Use the new function sleep_millisec() to delay execution for a short
time. This avoids the invocation of select() with just a timeout, but
no file descriptors. Such a use of select() is quit with EINVAL on
Windows, leading to no delay at all.
Signed-off-by: NJohannes Sixt <j6t@kdbg.org>
Reviewed-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

When the goal is to wait for some random amount of time up to one
second, it is not necessary to compute with microsecond precision.
This is a preparation to re-use sleep_millisec().
Signed-off-by: NJohannes Sixt <j6t@kdbg.org>
Reviewed-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

On Windows, we do not have functions srandom() and random(). Use srand()
and rand(). These functions produce random numbers of lesser quality,
but for the purpose (a retry time-out) they are still good enough.
Signed-off-by: NJohannes Sixt <j6t@kdbg.org>
Reviewed-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

When copy_fd() function encounters errors, it emits error messages
itself, which makes it impossible for callers to take responsibility
for reporting errors, especially when they want to ignore certain
errors.

Move the error reporting to its callers in preparation.

 - copy_file() and copy_file_with_time() by indirection get their
   own calls to error().

 - hold_lock_file_for_append(), when told to die on error, used to
   exit(128) relying on the error message from copy_fd(), but now it
   does its own die() instead.  Note that the callers that do not
   pass LOCK_DIE_ON_ERROR need to be adjusted for this change, but
   fortunately there is none ;-)

 - filter_buffer_or_fd() has its own error() already, in addition to
   the message from copy_fd(), so this will change the output but
   arguably in a better way.
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Currently, there is only one attempt to lock a file. If it fails, the
whole operation fails.

But it might sometimes be advantageous to try acquiring a file lock a
few times before giving up. So add a new function,
hold_lock_file_for_update_timeout(), that allows a timeout to be
specified. Make hold_lock_file_for_update() a thin wrapper around the
new function.

If timeout_ms is positive, then retry for at least that many
milliseconds to acquire the lock. On each failed attempt, use select()
to wait for a backoff time that increases quadratically (capped at 1
second) and has a random component to prevent two processes from
getting synchronized. If timeout_ms is negative, retry indefinitely.

In a moment we will switch to using the new function when locking
packed-refs.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Locked paths can be saved in a linked list so that if something wrong
happens, *.lock are removed. For relative paths, this works fine if we
keep cwd the same, which is true 99% of time except:

- update-index and read-tree hold the lock on $GIT_DIR/index really
  early, then later on may call setup_work_tree() to move cwd.

- Suppose a lock is being held (e.g. by "git add") then somewhere
  down the line, somebody calls real_path (e.g. "link_alt_odb_entry"),
  which temporarily moves cwd away and back.

During that time when cwd is moved (either permanently or temporarily)
and we decide to die(), attempts to remove relative *.lock will fail,
and the next operation will complain that some files are still locked.

Avoid this case by turning relative paths to absolute before storing
the path in "filename" field.
Reported-by: NYue Lin Ho <yuelinho777@gmail.com>
Helped-by: NRamsay Jones <ramsay@ramsay1.demon.co.uk>
Helped-by: NJohannes Sixt <j6t@kdbg.org>
Signed-off-by: NNguyễn Thái Ngọc Duy <pclouds@gmail.com>
Adapted-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

The former caller uses unable_to_lock_message now.
Signed-off-by: NJonathan Nieder <jrnieder@gmail.com>
Reviewed-by: NRonnie Sahlberg <sahlberg@google.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Add a new function, fdopen_lock_file(), which returns a FILE pointer
open to the lockfile. If a stream is open on a lock_file object, it is
closed using fclose() on commit, rollback, or close_lock_file().

This change will allow callers to use stdio to write to a lockfile
without having to muck around in the internal representation of the
lock_file object (callers will be rewritten in upcoming commits).
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Move the interface declaration for the functions in lockfile.c from
cache.h to a new file, lockfile.h. Add #includes where necessary (and
remove some redundant includes of cache.h by files that already
include builtin.h).

Move the documentation of the lock_file state diagram from lockfile.c
to the new header file.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

lockfile.c contains the general API for locking any file. Code
specifically about the index file doesn't belong here.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Callers who don't pass LOCK_DIE_ON_ERROR might want to examine errno
to see what went wrong, so restore errno before returning.

In fact this function only has one caller, add_to_alternates_file(),
and it *does* use LOCK_DIE_ON_ERROR, but, you know, think of future
generations.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Add a function to return the path of the file that is locked by a
lock_file object. This reduces the knowledge that callers have to have
about the lock_file layout.
Suggested-by: NRonnie Sahlberg <sahlberg@google.com>
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

* remove_lock_file() -> remove_lock_files()
* remove_lock_file_on_signal() -> remove_lock_files_on_signal()
Suggested-by: NTorsten Bögershausen <tboegi@web.de>
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

This makes it harder to misread the name as LOCK_NODE_REF.
Suggested-by: NTorsten Bögershausen <tboegi@web.de>
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

commit_locked_index(), when writing to an alternate index file,
duplicates (poorly) the code in commit_lock_file(). And anyway, it
shouldn't have to know so much about the internal workings of lockfile
objects. So extract a new function commit_lock_file_to() that does the
work common to the two functions, and call it from both
commit_lock_file() and commit_locked_index().
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Rewrite last_path_elm() to take a strbuf parameter and to trim off the
last path name element in place rather than returning a pointer to the
beginning of the last path name element. This simplifies the function
a bit and makes it integrate better with its caller, which is now also
strbuf-based. Rename the function accordingly and a bit less tersely.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Change resolve_symlink() to take a strbuf rather than a string as
parameter.  This simplifies the code and removes an arbitrary pathname
length restriction.  It also means that lock_file's filename field no
longer needs to be initialized to a large size.
Helped-by: NTorsten Bögershausen <tboegi@web.de>
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Aside from shortening and simplifying the code, this removes another
place where the path name length is arbitrarily limited.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

For now, we still make sure to allocate at least PATH_MAX characters
for the strbuf because resolve_symlink() doesn't know how to expand
the space for its return value.  (That will be fixed in a moment.)

Another alternative would be to just use a strbuf as scratch space in
lock_file() but then store a pointer to the naked string in struct
lock_file.  But lock_file objects are often reused.  By reusing the
same strbuf, we can avoid having to reallocate the string most times
when a lock_file object is reused.
Helped-by: NTorsten Bögershausen <tboegi@web.de>
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Avoid relying on the filename length restrictions that are currently
checked by lock_file().
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

The function remove_lock_file_on_signal() is used as a signal handler.
It is not realistic to make the signal handler conform strictly to the
C standard, which is very restrictive about what a signal handler is
allowed to do.  But let's increase the likelihood that it will work:

The lock_file_list global variable and several fields from struct
lock_file are used by the signal handler.  Declare those values
"volatile" to (1) force the main process to write the values to RAM
promptly, and (2) prevent updates to these fields from being reordered
in a way that leaves an opportunity for a jump to the signal handler
while the object is in an inconsistent state.

We don't mark the filename field volatile because that would prevent
the use of strcpy(), and it is anyway unlikely that a compiler
re-orders a strcpy() call across other expressions.  So in practice it
should be possible to get away without "volatile" in the "filename"
case.
Suggested-by: NJohannes Sixt <j6t@kdbg.org>
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Because remove_lock_file() can be called any time by the signal
handler, it is important that any lock_file objects that are in the
lock_file_list are always in a valid state.  And since lock_file
objects are often reused (but are never removed from lock_file_list),
that means we have to be careful whenever mutating a lock_file object
to always keep it in a well-defined state.

This was formerly not the case, because part of the state was encoded
by setting lk->filename to the empty string vs. a valid filename.  It
is wrong to assume that this string can be updated atomically; for
example, even

    strcpy(lk->filename, value)

is unsafe.  But the old code was even more reckless; for example,

    strcpy(lk->filename, path);
    if (!(flags & LOCK_NODEREF))
            resolve_symlink(lk->filename, max_path_len);
    strcat(lk->filename, ".lock");

During the call to resolve_symlink(), lk->filename contained the name
of the file that was being locked, not the name of the lockfile.  If a
signal were raised during that interval, then the signal handler would
have deleted the valuable file!

We could probably continue to use the filename field to encode the
state by being careful to write characters 1..N-1 of the filename
first, and then overwrite the NUL at filename[0] with the first
character of the filename, but that would be awkward and error-prone.

So, instead of using the filename field to determine whether the
lock_file object is active, add a new field "lock_file::active" for
this purpose.  Be careful to set this field only when filename really
contains the name of a file that should be deleted on cleanup.
Helped-by: NJohannes Sixt <j6t@kdbg.org>
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

If rename() fails, call rollback_lock_file() to delete the lock file
(in case it is still present) and reset the filename field to the
empty string so that the lockfile object is left in a valid state.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

If closing an open lockfile fails, then we cannot be sure of the
contents of the lockfile, so there is nothing sensible to do but
delete it. This change also insures that the lock_file object is left
in a defined state in this error path (namely, unlocked).

The only caller that is ultimately affected by this change is
try_merge_strategy() -> write_locked_index(), which can call
close_lock_file() via various execution paths. This caller uses a
static lock_file object which previously could have been reused after
a failed close_lock_file() even though it was still in locked state.
This change causes the lock_file object to be unlocked on failure,
thus fixing this error-handling path.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

It was previously a bug to call commit_lock_file() with a lock_file
object that was not active (an illegal access would happen within the
function).  It was presumably never done, but this would be an easy
programming error to overlook.  So before continuing, do a consistency
check that the lock_file object really is locked.
Helped-by: NJohannes Sixt <j6t@kdbg.org>
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

It does just what we need.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: NJonathan Nieder <jrnieder@gmail.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

This is a bit easier to read than the old version, which nested part
of the non-error code in an "if" block.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: NRonnie Sahlberg <sahlberg@google.com>
Reviewed-by: NJonathan Nieder <jrnieder@gmail.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

There are a few places that use these values, so define constants for
them.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Document the valid states of lock_file objects, how they get into each
state, and how the state is encoded in the object's fields.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: NRonnie Sahlberg <sahlberg@google.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

The purpose of this change is to make the state diagram for
lock_file objects simpler and deterministic.

If locking fails, lock_file() sometimes leaves the lock_file object
partly initialized, but sometimes not. It sometimes registers the
object in lock_file_list, but sometimes not. This makes the state
diagram for lock_file objects effectively indeterministic and hard
to reason about. A future patch will also change the filename field
into a strbuf, which needs more involved initialization, so it will
become even more important that the state of a lock_file object is
well-defined after a failed attempt to lock.

The ambiguity doesn't currently have any ill effects, because
lock_file objects cannot be removed from the lock_file_list anyway.
But to make it easier to document and reason about the code, make
this behavior consistent: *always* initialize the lock_file object
and *always* register it in lock_file_list the first time it is
used, regardless of whether an error occurs.

While we're at it, make sure that all of the lock_file fields are
initialized to values appropriate for an unlocked object; the caller
is only responsible for making sure that on_list is set to zero before
the first time it is used.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

If there is an error copying the old contents to the lockfile, roll
back the lockfile before exiting so that the lockfile is not held
until process cleanup.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: NJonathan Nieder <jrnieder@gmail.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

If the call to adjust_shared_perm() fails, lock_file returns -1, which
to the caller looks like any other failure to lock the file.  So in
this case, roll back the lockfile before returning so that the lock
file is deleted immediately and the lockfile object is left in a
predictable state (namely, unlocked).  Previously, the lockfile was
retained until process cleanup in this situation.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: NJonathan Nieder <jrnieder@gmail.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

When rolling back the lockfile, call close_lock_file() so that the
lock_file's fd field gets set back to -1. This keeps the lock_file
object in a valid state, which is important because these objects are
allowed to be reused. It also makes it unnecessary to check whether
the file has already been closed, because close_lock_file() takes care
of that.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: NJonathan Nieder <jrnieder@gmail.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Eliminate a layer of nesting.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: NRonnie Sahlberg <sahlberg@google.com>
Reviewed-by: NJonathan Nieder <jrnieder@gmail.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

It is only necessary to clear the lock_file's filename field if it was
not already clear.
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: NRonnie Sahlberg <sahlberg@google.com>
Reviewed-by: NJonathan Nieder <jrnieder@gmail.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Suggested-by: NJonathan Nieder <jrnieder@gmail.com>
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

This function is used for other things besides the index, so rename it
accordingly.
Suggested-by: NJeff King <peff@peff.net>
Signed-off-by: NMichael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: NRonnie Sahlberg <sahlberg@google.com>
Reviewed-by: NJonathan Nieder <jrnieder@gmail.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

The caller, not this function, opened the file descriptor; it is
selfish for the callee to close it when it is done reading from it.
The caller may want an option to rewind and re-read the contents
after it returns.

Simplify the loop to copy the input in full to the output; its
body essentially is what a call to write_in_full() helper does.
Signed-off-by: NSteffen Prohaska <prohaska@zib.de>
Helped-by: NJeff King <peff@peff.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

In some code paths (e.g. giving "add -i" to prepare the contents to
be committed interactively inside "commit -p") where a caller takes
a lock, writes the new content, give chance for others to use it
while still holding the lock, and then releases the lock when all is
done.  As an extension, allow the caller to re-update an already
closed file while still holding the lock (i.e. not yet committed) by
re-opening the file, to be followed by updating the contents and
then by the usual close_lock_file() or commit_lock_file().

This is necessary if we want to add code to rebuild the cache-tree
and write the resulting index out after "add -i" returns the control
to "commit -p", for example.
Signed-off-by: NJunio C Hamano <gitster@pobox.com>