 2017-08-21). In practice if that
value is high enough we'll probably never have ref updates or reflog
expiry failing, since the clients involved will retry for far longer
than the time any of those operations could take.
See [1] for an initial report of how this impacted "git gc" and a
large discussion about this change in early 2019. In particular patch
looked good to Michael Haggerty, see his[2]. That message seems to not
have made it to the ML archive, its content is quoted in full in my
[3].
I'm leaving behind now-unused code the refs API etc. that takes the
now-NULL "unused_oid" argument, and other code that can be simplified now
that we never have on OID in that context, that'll be cleaned up in
subsequent commits, but for now let's narrowly focus on fixing the
"git gc" issue. As the modified assert() shows we always pass a NULL
oid to reflog_expire() now.
Unfortunately this sort of probabilistic contention is hard to turn
into a test. I've tested this by running the following three subshells
in concurrent terminals:
(
rm -rf /tmp/git &&
git init /tmp/git &&
while true
do
head -c 10 /dev/urandom | hexdump >/tmp/git/out &&
git -C /tmp/git add out &&
git -C /tmp/git commit -m"out"
done
)
(
rm -rf /tmp/git-clone &&
git clone file:///tmp/git /tmp/git-clone &&
while git -C /tmp/git-clone pull
do
date
done
)
(
while git -C /tmp/git-clone reflog expire --all
do
date
done
)
Before this change the "reflog expire" would fail really quickly with
the "but expected" error noted above.
After this change both the "pull" and "reflog expire" will run for a
while, but eventually fail because I get unlucky with
core.filesRefLockTimeout (the "reflog expire" is in a really tight
loop). As noted above that can in turn be mitigated with higher values
of core.filesRefLockTimeout than the 100ms default.
As noted in the commentary added in the preceding commit there's also
the case of branches being racily deleted, that can be tested by
adding this to the above:
(
while git -C /tmp/git-clone branch topic master &&
git -C /tmp/git-clone branch -D topic
do
date
done
)
With core.filesRefLockTimeout set to 10 seconds (it can probably be a
lot lower) I managed to run all four of these concurrently for about
an hour, and accumulated ~125k commits, auto-gc's and all, and didn't
have a single failure. The loops visibly stall while waiting for the
lock, but that's expected and desired behavior.
0. https://en.wikipedia.org/wiki/Time-of-check_to_time-of-use
1. https://lore.kernel.org/git/87tvg7brlm.fsf@evledraar.gmail.com/
2. http://lore.kernel.org/git/b870a17d-2103-41b8-3cbc-7389d5fff33a@alum.mit.edu
3. https://lore.kernel.org/git/87pnqkco8v.fsf@evledraar.gmail.com/
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add a comment about why it is that we need to check for the the
existence of a reflog we're deleting after we've successfully acquired
the lock in files_reflog_expire(). As noted in [1] the lock protocol
for reflogs is somewhat intuitive.
This early exit code the comment applies to dates all the way back to
4264dc15e1 (git reflog expire, 2006-12-19).
1. https://lore.kernel.org/git/54DCDA42.2060800@alum.mit.edu/
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Re-indent this argument list that's been mis-indented since it was
added in 34c319970d (refs/debug: trace into reflog expiry too,
2021-04-23). This makes a subsequent change smaller.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Remove the unused "skip" parameter to lock_raw_ref(), it was never
used. We do use it when passing "skip" to the
refs_rename_ref_available() function in files_copy_or_rename_ref(),
but not here.
This is part of a larger series that modifies lock_ref_oid_basic()
extensively, there will be no more modifications of this function in
this series, but since the preceding commit removed this unused
parameter from lock_ref_oid_basic(), let's do it here too for
consistency.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The lock_ref_oid_basic() function has gradually been replaced by use
of the file transaction API, there are only 4 remaining callers of
it.
None of those callers pass non-NULL "extras" and "skip" parameters,
the last such caller went away in 92b1551b1d (refs: resolve symbolic
refs first, 2016-04-25), so let's remove the parameters.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In the last commit we removed the REF_DELETING flag from
lock_ref_oid_basic(). Since then all of the remaining callers do pass
REF_NO_DEREF, but that has been ignored completely since
7a418f3a17 (lock_ref_sha1_basic(): only handle REF_NODEREF mode,
2016-04-22).
So we can simply get rid of the parameter entirely.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The lock_ref_oid_basic() function has gradually been replaced by
most callers no longer performing a low-level "acquire lock,
update and release", and instead using the ref transaction API.
So there are only 4 remaining callers of lock_ref_oid_basic().
None of those callers pass REF_DELETING anymore, the last caller went
away in 92b1551b1d (refs: resolve symbolic refs first,
2016-04-25).
Before that we'd refactored and moved this code in:
- 8df4e51138 (struct ref_update: move "have_old" into "flags",
2015-02-17)
- 7bd9bcf372 (refs: split filesystem-based refs code into a new
file, 2015-11-09)
- 165056b2fc (lock_ref_for_update(): new function, 2016-04-24)
We then finally stopped using it in 92b1551b1d (noted above). So let's
remove the handling of this parameter.
By itself this change doesn't benefit us much, but it's the start of
even more removal of unused code in and around this function in
subsequent commits.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In e0cc8ac820 (packed_ref_store: make class into a subclass of
`ref_store`, 2017-06-23) a die() was added to packed_create_reflog(),
but not to any of the other reflog callbacks, let's do that.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Code cleanup around struct_type_init() functions.
* ab/struct-init:
string-list.h users: change to use *_{nodup,dup}()
string-list.[ch]: add a string_list_init_{nodup,dup}()
dir.[ch]: replace dir_init() with DIR_INIT
*.c *_init(): define in terms of corresponding *_INIT macro
*.h: move some *_INIT to designated initializers
Change all in-tree users of the string_list_init(LIST, BOOL) API to
use string_list_init_{nodup,dup}(LIST) instead.
As noted in the preceding commit let's leave the now-unused
string_list_init() wrapper in-place for any in-flight users, it can be
removed at some later date.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Use -1 as error return value throughout.
This removes spurious differences in the GIT_TRACE_REFS output, depending on the
ref storage backend active.
Before, the cached ref_iterator (but only that iterator!) would return
peel_object() output directly. No callers relied on the peel_status values
beyond success/failure. All calls to these functions go through
peel_iterated_oid(), which returns peel_object() as a fallback, but also
squashing the error values.
The iteration interface already passes REF_ISSYMREF and REF_ISBROKEN through the
flags argument, so the additional error values in enum peel_status provide no
value.
The ref iteration interface provides a separate peel() function because certain
formats (eg. packed-refs and reftable) can store the peeled object next to the
tag SHA1. Passing the peeled SHA1 as an optional argument to each_ref_fn maps
more naturally to the implementation of ref databases. Changing the code in this
way is left for a future refactoring.
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When "git update-ref -d" removes a ref that is packed, it left
empty directories under $GIT_DIR/refs/ for
* wc/packed-ref-removal-cleanup:
refs: cleanup directories when deleting packed ref
When deleting a packed ref via 'update-ref -d', a lockfile is made in
the directory that would contain the loose copy of that ref, creating
any directories in the ref's path that do not exist. When the
transaction completes, the lockfile is deleted, but any empty parent
directories made when creating the lockfile are left in place. These
empty directories are not removed by 'pack-refs' or other housekeeping
tasks and will accumulate over time.
When deleting a loose ref, we remove all empty parent directories at the
end of the transaction.
This commit applies the parent directory cleanup logic used when
deleting loose refs to packed refs as well.
Signed-off-by: Will Chandler <wfc@wfchandler.org>
Reviewed-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
SHA-256 transition.
* bc/hash-transition-interop-part-1:
hex: print objects using the hash algorithm member
hex: default to the_hash_algo on zero algorithm value
builtin/pack-objects: avoid using struct object_id for pack hash
commit-graph: don't store file hashes as struct object_id
builtin/show-index: set the algorithm for object IDs
hash: provide per-algorithm null OIDs
hash: set, copy, and use algo field in struct object_id
builtin/pack-redundant: avoid casting buffers to struct object_id
Use the final_oid_fn to finalize hashing of object IDs
hash: add a function to finalize object IDs
http-push: set algorithm when reading object ID
Always use oidread to read into struct object_id
hash: add an algo member to struct object_id
Up until recently, object IDs did not have an algorithm member, only a
hash. Consequently, it was possible to share one null (all-zeros)
object ID among all hash algorithms. Now that we're going to be
handling objects from multiple hash algorithms, it's important to make
sure that all object IDs have a correct algorithm field.
Introduce a per-algorithm null OID, and add it to struct hash_algo.
Introduce a wrapper function as well, and use it everywhere we used to
use the null_oid constant.
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The ref backend API uses errno as a sideband error channel.
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add and apply a semantic patch for converting code that open-codes
CALLOC_ARRAY to use it instead. It shortens the code and infers the
element size automatically.
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Similar to the previous commits, avoid peeking into the `struct
lock_file`. Use the lock file API instead. Note how we obtain the path
to the lock file if `fdopen_lock_file()` failed and that this is not a
problem: as documented in lockfile.h, failure to "fdopen" does not roll
back the lock file and we're free to, e.g., query it for its path.
Signed-off-by: Martin Ågren <martin.agren@gmail.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When set in the environment, GIT_TRACE_REFS makes git print operations and
results as they flow through the ref storage backend. This helps debug
discrepancies between different ref backends.
Example:
$ GIT_TRACE_REFS="1" ./git branch
15:42:09.769631 refs/debug.c:26 ref_store for .git
15:42:09.769681 refs/debug.c:249 read_raw_ref: HEAD: 0000000000000000000000000000000000000000 (=> refs/heads/ref-debug) type 1: 0
15:42:09.769695 refs/debug.c:249 read_raw_ref: refs/heads/ref-debug: 3a238e539b (=> refs/heads/ref-debug) type 0: 0
15:42:09.770282 refs/debug.c:233 ref_iterator_begin: refs/heads/ (0x1)
15:42:09.770290 refs/debug.c:189 iterator_advance: refs/heads/b4 (0)
15:42:09.770295 refs/debug.c:189 iterator_advance: refs/heads/branch3 (0)
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
REF_LOG_ONLY is used in the transaction preparation: if a symref is involved in
a transaction, the referent of the symref should be updated, and the symref
itself should only be updated in the reflog.
Other ref backends will need to duplicate this logic too, so move it to a
central place.
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This prepares for handling FETCH_HEAD (which is not a regular ref)
separately from the ref backend.
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
523fa69c (reflog: cleanse messages in the refs.c layer, 2020-07-10)
centralized reflog normalizaton. However, the normalizaton added a
leading "\t" to the message. This is an artifact of the reflog
storage format in the files backend, so it should be added there.
Routines that parse back the reflog (such as grab_nth_branch_switch)
expect the "\t" to not be in the message, so without this fix, git
with reftable cannot process the "@{-1}" syntax.
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Regarding reflog messages:
- We expect that a reflog message consists of a single line. The
file format used by the files backend may add a LF after the
message as a delimiter, and output by commands like "git log -g"
may complete such an incomplete line by adding a LF at the end,
but philosophically, the terminating LF is not a part of the
message.
- We however allow callers of refs API to supply a random sequence
of NUL terminated bytes. We cleanse caller-supplied message by
squashing a run of whitespaces into a SP, and by trimming trailing
whitespace, before storing the message. This is how we tolerate,
instead of erring out, a message with LF in it (be it at the end,
in the middle, or both).
Currently, the cleansing of the reflog message is done by the files
backend, before the log is written out. This is sufficient with the
current code, as that is the only backend that writes reflogs. But
new backends can be added that write reflogs, and we'd want the
resulting log message we would read out of "log -g" the same no
matter what backend is used, and moving the code to do so to the
generic layer is a way to do so.
An added benefit is that the "cleansing" function could be updated
later, independent from individual backends, to e.g. allow
multi-line log messages if we wanted to, and when that happens, it
would help a lot to ensure we covered all bases if the cleansing
function (which would be updated) is called from the generic layer.
Side note: I am not interested in supporting multi-line reflog
messages right at the moment (nobody is asking for it), but I
envision that instead of the "squash a run of whitespaces into a SP
and rtrim" cleansing, we can %urlencode problematic bytes in the
message *AND* append a SP at the end, when a new version of Git that
supports multi-line and/or verbatim reflog messages writes a reflog
record. The reading side can detect the presense of SP at the end
(which should have been rtrimmed out if it were written by existing
versions of Git) as a signal that decoding %urlencode recovers the
original reflog message.
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Document some of the flag options in refs_ref_iterator_begin, and explain how
ref_iterator_advance_fn should handle them.
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When cleaning up a transaction that has no updates queued, then the
transaction's backend data will not have been allocated. We correctly
handle this for the packed backend, where the cleanup function checks
whether the backend data has been allocated at all -- if not, then there
is nothing to clean up. For the files backend we do not check this and
as a result will hit a segfault due to dereferencing a `NULL` pointer
when cleaning up such a transaction.
Fix the issue by checking whether `backend_data` is set in the files
backend, too.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We often skip an optional prefix in a string with a hardcoded
constant, e.g.
if (starts_with(string, "prefix"))
string += 6;
which is less error prone when written
skip_prefix(string, "prefix", &string);
Note that this changes a few error messages from "git reflog expire
--expire=nonsense.timestamp", which used to complain by saying
'--expire=nonsense.timestamp' is not a valid timestamp
but with this change, we say
'nonsense.timestamp' is not a valid timestamp
which is more technically correct (the string with --expire= as
a prefix obviously cannot be a valid timestamp, but the error is
about the part of the input without that prefix).
Helped-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
refs_read_ref_full() wraps refs_resolve_ref_unsafe(), which handles a
NULL oid pointer of callers not interested in the resolved object ID.
Pass NULL from files_copy_or_rename_ref() to clarify that it is one
such caller.
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git pack-refs" can lose refs that are created while running, which
is getting corrected.
* sc/pack-refs-deletion-racefix:
pack-refs: always refresh after taking the lock file
When a packed ref is deleted, the whole packed-refs file is
rewritten to omit the ref that no longer exists. However if another
gc command is running and calls `pack-refs --all` simultaneously,
there is a chance that a ref that was just updated lose the newly
created commits.
Through these steps, losing commits on newly updated refs can be
demonstrated:
# step 1: compile git without `USE_NSEC` option
Some kernel releases do enable it by default while some do
not. And if we compile git without `USE_NSEC`, it will be easier
demonstrated by the following steps.
# step 2: setup a repository and add the first commit
git init repo &&
(cd repo &&
git config core.logallrefupdates true &&
git commit --allow-empty -m foo)
# step 3: in one terminal, repack the refs repeatedly
cd repo &&
while true
do
git pack-refs --all
done
# step 4: in another terminal, simultaneously update the
# master with update-ref, and create and delete an
# unrelated ref also with update-ref
cd repo &&
while true
do
us=$(git commit-tree -m foo -p HEAD HEAD^{tree}) &&
git update-ref refs/heads/newbranch $us &&
git update-ref refs/heads/master $us &&
git update-ref -d refs/heads/newbranch &&
them=$(git rev-parse master) &&
if test "$them" != "$us"
then
echo >&2 "lost commit: $us"
exit 1
fi
# eye candy
printf .
done
Though we have the packed-refs lock file and loose refs lock
files to avoid updating conflicts, a ref will lost its newly
commits if racy stat-validity of `packed-refs` file happens
(which is quite same as the racy-git described in
`Documentation/technical/racy-git.txt`), the following
specific set of operations demonstrates the problem:
1. Call `pack-refs --all` to pack all the loose refs to
packed-refs, and let say the modify time of the
packed-refs is DATE_M.
2. Call `update-ref` to update a new commit to master while
it is already packed. the old value (let us call it
OID_A) remains in the packed-refs file and write the new
value (let us call it OID_B) to $GIT_DIR/refs/heads/master.
3. Call `update-ref -d` within the same DATE_M from the 1th
step to delete a different ref newbranch which is packed
in the packed-refs file. It check newbranch's oid from
packed-refs file without locking it.
Meanwhile it keeps a snapshot of the packed-refs file in
memory and record the file's attributes with the snapshot.
The oid of master in the packed-refs's snapshot is OID_A.
4. Call a new `pack-refs --all` to pack the loose refs, the
oid of master in packe-refs file is OID_B, and the loose
refs $GIT_DIR/refs/heads/master is removed. Let's say
the `pack-refs --all` is very quickly done and the new
packed-refs file's modify time is still DATE_M, and it
has the same file size, even the same inode.
5. 3th step now goes on after checking the newbranch, it
begin to rewrite the packed-refs file. After get the
lock file of packed-ref file, it checks it's on-disk
file attributes with the snapshot, suck as the timestamp,
the file size and the inode value. If they are both the
same values, and the snapshot is not refreshed.
Because the loose ref of master is removed by 4th step,
`update-ref -d` will updates the new packed-ref to disk
which contains master with the oid OID_A. So now the
newly commit OID_B of master is lost.
The best path forward is just always refreshing after take
the lock file of `packed-refs` file. Traditionally we avoided
that because refreshing it implied parsing the whole file.
But these days we mmap it, so it really is just an extra
open()/mmap() and a quick read of the header. That doesn't seem
like an outrageous cost to pay when we're already taking the lock.
Signed-off-by: Sun Chao <sunchao9@huawei.com>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Sun Chao <sunchao9@huawei.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add the possibility of giving flags to dir_iterator_begin to initialize
a dir-iterator with special options.
Currently possible flags are:
- DIR_ITERATOR_PEDANTIC, which makes dir_iterator_advance abort
immediately in the case of an error, instead of keep looking for the
next valid entry;
- DIR_ITERATOR_FOLLOW_SYMLINKS, which makes the iterator follow
symlinks and include linked directories' contents in the iteration.
These new flags will be used in a subsequent patch.
Also add tests for the flags' usage and adjust refs/files-backend.c to
the new dir_iterator_begin signature.
Signed-off-by: Matheus Tavares <matheus.bernardino@usp.br>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
dir_iterator_advance() is a large function with two nested loops. Let's
improve its readability factoring out three functions and simplifying
its mechanics. The refactored model will no longer depend on
level.initialized and level.dir_state to keep track of the iteration
state and will perform on a single loop.
Also, dir_iterator_begin() currently does not check if the given string
represents a valid directory path. Since the refactored model will have
to stat() the given path at initialization, let's also check for this
kind of error and make dir_iterator_begin() return NULL, on failures,
with errno appropriately set. And add tests for this new behavior.
Improve documentation at dir-iteration.h and code comments at
dir-iterator.c to reflect the changes and eliminate possible
ambiguities.
Finally, adjust refs/files-backend.c to check for now possible
dir_iterator_begin() failures.
Original-patch-by: Daniel Ferreira <bnmvco@gmail.com>
Signed-off-by: Matheus Tavares <matheus.bernardino@usp.br>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A corner case bug in the refs API has been corrected.
* jk/refs-double-abort:
refs/files-backend: don't look at an aborted transaction
refs/files-backend: handle packed transaction prepare failure
"git rebase" uses the refs/rewritten/ hierarchy to store its
intermediate states, which inherently makes the hierarchy per
worktree, but it didn't quite work well.
* nd/rewritten-ref-is-per-worktree:
Make sure refs/rewritten/ is per-worktree
files-backend.c: reduce duplication in add_per_worktree_entries_to_dir()
files-backend.c: factor out per-worktree code in loose_fill_ref_dir()
When deleting refs, we hold packed-refs.lock and prepare a packed
transaction to drop the refs from the packed-refs file. If it turns out
that we don't need to rewrite the packed refs (e.g., because none of the
deletions were present in the file), then we abort the transaction.
If that abort succeeds, then the transaction struct will have been
freed, and we set our local pointer to NULL so we don't look at it
again.
However, if it fails, then the struct will _still_ have been freed
(because ref_transaction_abort() always frees). But we don't clean up
the pointer, and will jump to our cleanup code, which will try to abort
it again, causing a use-after-free.
It's actually impossible for this to trigger in practice, since
packed_transaction_abort() will never return anything but success. But
let's fix it anyway, since that's more than we should assume about the
packed-refs code (after all, we are already bothering to check for an
error result which cannot be triggered).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In files_transaction_prepare(), if we have to delete some refs, we use a
subordinate packed_transaction to do so. It's rare for that
sub-transaction's prepare step to fail, since we hold the packed-refs
lock. But if it does, we trigger a BUG() due to these steps:
- we've attached the packed transaction to the files transaction as
backend_data->packed_transaction
- when the prepare step fails, the packed transaction cleans itself
up, putting itself into the CLOSED state
- the error value from preparing the packed transaction lets us know
in files_transaction_prepare() that we should also clean up and
return an error. We call files_transaction_cleanup(), which tries to
abort backend_data->packed_transaction. Since it's already CLOSED,
that triggers an assertion in ref_transaction_abort().
We can fix that by disconnecting the packed transaction from the outer
files transaction, and then free-ing (not aborting!) it ourselves.
A few other options/alternatives I considered:
- we could just make it a noop to abort a CLOSED transaction. But that
seems less safe, since clearly this code expects (and enforces) a
particular set of state transitions.
- we could have files_transaction_cleanup() selectively call abort()
vs free() based on the state of the on the packed transaction.
That's basically a more restricted version of the above, but also
potentially unsafe.
- instead of disconnecting backend_data->packed_transaction on error,
we could wait to install it until we successfully prepare. That
might make the flow a little simpler, but it introduces a hassle.
Earlier parts of files_transaction_prepare() that encounter an error
will jump to the cleanup label, and expect that cleaning up the
outer transaction will clean up the packed transaction, too. We'd
have to adjust those sites to clean up the packed transaction.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
a9be29c981 (sequencer: make refs generated by the `label` command
worktree-local, 2018-04-25) adds refs/rewritten/ as per-worktree
reference space. Unfortunately (my bad) there are a couple places that
need update to make sure it's really per-worktree.
- add_per_worktree_entries_to_dir() is updated to make sure ref listing
look at per-worktree refs/rewritten/ instead of per-repo one [1]
- common_list[] is updated so that git_path() returns the correct
location. This includes "rev-parse --git-path".
This mess is created by me. I started trying to fix it with the
introduction of refs/worktree, where all refs will be per-worktree
without special treatments. Unfortunate refs/rewritten came before
refs/worktree so this is all we can do.
This also fixes logs/refs/worktree not being per-worktree.
[1] note that ref listing still works sometimes. For example, if you
have .git/worktrees/foo/refs/rewritten/bar AND the directory
.git/worktrees/refs/rewritten, refs/rewritten/bar will show up.
add_per_worktree_entries_to_dir() is only needed when the directory
.git/worktrees/refs/rewritten is missing.
Reported-by: Phillip Wood <phillip.wood123@gmail.com>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This function is duplicated to handle refs/bisect/ and refs/worktree/
and a third prefix is coming. Time to clean up.
This also fixes incorrect "refs/worktrees/" length in this code. The
correct length is 14 not 11. The test in the next patch will also cover
this.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is the first step for further cleaning up and extending this
function.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This parameter was added in fcc42ea0c9 (split_symref_update(): add a
files_ref_store argument, 2016-09-04) without comment, but never used.
The splitting is purely mechanical, and doesn't depend on the particular
ref-store. Let's drop this parameter in the name of simplicity.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
