"git pack-refs" that races with new ref creation or deletion have
been susceptible to lossage of refs under right conditions, which
has been tightened up.
* mh/ref-races:
for_each_ref: load all loose refs before packed refs
get_packed_ref_cache: reload packed-refs file when it changes
add a stat_validity struct
Extract a struct stat_data from cache_entry
packed_ref_cache: increment refcount when locked
do_for_each_entry(): increment the packed refs cache refcount
refs: manage lifetime of packed refs cache via reference counting
refs: implement simple transactions for the packed-refs file
refs: wrap the packed refs cache in a level of indirection
pack_refs(): split creation of packed refs and entry writing
repack_without_ref(): split list curation and entry writing
If we are iterating through the refs using for_each_ref (or
any of its sister functions), we can get into a race
condition with a simultaneous "pack-refs --prune" that looks
like this:
0. We have a large number of loose refs, and a few packed
refs. refs/heads/z/foo is loose, with no matching entry
in the packed-refs file.
1. Process A starts iterating through the refs. It loads
the packed-refs file from disk, then starts lazily
traversing through the loose ref directories.
2. Process B, running "pack-refs --prune", writes out the
new packed-refs file. It then deletes the newly packed
refs, including refs/heads/z/foo.
3. Meanwhile, process A has finally gotten to
refs/heads/z (it traverses alphabetically). It
descends, but finds nothing there. It checks its
cached view of the packed-refs file, but it does not
mention anything in "refs/heads/z/" at all (it predates
the new file written by B in step 2).
The traversal completes successfully without mentioning
refs/heads/z/foo at all (the name, of course, isn't
important; but the more refs you have and the farther down
the alphabetical list a ref is, the more likely it is to hit
the race). If refs/heads/z/foo did exist in the packed refs
file at state 0, we would see an entry for it, but it would
show whatever sha1 the ref had the last time it was packed
(which could be an arbitrarily long time ago).
This can be especially dangerous when process A is "git
prune", as it means our set of reachable tips will be
incomplete, and we may erroneously prune objects reachable
from that tip (the same thing can happen if "repack -ad" is
used, as it simply drops unreachable objects that are
packed).
This patch solves it by loading all of the loose refs for
our traversal into our in-memory cache, and then refreshing
the packed-refs cache. Because a pack-refs writer will
always put the new packed-refs file into place before
starting the prune, we know that any loose refs we fail to
see will either truly be missing, or will have already been
put in the packed-refs file by the time we refresh.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Once we read the packed-refs file into memory, we cache it
to save work on future ref lookups. However, our cache may
be out of date with respect to what is on disk if another
process is simultaneously packing the refs. Normally it
is acceptable for us to be a little out of date, since there
is no guarantee whether we read the file before or after the
simultaneous update. However, there is an important special
case: our packed-refs file must be up to date with respect
to any loose refs we read. Otherwise, we risk the following
race condition:
0. There exists a loose ref refs/heads/master.
1. Process A starts and looks up the ref "master". It
first checks $GIT_DIR/master, which does not exist. It
then loads (and caches) the packed-refs file to see if
"master" exists in it, which it does not.
2. Meanwhile, process B runs "pack-refs --all --prune". It
creates a new packed-refs file which contains
refs/heads/master, and removes the loose copy at
$GIT_DIR/refs/heads/master.
3. Process A continues its lookup, and eventually tries
$GIT_DIR/refs/heads/master. It sees that the loose ref
is missing, and falls back to the packed-refs file. But
it examines its cached version, which does not have
refs/heads/master. After trying a few other prefixes,
it reports master as a non-existent ref.
There are many variants (e.g., step 1 may involve process A
looking up another ref entirely, so even a fully qualified
refname can fail). One of the most interesting ones is if
"refs/heads/master" is already packed. In that case process
A will not see it as missing, but rather will report
whatever value happened to be in the packed-refs file before
process B repacked (which might be an arbitrarily old
value).
We can fix this by making sure we reload the packed-refs
file from disk after looking at any loose refs. That's
unacceptably slow, so we can check its stat()-validity as a
proxy, and read it only when it appears to have changed.
Reading the packed-refs file after performing any loose-ref
system calls is sufficient because we know the ordering of
the pack-refs process: it always makes sure the newly
written packed-refs file is installed into place before
pruning any loose refs. As long as those operations by B
appear in their executed order to process A, by the time A
sees the missing loose ref, the new packed-refs file must be
in place.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Increment the packed_ref_cache reference count while it is locked to
prevent its being freed.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This function calls a user-supplied callback function which could do
something that causes the packed refs cache to be invalidated. So
acquire a reference count on the data structure to prevent our copy
from being freed while we are iterating over it.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In struct packed_ref_cache, keep a count of the number of users of the
data structure. Only free the packed ref cache when the reference
count goes to zero rather than when the packed ref cache is cleared.
This mechanism will be used to prevent the cache data structure from
being freed while it is being iterated over.
So far, only the reference in struct ref_cache::packed is counted;
other users will be adjusted in separate commits.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Handle simple transactions for the packed-refs file at the
packed_ref_cache level via new functions lock_packed_refs(),
commit_packed_refs(), and rollback_packed_refs().
Only allow the packed ref cache to be modified (via add_packed_ref())
while the packed refs file is locked.
Change clone to add the new references within a transaction.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
As we know, we can solve any problem in this manner. In this case,
the problem is to avoid freeing a packed refs cache while somebody is
using it. So add a level of indirection as a prelude to
reference-counting the packed refs cache.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Split pack_refs() into multiple passes:
* Iterate over loose refs. For each one that can be turned into a
packed ref, create a corresponding entry in the packed refs cache.
* Write the packed refs to the packed-refs file.
This change isolates the mutation of the packed-refs file to a single
place.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The repack_without_ref() function first removes the deleted ref from
the internal packed-refs list, then writes the packed-refs list to
disk, omitting any broken or stale entries. This patch splits that
second step into multiple passes:
* collect the list of refnames that should be deleted from packed_refs
* delete those refnames from the cache
* write the remainder to the packed-refs file
The purpose of this change is to make the "write the remainder" part
reusable.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We read loose references in two steps. The code is roughly:
lstat()
if error ENOENT:
loose ref is missing; look for corresponding packed ref
else if S_ISLNK:
readlink()
if error:
report failure
else if S_ISDIR:
report failure
else
open()
if error:
report failure
read()
The problem is that the first filesystem call, to lstat(), is not
atomic with the second filesystem call, to readlink() or open().
Therefore it is possible for another process to change the file
between our two calls, for example:
* If the other process deletes the file, our second call will fail
with ENOENT, which we *should* interpret as "loose ref is missing;
look for corresponding packed ref". This can arise if the other
process is pack-refs; it might have just written a new packed-refs
file containing the old contents of the reference then deleted the
loose ref.
* If the other process changes a symlink into a plain file, our call
to readlink() will fail with EINVAL, which we *should* respond to by
trying to open() and read() the file.
The old code treats the reference as missing in both of these cases,
which is incorrect.
So instead, handle errors more selectively: if the result of
readline()/open() is a failure that is inconsistent with the result of
the previous lstat(), then something is fishy. In this case jump back
and start over again with a fresh call to lstat().
One race is still possible and undetected: another process could
change the file from a regular file into a symlink between the call to
lstat and the call to open(). The open() call would silently follow
the symlink and not know that something is wrong. This situation
could be detected in two ways:
* On systems that support O_NOFOLLOW, pass that option to the open().
* On other systems, call fstat() on the fd returned by open() and make
sure that it agrees with the stat info from the original lstat().
However, we don't use symlinks anymore, so this situation is unlikely.
Moreover, it doesn't appear that treating a symlink as a regular file
would have grave consequences; after all, this is exactly how the code
handles non-relative symlinks. So this commit leaves that race
unaddressed.
Note that this solves only the part of the race within
resolve_ref_unsafe. In the situation described above, we may still be
depending on a cached view of the packed-refs file; that race will be
dealt with in a future patch.
This problem was reported and diagnosed by Jeff King <peff@peff.net>,
and this solution is derived from his patch.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There is only one "break" statement within the loop, which jumps to
the code after the loop that handles the case of a file that holds a
SHA-1. So move that code from below the loop into the if statement
where the break was previously located. This makes the logic flow
more local.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The nesting was getting a bit out of hand, and it's about to get
worse.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Update reading and updating packed-refs file, correcting corner case
bugs.
* mh/packed-refs-various: (33 commits)
refs: handle the main ref_cache specially
refs: change do_for_each_*() functions to take ref_cache arguments
pack_one_ref(): do some cheap tests before a more expensive one
pack_one_ref(): use write_packed_entry() to do the writing
pack_one_ref(): use function peel_entry()
refs: inline function do_not_prune()
pack_refs(): change to use do_for_each_entry()
refs: use same lock_file object for both ref-packing functions
pack_one_ref(): rename "path" parameter to "refname"
pack-refs: merge code from pack-refs.{c,h} into refs.{c,h}
pack-refs: rename handle_one_ref() to pack_one_ref()
refs: extract a function write_packed_entry()
repack_without_ref(): write peeled refs in the rewritten file
t3211: demonstrate loss of peeled refs if a packed ref is deleted
refs: change how packed refs are deleted
search_ref_dir(): return an index rather than a pointer
repack_without_ref(): silence errors for dangling packed refs
t3210: test for spurious error messages for dangling packed refs
refs: change the internal reference-iteration API
refs: extract a function peel_entry()
...
Typing 'HEAD' is tedious, especially when we can use '@' instead.
The reason for choosing '@' is that it follows naturally from the
ref@op syntax (e.g. HEAD@{u}), except we have no ref, and no
operation, and when we don't have those, it makes sens to assume
'HEAD'.
So now we can use 'git show @~1', and all that goody goodness.
Until now '@' was a valid name, but it conflicts with this idea, so
let's make it invalid. Probably very few people, if any, used this name.
Signed-off-by: Felipe Contreras <felipe.contreras@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Hold the ref_cache instance for the main repository in a dedicated,
statically-allocated instance to avoid the need for a function call
and a linked-list traversal when it is needed.
Suggested by: Heiko Voigt <hvoigt@hvoigt.net>
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Change the callers convert submodule names into ref_cache pointers.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Change pack_refs() to work with a file descriptor instead of a FILE*
(making the file-locking code less awkward) and use
write_packed_entry() to do the writing.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Change pack_one_ref() to call peel_entry() rather than using its own
code for peeling references. Aside from sharing code, this lets it
take advantage of the optimization introduced by 6c4a060d7d.
Please note that we *could* use any peeled values that happen to
already be stored in the ref_entries, which would avoid some object
lookups for references that were already packed. But doing so would
also propagate any peeling errors across runs of "git pack-refs" and
give no way to recover from such errors. And "git pack-refs" isn't
run often enough that the performance cost is a problem. So instead,
add a new option to peel_entry() to force the entry to be re-peeled,
and call it with that option set.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Function do_not_prune() was redundantly checking REF_ISSYMREF, which
was already tested at the top of pack_one_ref(), so remove that check.
And the rest was trivial, so inline the function.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
pack_refs() was not using any of the extra features of for_each_ref(),
so change it to use do_for_each_entry(). This also gives it access to
the ref_entry and in particular its peeled field, which will be taken
advantage of in the next commit.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Use a single struct lock_file for both pack_refs() and
repack_without_ref().
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Make this function conform to the naming convention established in
65385ef7d4 for the rest of the refs.c file.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
pack-refs.c doesn't contain much code, and the code it does contain is
closely related to reference handling. Moreover, there is some
duplication between pack_refs() and repack_without_ref(). Therefore,
merge pack-refs.c into refs.c and pack-refs.h into refs.h.
The code duplication will be addressed in future commits.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Extract the I/O code from the "business logic" in repack_ref_fn().
Later there will be another caller for this function.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When a reference that existed in the packed-refs file is deleted, the
packed-refs file must be rewritten. Previously, the file was
rewritten without any peeled refs, even if the file contained peeled
refs when it was read. This was not a bug, because the packed-refs
file header didn't claim that the file contained peeled values. But
it had a performance cost, because the repository would lose the
benefit of having precomputed peeled references until pack-refs was
run again.
Teach repack_without_ref() to write peeled refs to the packed-refs
file (regardless of whether they were present in the old version of
the file).
This means that if the old version of the packed-refs file was not
fully peeled, then repack_without_ref() will have to peel references.
To avoid the expense of reading lots of loose references, we take two
shortcuts relative to pack-refs:
* If the peeled value of a reference is already known (i.e., because
it was read from the old version of the packed-refs file), then
output that peeled value again without any checks. This is the
usual code path and should avoid any noticeable overhead. (This is
different than pack-refs, which always re-peels references.)
* We don't verify that the packed ref is still current. It could be
that a packed references is overridden by a loose reference, in
which case the packed ref is no longer needed and might even refer
to an object that has been garbage collected. But we don't check;
instead, we just try to peel all references. If peeling is
successful, the peeled value is written out (even though it might
not be needed any more); if not, then the reference is silently
omitted from the output.
The extra overhead of peeling references in repack_without_ref()
should only be incurred the first time the packed-refs file is written
by a version of Git that knows about the "fully-peeled" attribute.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add a function remove_ref(), which removes a single entry from a
reference cache.
Use this function to reimplement repack_without_ref(). The old
version iterated over all refs, packing all of them except for the one
to be deleted, then discarded the entire packed reference cache. The
new version deletes the doomed reference from the cache *before*
iterating.
This has two advantages:
* the code for writing packed-refs becomes simpler, because it doesn't
have to exclude one of the references.
* it is no longer necessary to discard the packed refs cache after
deleting a reference: symbolic refs cannot be packed, so packed
references cannot depend on each other, so the rest of the packed
refs cache remains valid after a reference is deleted.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Change search_ref_dir() to return the index of the sought entry (or -1
on error) rather than a pointer to the entry. This will make it more
natural to use the function for removing an entry from the list.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Stop emitting an error message when deleting a packed reference if we
find another dangling packed reference that is overridden by a loose
reference. See the previous commit for a longer explanation of the
issue.
We have to be careful to make sure that the invalid packed reference
really *is* overridden by a loose reference; otherwise what we have
found is repository corruption, which we *should* report.
Please note that this approach is vulnerable to a race condition
similar to the race conditions already known to affect packed
references [1]:
* Process 1 tries to peel packed reference X as part of deleting
another packed reference. It discovers that X does not refer to a
valid object (because the object that it referred to has been
garbage collected).
* Process 2 tries to delete reference X. It starts by deleting the
loose reference X.
* Process 1 checks whether there is a loose reference X. There is not
(it has just been deleted by process 2), so process 1 reports a
spurious error "X does not point to a valid object!"
The worst case seems relatively harmless, and the fix is identical to
the fix that will be needed for the other race conditions (namely
holding a lock on the packed-refs file during *all* reference
deletions), so we leave the cleaning up of all of them as a future
project.
[1] http://thread.gmane.org/gmane.comp.version-control.git/211956
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Establish an internal API for iterating over references, which gives
the callback functions direct access to the ref_entry structure
describing the reference. (Do not change the iteration API that is
exposed outside of the module.)
Define a new internal callback signature
int each_ref_entry_fn(struct ref_entry *entry, void *cb_data)
Change do_for_each_ref_in_dir() and do_for_each_ref_in_dirs() to
accept each_ref_entry_fn callbacks, and rename them to
do_for_each_entry_in_dir() and do_for_each_entry_in_dirs(),
respectively. Adapt their callers accordingly.
Add a new function do_for_each_entry() analogous to do_for_each_ref()
but using the new callback style.
Change do_one_ref() into an each_ref_entry_fn that does some
bookkeeping and then calls a wrapped each_ref_fn.
Reimplement do_for_each_ref() in terms of do_for_each_entry(), using
do_one_ref() as an adapter.
Please note that the responsibility for setting current_ref remains in
do_one_ref(), which means that current_ref is *not* set when iterating
over references via the new internal API. This is not a disadvantage,
because current_ref is not needed by callers of the internal API (they
receive a pointer to the current ref_entry anyway). But more
importantly, this change prevents peel_ref() from returning invalid
results in the following scenario:
When iterating via the external API, the iteration always includes
both packed and loose references, and in particular never presents a
packed ref if there is a loose ref with the same name. The internal
API, on the other hand, gives the option to iterate over only the
packed references. During such an iteration, there is no check
whether the packed ref might be hidden by a loose ref of the same
name. But until now the packed ref was recorded in current_ref during
the iteration. So if peel_ref() were called with the reference name
corresponding to current ref, it would return the peeled version of
the packed ref even though there might be a loose ref that peels to a
different value. This scenario doesn't currently occur in the code,
but fix it to prevent things from breaking in a very confusing way in
the future.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Peel the entry, and as a side effect store the peeled value in the
entry. Use this function from two places in peel_ref(); a third
caller will be added soon.
Please note that this change can lead to ref_entries for unpacked refs
being peeled. This has no practical benefit but is harmless.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The old version was inconsistent: when a reference was
REF_KNOWS_PEELED but with a null peeled value, it returned non-zero
for the current reference but zero for other references. Change the
behavior for non-current references to match that of current_ref,
which is what callers expect. Document the behavior.
Current callers only call peel_ref() from within a for_each_ref-style
iteration and only for the current ref; therefore, the buggy code path
was never reached.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Instead of just returning a success/failure bit, return an enumeration
value that explains the reason for any failure. This will come in
handy shortly.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
It is a nice, logical unit of work, and putting it in a function
removes the need to use a goto in peel_ref(). Soon it will also have
other uses.
The algorithm is unchanged.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
It is a nice unit of work and soon will be needed from multiple
locations.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Instead of copying the reference's SHA1 into a caller-supplied
variable, just return the ref_entry itself (or NULL if there is no
such entry). This change will allow the function to be used from
elsewhere.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There is no way to drop out of the while loop. This code has been
dead since 432ad41e.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Document the bits that can appear in the "flags" parameter passed to
an each_ref_function and/or in the ref_entry::flag field.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
An internal function used to implement "git checkout @{-1}" was
hard to use correctly.
* jc/reflog-reverse-walk:
refs.c: fix fread error handling
reflog: add for_each_reflog_ent_reverse() API
for_each_recent_reflog_ent(): simplify opening of a reflog file
for_each_reflog_ent(): extract a helper to process a single entry
Not that we do not actively encourage having annotated tags outside
refs/tags/ hierarchy, but they were not advertised correctly to the
ls-remote and fetch with recent version of Git.
* jk/fully-peeled-packed-ref:
pack-refs: add fully-peeled trait
pack-refs: write peeled entry for non-tags
use parse_object_or_die instead of die("bad object")
avoid segfaults on parse_object failure
fread returns the number of items read, with no special error return.
Commit 98f85ff (reflog: add for_each_reflog_ent_reverse() API -
2013-03-08) introduced a call to fread which checks for an error with
"nread < 0" which is tautological since nread is unsigned. The correct
check in this case (which tries to read a single item) is "nread != 1".
Signed-off-by: John Keeping <john@keeping.me.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
