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>
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>
This function files_reflog_path returns void, which usually means
"return;" not returning "void value" from another function.
Reported-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
One of the problems with multiple worktree is accessing per-worktree
refs of one worktree from another worktree. This was sort of solved by
multiple ref store, where the code can open the ref store of another
worktree and has access to the ref space of that worktree.
The problem with this is reporting. "HEAD" in another ref space is
also called "HEAD" like in the current ref space. In order to
differentiate them, all the code must somehow carry the ref store
around and print something like "HEAD from this ref store".
But that is not feasible (or possible with a _lot_ of work). With the
current design, we pass a reference around as a string (so called
"refname"). Extending this design to pass a string _and_ a ref store
is a nightmare, especially when handling extended SHA-1 syntax.
So we do it another way. Instead of entering a separate ref space, we
make refs from other worktrees available in the current ref space. So
"HEAD" is always HEAD of the current worktree, but then we can have
"worktrees/blah/HEAD" to denote HEAD from a worktree named
"blah". This syntax coincidentally matches the underlying directory
structure which makes implementation a bit easier.
The main worktree has to be treated specially because well... it's
special from the beginning. So HEAD from the main worktree is
acccessible via the name "main-worktree/HEAD" instead of
"worktrees/main/HEAD" because "main" could be just another secondary
worktree.
This patch also makes it possible to specify refs from one worktree in
another one, e.g.
git log worktrees/foo/HEAD
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When multiple worktrees are used, we need rules to determine if
something belongs to one worktree or all of them. Instead of keeping
adding rules when new stuff comes (*), have a generic rule:
- Inside $GIT_DIR, which is per-worktree by default, add
$GIT_DIR/common which is always shared. New features that want to
share stuff should put stuff under this directory.
- Inside refs/, which is shared by default except refs/bisect, add
refs/worktree/ which is per-worktree. We may eventually move
refs/bisect to this new location and remove the exception in refs
code.
(*) And it may also include stuff from external commands which will
have no way to modify common/per-worktree rules.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is the flip side of the previous two patches: checking
for a non-zero oidcmp() can be more strictly expressed as
inequality. Like those patches, we write "!= 0" in the
coccinelle transformation, which covers by isomorphism the
more common:
if (oidcmp(E1, E2))
As with the previous two patches, this patch can be achieved
almost entirely by running "make coccicheck"; the only
differences are manual line-wrap fixes to match the original
code.
There is one thing to note for anybody replicating this,
though: coccinelle 1.0.4 seems to miss the case in
builtin/tag.c, even though it's basically the same as all
the others. Running with 1.0.7 does catch this, so
presumably it's just a coccinelle bug that was fixed in the
interim.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Using the more restrictive oideq() should, in the long run,
give the compiler more opportunities to optimize these
callsites. For now, this conversion should be a complete
noop with respect to the generated code.
The result is also perhaps a little more readable, as it
avoids the "zero is equal" idiom. Since it's so prevalent in
C, I think seasoned programmers tend not to even notice it
anymore, but it can sometimes make for awkward double
negations (e.g., we can drop a few !!oidcmp() instances
here).
This patch was generated almost entirely by the included
coccinelle patch. This mechanical conversion should be
completely safe, because we check explicitly for cases where
oidcmp() is compared to 0, which is what oideq() is doing
under the hood. Note that we don't have to catch "!oidcmp()"
separately; coccinelle's standard isomorphisms make sure the
two are treated equivalently.
I say "almost" because I did hand-edit the coccinelle output
to fix up a few style violations (it mostly keeps the
original formatting, but sometimes unwraps long lines).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Code clean-up to use size_t/ssize_t when they are the right type.
* jk/size-t:
strbuf_humanise: use unsigned variables
pass st.st_size as hint for strbuf_readlink()
strbuf_readlink: use ssize_t
strbuf: use size_t for length in intermediate variables
reencode_string: use size_t for string lengths
reencode_string: use st_add/st_mult helpers
Conversion from uchar[40] to struct object_id continues.
* bc/object-id:
pretty: switch hard-coded constants to the_hash_algo
sha1-file: convert constants to uses of the_hash_algo
log-tree: switch GIT_SHA1_HEXSZ to the_hash_algo->hexsz
diff: switch GIT_SHA1_HEXSZ to use the_hash_algo
builtin/merge-recursive: make hash independent
builtin/merge: switch to use the_hash_algo
builtin/fmt-merge-msg: make hash independent
builtin/update-index: simplify parsing of cacheinfo
builtin/update-index: convert to using the_hash_algo
refs/files-backend: use the_hash_algo for writing refs
sha1-name: use the_hash_algo when parsing object names
strbuf: allocate space with GIT_MAX_HEXSZ
commit: express tree entry constants in terms of the_hash_algo
hex: switch to using the_hash_algo
tree-walk: replace hard-coded constants with the_hash_algo
cache: update object ID functions for the_hash_algo
When we initially added the strbuf_readlink() function in
b11b7e13f4 (Add generic 'strbuf_readlink()' helper function,
2008-12-17), the point was that we generally have a _guess_
as to the correct size based on the stat information, but we
can't necessarily trust it.
Over the years, a few callers have grown up that simply pass
in 0, even though they have the stat information. Let's have
them pass in their hint for consistency (and in theory
efficiency, since it may avoid an extra resize/syscall loop,
but neither location is probably performance critical).
Note that st.st_size is actually an off_t, so in theory we
need xsize_t() here. But none of the other callsites use it,
and since this is just a hint, it doesn't matter either way
(if we wrap we'll simply start with a too-small hint and
then eventually complain when we cannot allocate the
memory).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In order to ensure we write the correct amount, use the_hash_algo to
find the correct number of bytes for the current hash.
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since we don't care about how many bytes were written, simplify the return
value logic.
log_ref_write_fd() was written long before strbuf was fleshed out. Remove
the old manual buffer management code and replace it with strbuf(). Also
update copy_reflog_msg() which is called only by log_ref_write_fd() to use
strbuf as it keeps things consistent.
Signed-off-by: Ben Peart <Ben.Peart@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add a repository argument to allow the callers of parse_object
to be more specific about which repository to act on. This is a small
mechanical change; it doesn't change the implementation to handle
repositories other than the_repository yet.
As with the previous commits, use a macro to catch callers passing a
repository other than the_repository at compile time.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Developer support update, by using BUG() macro instead of die() to
mark codepaths that should not happen more clearly.
* js/use-bug-macro:
BUG_exit_code: fix sparse "symbol not declared" warning
Convert remaining die*(BUG) messages
Replace all die("BUG: ...") calls by BUG() ones
run-command: use BUG() to report bugs, not die()
test-tool: help verifying BUG() code paths
Code clean-up to adjust to a more recent lockfile API convention that
allows lockfile instances kept on the stack.
* ma/lockfile-cleanup:
lock_file: move static locks into functions
lock_file: make function-local locks non-static
refs.c: do not die if locking fails in `delete_pseudoref()`
refs.c: do not die if locking fails in `write_pseudoref()`
t/helper/test-write-cache: clean up lock-handling
Placing `struct lock_file`s on the stack used to be a bad idea, because
the temp- and lockfile-machinery would keep a pointer into the struct.
But after 076aa2cbd (tempfile: auto-allocate tempfiles on heap,
2017-09-05), we can safely have lockfiles on the stack. (This applies
even if a user returns early, leaving a locked lock behind.)
These `struct lock_file`s are local to their respective functions and we
can drop their staticness.
For good measure, I have inspected these sites and come to believe that
they always release the lock, with the possible exception of bailing out
using `die()` or `exit()` or by returning from a `cmd_foo()`.
As pointed out by Jeff King, it would be bad if someone held on to a
`struct lock_file *` for some reason. After some grepping, I agree with
his findings: no-one appears to be doing that.
Signed-off-by: Martin Ågren <martin.agren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The effort to pass the repository in-core structure throughout the
API continues. This round deals with the code that implements the
refs/replace/ mechanism.
* sb/object-store-replace:
replace-object: allow lookup_replace_object to handle arbitrary repositories
replace-object: allow do_lookup_replace_object to handle arbitrary repositories
replace-object: allow prepare_replace_object to handle arbitrary repositories
refs: allow for_each_replace_ref to handle arbitrary repositories
refs: store the main ref store inside the repository struct
replace-object: add repository argument to lookup_replace_object
replace-object: add repository argument to do_lookup_replace_object
replace-object: add repository argument to prepare_replace_object
refs: add repository argument to for_each_replace_ref
refs: add repository argument to get_main_ref_store
replace-object: check_replace_refs is safe in multi repo environment
replace-object: eliminate replace objects prepared flag
object-store: move lookup_replace_object to replace-object.h
replace-object: move replace_map to object store
replace_object: use oidmap
In d8193743e0 (usage.c: add BUG() function, 2017-05-12), a new macro
was introduced to use for reporting bugs instead of die(). It was then
subsequently used to convert one single caller in 588a538ae5
(setup_git_env: convert die("BUG") to BUG(), 2017-05-12).
The cover letter of the patch series containing this patch
(cf 20170513032414.mfrwabt4hovujde2@sigill.intra.peff.net) is not
terribly clear why only one call site was converted, or what the plan
is for other, similar calls to die() to report bugs.
Let's just convert all remaining ones in one fell swoop.
This trick was performed by this invocation:
sed -i 's/die("BUG: /BUG("/g' $(git grep -l 'die("BUG' \*.c)
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This moves the 'main_ref_store', which was a global variable in refs.c
into the repository struct.
This patch does not deal with the parts in the refs subsystem which deal
with the submodules there. A later patch needs to get rid of the submodule
exposure in the refs API, such as 'get_submodule_ref_store(path)'.
Acked-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Commit f57f37e2e1 (files-backend: remove the use of
git_path(), 2017-03-26) introduced a regression when a
relative $GIT_DIR is used in a working tree:
- when we initialize the ref backend, we make a copy of
get_git_dir(), which may be relative
- later, we may call setup_work_tree() and chdir to the
root of the working tree
- further calls to the ref code will use the stored git
directory, but relative paths will now point to the
wrong place
The new test in t1501 demonstrates one such instance (the
bug causes us to write the ref update to the nonsense
"relative/relative/.git").
Since setup_work_tree() now uses chdir_notify, we can just
ask it update our relative paths when necessary.
Reported-by: Rafael Ascensao <rafa.almas@gmail.com>
Helped-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Crash fix for a corner case where an error codepath tried to unlock
what it did not acquire lock on.
* mr/packed-ref-store-fix:
files_initial_transaction_commit(): only unlock if locked
Running git clone --single-branch --mirror -b TAGNAME previously
triggered the following error message:
fatal: multiple updates for ref 'refs/tags/TAGNAME' not allowed.
This error condition is handled in files_initial_transaction_commit().
42c7f7ff9 ("commit_packed_refs(): remove call to `packed_refs_unlock()`", 2017-06-23)
introduced incorrect unlocking in the error path of this function,
which changes the error message to
fatal: BUG: packed_refs_unlock() called when not locked
Move the call to packed_refs_unlock() above the "cleanup:" label
since the unlocking should only be done in the last error path.
Signed-off-by: Mathias Rav <m@git.strova.dk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Recent update to the refs infrastructure implementation started
rewriting packed-refs file more often than before; this has been
optimized again for most trivial cases.
* mh/avoid-rewriting-packed-refs:
files-backend: don't rewrite the `packed-refs` file unnecessarily
t1409: check that `packed-refs` is not rewritten unnecessarily
Code clean-up in refs API implementation.
* mh/tidy-ref-update-flags:
refs: update some more docs to use "oid" rather than "sha1"
write_packed_entry(): take `object_id` arguments
refs: rename constant `REF_ISPRUNING` to `REF_IS_PRUNING`
refs: rename constant `REF_NODEREF` to `REF_NO_DEREF`
refs: tidy up and adjust visibility of the `ref_update` flags
ref_transaction_add_update(): remove a check
ref_transaction_update(): die on disallowed flags
prune_ref(): call `ref_transaction_add_update()` directly
files_transaction_prepare(): don't leak flags to packed transaction
Recent update to the refs infrastructure implementation started
rewriting packed-refs file more often than before; this has been
optimized again for most trivial cases.
* mh/avoid-rewriting-packed-refs:
files-backend: don't rewrite the `packed-refs` file unnecessarily
t1409: check that `packed-refs` is not rewritten unnecessarily
Conversion from uchar[20] to struct object_id continues.
* bc/object-id: (25 commits)
refs/files-backend: convert static functions to object_id
refs: convert read_raw_ref backends to struct object_id
refs: convert peel_object to struct object_id
refs: convert resolve_ref_unsafe to struct object_id
worktree: convert struct worktree to object_id
refs: convert resolve_gitlink_ref to struct object_id
Convert remaining callers of resolve_gitlink_ref to object_id
sha1_file: convert index_path and index_fd to struct object_id
refs: convert reflog_expire parameter to struct object_id
refs: convert read_ref_at to struct object_id
refs: convert peel_ref to struct object_id
builtin/pack-objects: convert to struct object_id
pack-bitmap: convert traverse_bitmap_commit_list to object_id
refs: convert dwim_log to struct object_id
builtin/reflog: convert remaining unsigned char uses to object_id
refs: convert dwim_ref and expand_ref to struct object_id
refs: convert read_ref and read_ref_full to object_id
refs: convert resolve_refdup and refs_resolve_refdup to struct object_id
Convert check_connected to use struct object_id
refs: update ref transactions to use struct object_id
...
Even after working with this code for years, I still see this constant
name as "ref node ref". Rename it to make it's meaning clearer.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The constants used for `ref_update::flags` were rather disorganized:
* The definitions in `refs.h` were not close to the functions that
used them.
* Maybe constants were defined in `refs-internal.h`, making them
visible to the whole refs module, when in fact they only made sense
for the files backend.
* Their documentation wasn't very consistent and partly still referred
to sha1s rather than oids.
* The numerical values followed no rational scheme
Fix all of these problems. The main functional improvement is that
some constants' visibility is now limited to `files-backend.c`.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We want to make `REF_ISPRUNING` internal to the files backend. For
this to be possible, `ref_transaction_add_update()` mustn't know about
it. So move the check that `REF_ISPRUNING` is only used with
`REF_NODEREF` from this function to `files_transaction_prepare()`.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
`prune_ref()` needs to use the `REF_ISPRUNING` flag, but we want to
make that flag private to the files backend. So instead of calling
`ref_transaction_delete()`, which is a public function and therefore
shouldn't allow the `REF_ISPRUNING` flag, change `prune_ref()` to call
`ref_transaction_add_update()`, which is private to the refs
module. (Note that we don't need any of the other services provided by
`ref_transaction_delete()`.)
This allows us to change `ref_transaction_update()` to reject the
`REF_ISPRUNING` flag. Do so by adjusting
`REF_TRANSACTION_UPDATE_ALLOWED_FLAGS`. Also add parentheses to its
definition to avoid potential future mishaps.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The files backend uses `ref_update::flags` for several internal flags.
But those flags have no meaning to the packed backend. So when adding
updates for the packed-refs transaction, only use flags that make
sense to the packed backend.
`REF_NODEREF` is part of the public interface, and it's logically what
we want, so include it. In fact it is actually ignored by the packed
backend (which doesn't support symbolic references), but that's its
own business.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Even when we are deleting references, we needn't overwrite the
`packed-refs` file if the references that we are deleting only exist
as loose references. Implement this optimization as follows:
* Add a function `is_packed_transaction_needed()`, which checks
whether a given packed-refs transaction actually needs to be carried
out (i.e., it returns false if the transaction obviously wouldn't
have any effect). This function must be called while holding the
`packed-refs` lock to avoid races.
* Change `files_transaction_prepare()` to check whether the
packed-refs transaction is actually needed. If not, squelch it, but
continue holding the `packed-refs` lock until the end of the
transaction to avoid races.
This fixes a mild regression caused by dc39e09942 (files_ref_store:
use a transaction to update packed refs, 2017-09-08). Before that
commit, unnecessary rewrites of `packed-refs` were suppressed by
`repack_without_refs()`. But the transaction-based writing introduced
by that commit didn't perform that optimization.
Note that the pre-dc39e09942 code still had to *read* the whole
`packed-refs` file to determine that the rewrite could be skipped, so
the performance for the cases that the write could be elided was
`O(N)` in the number of packed references both before and after
dc39e09942. But after that commit the constant factor increased.
This commit reimplements the optimization of eliding unnecessary
`packed-refs` rewrites. That, plus the fact that since
cfa2e29c34 (packed_ref_store: get rid of the `ref_cache` entirely,
2017-03-17) we don't necessarily have to read the whole `packed-refs`
file at all, means that deletes of one or a few loose references can
now be done with `O(n lg N)` effort, where `n` is the number of loose
references being deleted and `N` is the total number of packed
references.
This commit fixes two tests in t1409.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Transactions to update multiple references that involves a deletion
was quite broken in an error codepath and did not abort everything
correctly.
* mh/ref-locking-fix:
files_transaction_prepare(): fix handling of ref lock failure
t1404: add a bunch of tests of D/F conflicts
Since dc39e09942 (files_ref_store: use a transaction to update packed
refs, 2017-09-08), failure to lock a reference has been handled
incorrectly by `files_transaction_prepare()`. If
`lock_ref_for_update()` fails in the lock-acquisition loop of that
function, it sets `ret` then breaks out of that loop. Prior to
dc39e09942, that was OK, because the only thing following the loop was
the cleanup code. But dc39e09942 added another blurb of code between
the loop and the cleanup. That blurb sometimes resets `ret` to zero,
making the cleanup code think that the locking was successful.
Specifically, whenever
* One or more reference deletions have been processed successfully in
the lock-acquisition loop. (Processing the first such reference
causes a packed-ref transaction to be initialized.)
* Then `lock_ref_for_update()` fails for a subsequent reference. Such
a failure can happen for a number of reasons, such as the old SHA-1
not being correct, lock contention, etc. This causes a `break` out
of the lock-acquisition loop.
* The `packed-refs` lock is acquired successfully and
`ref_transaction_prepare()` succeeds for the packed-ref transaction.
This has the effect of resetting `ret` back to 0, and making the
cleanup code think that lock acquisition was successful.
In that case, any reference updates that were processed prior to
breaking out of the loop would be carried out (loose and packed), but
the reference that couldn't be locked and any subsequent references
would silently be ignored.
This can easily cause data loss if, for example, the user was trying
to push a new name for an existing branch while deleting the old name.
After the push, the branch could be left unreachable, and could even
subsequently be garbage-collected.
This problem was noticed in the context of deleting one reference and
creating another in a single transaction, when the two references D/F
conflict with each other, like
git update-ref --stdin <<EOF
delete refs/foo
create refs/foo/bar HEAD
EOF
This triggers the above bug because the deletion is processed
successfully for `refs/foo`, then the D/F conflict causes
`lock_ref_for_update()` to fail when `refs/foo/bar` is processed. In
this case the transaction *should* fail, but instead it causes
`refs/foo` to be deleted without creating `refs/foo`. This could
easily result in data loss.
The fix is simple: instead of just breaking out of the loop, jump
directly to the cleanup code. This fixes some tests in t1404 that were
added in the previous commit.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
