def _golevel(level, f, ename, newentry, metastore, tmax):
# close nodes back up the tree
assert (level)
default_meta_ofs = metastore.store(metadata.Metadata())
while ename[:len(level.ename)] != level.ename:
n = BlankNewEntry(level.ename[-1], default_meta_ofs, tmax)
n.flags |= IX_EXISTS
(n.children_ofs, n.children_n) = level.write(f)
level.parent.list.append(n)
level = level.parent
# create nodes down the tree
while len(level.ename) < len(ename):
level = Level(ename[:len(level.ename) + 1], level)
# are we in precisely the right place?
assert (ename == level.ename)
n = newentry or \
BlankNewEntry(ename and level.ename[-1] or None, default_meta_ofs, tmax)
(n.children_ofs, n.children_n) = level.write(f)
if level.parent:
level.parent.list.append(n)
level = level.parent
return level
def add(self, name, st, meta_ofs, hashgen=None):
endswith = name.endswith(b'/')
ename = pathsplit(name)
basename = ename[-1]
#log('add: %r %r\n' % (basename, name))
flags = IX_EXISTS
sha = None
if hashgen:
(gitmode, sha) = hashgen(name)
flags |= IX_HASHVALID
else:
(gitmode, sha) = (0, EMPTY_SHA)
if st:
isdir = stat.S_ISDIR(st.st_mode)
assert (isdir == endswith)
e = NewEntry(basename, name, self.tmax, st.st_dev, st.st_ino,
st.st_nlink, st.st_ctime, st.st_mtime, st.st_atime,
st.st_size, st.st_mode, gitmode, sha, flags, meta_ofs,
0, 0)
else:
assert (endswith)
meta_ofs = self.metastore.store(metadata.Metadata())
e = BlankNewEntry(basename, meta_ofs, self.tmax)
e.gitmode = gitmode
e.sha = sha
e.flags = flags
self._add(ename, e)
def node_info(n, name,
show_hash = False,
long_fmt = False,
classification = None,
numeric_ids = False,
human_readable = False):
"""Return a string containing the information to display for the node
n. Classification may be "all", "type", or None."""
result = ''
if show_hash:
result += "%s " % n.hash.encode('hex')
if long_fmt:
meta = copy.copy(n.metadata())
if meta:
meta.path = name
meta.size = n.size()
else:
# Fake it -- summary_str() is designed to handle a fake.
meta = metadata.Metadata()
meta.size = n.size()
meta.mode = n.mode
meta.path = name
meta.atime, meta.mtime, meta.ctime = n.atime, n.mtime, n.ctime
if stat.S_ISLNK(meta.mode):
meta.symlink_target = n.readlink()
result += metadata.summary_str(meta,
numeric_ids = numeric_ids,
classification = classification,
human_readable = human_readable)
else:
result += name
if classification:
mode = n.metadata() and n.metadata().mode or n.mode
result += xstat.classification_str(mode, classification == 'all')
return result
if first_root == None:
first_root = dirp[0]
elif first_root != dirp[0]:
root_collision = True
# If switching to a new sub-tree, finish the current sub-tree.
while parts > [x[0] for x in dirp]:
_pop(force_tree = None)
# If switching to a new sub-tree, start a new sub-tree.
for path_component in dirp[len(parts):]:
dir_name, fs_path = path_component
# Not indexed, so just grab the FS metadata or use empty metadata.
try:
meta = metadata.from_path(fs_path, normalized=True) \
if fs_path else metadata.Metadata()
except (OSError, IOError) as e:
add_error(e)
lastskip_name = dir_name
meta = metadata.Metadata()
_push(dir_name, meta)
if not file:
if len(parts) == 1:
continue # We're at the top level -- keep the current root dir
# Since there's no filename, this is a subdir -- finish it.
oldtree = already_saved(ent) # may be None
newtree = _pop(force_tree = oldtree)
if not oldtree:
if lastskip_name and lastskip_name.startswith(ent.name):
ent.invalidate()
def index_dirty():
with no_lingering_errors():
with test_tempdir('bup-tindex-') as tmpdir:
orig_cwd = os.getcwd()
try:
os.chdir(tmpdir)
default_meta = metadata.Metadata()
ms1 = index.MetaStoreWriter('index.meta.tmp')
ms2 = index.MetaStoreWriter('index2.meta.tmp')
ms3 = index.MetaStoreWriter('index3.meta.tmp')
meta_ofs1 = ms1.store(default_meta)
meta_ofs2 = ms2.store(default_meta)
meta_ofs3 = ms3.store(default_meta)
ds = xstat.stat(lib_t_dir)
fs = xstat.stat(lib_t_dir + '/tindex.py')
tmax = (time.time() - 1) * 10**9
w1 = index.Writer('index.tmp', ms1, tmax)
w1.add('/a/b/x', fs, meta_ofs1)
w1.add('/a/b/c', fs, meta_ofs1)
w1.add('/a/b/', ds, meta_ofs1)
w1.add('/a/', ds, meta_ofs1)
#w1.close()
WVPASS()
w2 = index.Writer('index2.tmp', ms2, tmax)
w2.add('/a/b/n/2', fs, meta_ofs2)
#w2.close()
WVPASS()
w3 = index.Writer('index3.tmp', ms3, tmax)
w3.add('/a/c/n/3', fs, meta_ofs3)
#w3.close()
WVPASS()
r1 = w1.new_reader()
r2 = w2.new_reader()
r3 = w3.new_reader()
WVPASS()
r1all = [e.name for e in r1]
WVPASSEQ(r1all, ['/a/b/x', '/a/b/c', '/a/b/', '/a/', '/'])
r2all = [e.name for e in r2]
WVPASSEQ(r2all, ['/a/b/n/2', '/a/b/n/', '/a/b/', '/a/', '/'])
r3all = [e.name for e in r3]
WVPASSEQ(r3all, ['/a/c/n/3', '/a/c/n/', '/a/c/', '/a/', '/'])
all = [e.name for e in index.merge(r2, r1, r3)]
WVPASSEQ(all, [
'/a/c/n/3', '/a/c/n/', '/a/c/', '/a/b/x', '/a/b/n/2',
'/a/b/n/', '/a/b/c', '/a/b/', '/a/', '/'
])
fake_validate(r1)
dump(r1)
print[hex(e.flags) for e in r1]
WVPASSEQ([e.name for e in r1 if e.is_valid()], r1all)
WVPASSEQ([e.name for e in r1 if not e.is_valid()], [])
WVPASSEQ([
e.name
for e in index.merge(r2, r1, r3) if not e.is_valid()
], [
'/a/c/n/3', '/a/c/n/', '/a/c/', '/a/b/n/2', '/a/b/n/',
'/a/b/', '/a/', '/'
])
expect_invalid = ['/'] + r2all + r3all
expect_real = (set(r1all) - set(r2all) - set(r3all)) \
| set(['/a/b/n/2', '/a/c/n/3'])
dump(index.merge(r2, r1, r3))
for e in index.merge(r2, r1, r3):
print e.name, hex(e.flags), e.ctime
eiv = e.name in expect_invalid
er = e.name in expect_real
WVPASSEQ(eiv, not e.is_valid())
WVPASSEQ(er, e.is_real())
fake_validate(r2, r3)
dump(index.merge(r2, r1, r3))
WVPASSEQ([
e.name
for e in index.merge(r2, r1, r3) if not e.is_valid()
], [])
e = eget(index.merge(r2, r1, r3), '/a/b/c')
e.invalidate()
e.repack()
dump(index.merge(r2, r1, r3))
WVPASSEQ([
e.name
for e in index.merge(r2, r1, r3) if not e.is_valid()
], ['/a/b/c', '/a/b/', '/a/', '/'])
w1.close()
w2.close()
w3.close()
finally:
os.chdir(orig_cwd)
def main(argv):
# Hack around lack of nonlocal vars in python 2
_nonlocal = {}
o = options.Options(optspec)
opt, flags, extra = o.parse_bytes(argv[1:])
if opt.indexfile:
opt.indexfile = argv_bytes(opt.indexfile)
if opt.name:
opt.name = argv_bytes(opt.name)
if opt.remote:
opt.remote = argv_bytes(opt.remote)
if opt.strip_path:
opt.strip_path = argv_bytes(opt.strip_path)
git.check_repo_or_die()
if not (opt.tree or opt.commit or opt.name):
o.fatal("use one or more of -t, -c, -n")
if not extra:
o.fatal("no filenames given")
extra = [argv_bytes(x) for x in extra]
opt.progress = (istty2 and not opt.quiet)
opt.smaller = parse_num(opt.smaller or 0)
if opt.bwlimit:
client.bwlimit = parse_num(opt.bwlimit)
if opt.date:
date = parse_date_or_fatal(opt.date, o.fatal)
else:
date = time.time()
if opt.strip and opt.strip_path:
o.fatal("--strip is incompatible with --strip-path")
graft_points = []
if opt.graft:
if opt.strip:
o.fatal("--strip is incompatible with --graft")
if opt.strip_path:
o.fatal("--strip-path is incompatible with --graft")
for (option, parameter) in flags:
if option == "--graft":
parameter = argv_bytes(parameter)
splitted_parameter = parameter.split(b'=')
if len(splitted_parameter) != 2:
o.fatal(
"a graft point must be of the form old_path=new_path")
old_path, new_path = splitted_parameter
if not (old_path and new_path):
o.fatal("a graft point cannot be empty")
graft_points.append(
(resolve_parent(old_path), resolve_parent(new_path)))
is_reverse = environ.get(b'BUP_SERVER_REVERSE')
if is_reverse and opt.remote:
o.fatal("don't use -r in reverse mode; it's automatic")
name = opt.name
if name and not valid_save_name(name):
o.fatal("'%s' is not a valid branch name" % path_msg(name))
refname = name and b'refs/heads/%s' % name or None
if opt.remote or is_reverse:
try:
cli = client.Client(opt.remote)
except client.ClientError as e:
log('error: %s' % e)
sys.exit(1)
oldref = refname and cli.read_ref(refname) or None
w = cli.new_packwriter(compression_level=opt.compress)
else:
cli = None
oldref = refname and git.read_ref(refname) or None
w = git.PackWriter(compression_level=opt.compress)
handle_ctrl_c()
# Metadata is stored in a file named .bupm in each directory. The
# first metadata entry will be the metadata for the current directory.
# The remaining entries will be for each of the other directory
# elements, in the order they're listed in the index.
#
# Since the git tree elements are sorted according to
# git.shalist_item_sort_key, the metalist items are accumulated as
# (sort_key, metadata) tuples, and then sorted when the .bupm file is
# created. The sort_key should have been computed using the element's
# mangled name and git mode (after hashsplitting), but the code isn't
# actually doing that but rather uses the element's real name and mode.
# This makes things a bit more difficult when reading it back, see
# vfs.ordered_tree_entries().
# Maintain a stack of information representing the current location in
# the archive being constructed. The current path is recorded in
# parts, which will be something like ['', 'home', 'someuser'], and
# the accumulated content and metadata for of the dirs in parts is
# stored in parallel stacks in shalists and metalists.
parts = [] # Current archive position (stack of dir names).
shalists = [] # Hashes for each dir in paths.
metalists = [] # Metadata for each dir in paths.
def _push(part, metadata):
# Enter a new archive directory -- make it the current directory.
parts.append(part)
shalists.append([])
metalists.append([(b'', metadata)]) # This dir's metadata (no name).
def _pop(force_tree, dir_metadata=None):
# Leave the current archive directory and add its tree to its parent.
assert (len(parts) >= 1)
part = parts.pop()
shalist = shalists.pop()
metalist = metalists.pop()
# FIXME: only test if collision is possible (i.e. given --strip, etc.)?
if force_tree:
tree = force_tree
else:
names_seen = set()
clean_list = []
metaidx = 1 # entry at 0 is for the dir
for x in shalist:
name = x[1]
if name in names_seen:
parent_path = b'/'.join(parts) + b'/'
add_error('error: ignoring duplicate path %s in %s' %
(path_msg(name), path_msg(parent_path)))
if not stat.S_ISDIR(x[0]):
del metalist[metaidx]
else:
names_seen.add(name)
clean_list.append(x)
if not stat.S_ISDIR(x[0]):
metaidx += 1
if dir_metadata: # Override the original metadata pushed for this dir.
metalist = [(b'', dir_metadata)] + metalist[1:]
sorted_metalist = sorted(metalist, key=lambda x: x[0])
metadata = b''.join([m[1].encode() for m in sorted_metalist])
metadata_f = BytesIO(metadata)
mode, id = hashsplit.split_to_blob_or_tree(w.new_blob,
w.new_tree,
[metadata_f],
keep_boundaries=False)
clean_list.append((mode, b'.bupm', id))
tree = w.new_tree(clean_list)
if shalists:
shalists[-1].append((GIT_MODE_TREE,
git.mangle_name(part, GIT_MODE_TREE,
GIT_MODE_TREE), tree))
return tree
_nonlocal['count'] = 0
_nonlocal['subcount'] = 0
_nonlocal['lastremain'] = None
def progress_report(n):
_nonlocal['subcount'] += n
cc = _nonlocal['count'] + _nonlocal['subcount']
pct = total and (cc * 100.0 / total) or 0
now = time.time()
elapsed = now - tstart
kps = elapsed and int(cc / 1024. / elapsed)
kps_frac = 10**int(math.log(kps + 1, 10) - 1)
kps = int(kps / kps_frac) * kps_frac
if cc:
remain = elapsed * 1.0 / cc * (total - cc)
else:
remain = 0.0
if (_nonlocal['lastremain'] and (remain > _nonlocal['lastremain']) and
((remain - _nonlocal['lastremain']) / _nonlocal['lastremain'] <
0.05)):
remain = _nonlocal['lastremain']
else:
_nonlocal['lastremain'] = remain
hours = int(remain / 60 / 60)
mins = int(remain / 60 - hours * 60)
secs = int(remain - hours * 60 * 60 - mins * 60)
if elapsed < 30:
remainstr = ''
kpsstr = ''
else:
kpsstr = '%dk/s' % kps
if hours:
remainstr = '%dh%dm' % (hours, mins)
elif mins:
remainstr = '%dm%d' % (mins, secs)
else:
remainstr = '%ds' % secs
qprogress(
'Saving: %.2f%% (%d/%dk, %d/%d files) %s %s\r' %
(pct, cc / 1024, total / 1024, fcount, ftotal, remainstr, kpsstr))
indexfile = opt.indexfile or git.repo(b'bupindex')
r = index.Reader(indexfile)
try:
msr = index.MetaStoreReader(indexfile + b'.meta')
except IOError as ex:
if ex.errno != EACCES:
raise
log('error: cannot access %r; have you run bup index?' %
path_msg(indexfile))
sys.exit(1)
hlink_db = hlinkdb.HLinkDB(indexfile + b'.hlink')
def already_saved(ent):
return ent.is_valid() and w.exists(ent.sha) and ent.sha
def wantrecurse_pre(ent):
return not already_saved(ent)
def wantrecurse_during(ent):
return not already_saved(ent) or ent.sha_missing()
def find_hardlink_target(hlink_db, ent):
if hlink_db and not stat.S_ISDIR(ent.mode) and ent.nlink > 1:
link_paths = hlink_db.node_paths(ent.dev, ent.ino)
if link_paths:
return link_paths[0]
total = ftotal = 0
if opt.progress:
for (transname, ent) in r.filter(extra, wantrecurse=wantrecurse_pre):
if not (ftotal % 10024):
qprogress('Reading index: %d\r' % ftotal)
exists = ent.exists()
hashvalid = already_saved(ent)
ent.set_sha_missing(not hashvalid)
if not opt.smaller or ent.size < opt.smaller:
if exists and not hashvalid:
total += ent.size
ftotal += 1
progress('Reading index: %d, done.\n' % ftotal)
hashsplit.progress_callback = progress_report
# Root collisions occur when strip or graft options map more than one
# path to the same directory (paths which originally had separate
# parents). When that situation is detected, use empty metadata for
# the parent. Otherwise, use the metadata for the common parent.
# Collision example: "bup save ... --strip /foo /foo/bar /bar".
# FIXME: Add collision tests, or handle collisions some other way.
# FIXME: Detect/handle strip/graft name collisions (other than root),
# i.e. if '/foo/bar' and '/bar' both map to '/'.
first_root = None
root_collision = None
tstart = time.time()
fcount = 0
lastskip_name = None
lastdir = b''
for (transname, ent) in r.filter(extra, wantrecurse=wantrecurse_during):
(dir, file) = os.path.split(ent.name)
exists = (ent.flags & index.IX_EXISTS)
hashvalid = already_saved(ent)
wasmissing = ent.sha_missing()
oldsize = ent.size
if opt.verbose:
if not exists:
status = 'D'
elif not hashvalid:
if ent.sha == index.EMPTY_SHA:
status = 'A'
else:
status = 'M'
else:
status = ' '
if opt.verbose >= 2:
log('%s %-70s\n' % (status, path_msg(ent.name)))
elif not stat.S_ISDIR(ent.mode) and lastdir != dir:
if not lastdir.startswith(dir):
log('%s %-70s\n' %
(status, path_msg(os.path.join(dir, b''))))
lastdir = dir
if opt.progress:
progress_report(0)
fcount += 1
if not exists:
continue
if opt.smaller and ent.size >= opt.smaller:
if exists and not hashvalid:
if opt.verbose:
log('skipping large file "%s"\n' % path_msg(ent.name))
lastskip_name = ent.name
continue
assert (dir.startswith(b'/'))
if opt.strip:
dirp = stripped_path_components(dir, extra)
elif opt.strip_path:
dirp = stripped_path_components(dir, [opt.strip_path])
elif graft_points:
dirp = grafted_path_components(graft_points, dir)
else:
dirp = path_components(dir)
# At this point, dirp contains a representation of the archive
# path that looks like [(archive_dir_name, real_fs_path), ...].
# So given "bup save ... --strip /foo/bar /foo/bar/baz", dirp
# might look like this at some point:
# [('', '/foo/bar'), ('baz', '/foo/bar/baz'), ...].
# This dual representation supports stripping/grafting, where the
# archive path may not have a direct correspondence with the
# filesystem. The root directory is represented by an initial
# component named '', and any component that doesn't have a
# corresponding filesystem directory (due to grafting, for
# example) will have a real_fs_path of None, i.e. [('', None),
# ...].
if first_root == None:
first_root = dirp[0]
elif first_root != dirp[0]:
root_collision = True
# If switching to a new sub-tree, finish the current sub-tree.
while parts > [x[0] for x in dirp]:
_pop(force_tree=None)
# If switching to a new sub-tree, start a new sub-tree.
for path_component in dirp[len(parts):]:
dir_name, fs_path = path_component
# Not indexed, so just grab the FS metadata or use empty metadata.
try:
meta = metadata.from_path(fs_path, normalized=True) \
if fs_path else metadata.Metadata()
except (OSError, IOError) as e:
add_error(e)
lastskip_name = dir_name
meta = metadata.Metadata()
_push(dir_name, meta)
if not file:
if len(parts) == 1:
continue # We're at the top level -- keep the current root dir
# Since there's no filename, this is a subdir -- finish it.
oldtree = already_saved(ent) # may be None
newtree = _pop(force_tree=oldtree)
if not oldtree:
if lastskip_name and lastskip_name.startswith(ent.name):
ent.invalidate()
else:
ent.validate(GIT_MODE_TREE, newtree)
ent.repack()
if exists and wasmissing:
_nonlocal['count'] += oldsize
continue
# it's not a directory
if hashvalid:
id = ent.sha
git_name = git.mangle_name(file, ent.mode, ent.gitmode)
git_info = (ent.gitmode, git_name, id)
shalists[-1].append(git_info)
sort_key = git.shalist_item_sort_key((ent.mode, file, id))
meta = msr.metadata_at(ent.meta_ofs)
meta.hardlink_target = find_hardlink_target(hlink_db, ent)
# Restore the times that were cleared to 0 in the metastore.
(meta.atime, meta.mtime, meta.ctime) = (ent.atime, ent.mtime,
ent.ctime)
metalists[-1].append((sort_key, meta))
else:
id = None
hlink = find_hardlink_target(hlink_db, ent)
try:
meta = metadata.from_path(
ent.name,
hardlink_target=hlink,
normalized=True,
after_stat=after_nondir_metadata_stat)
except (OSError, IOError) as e:
add_error(e)
lastskip_name = ent.name
continue
if stat.S_IFMT(ent.mode) != stat.S_IFMT(meta.mode):
# The mode changed since we indexed the file, this is bad.
# This can cause two issues:
# 1) We e.g. think the file is a regular file, but now it's
# something else (a device, socket, FIFO or symlink, etc.)
# and _read_ from it when we shouldn't.
# 2) We then record it as valid, but don't update the index
# metadata, and on a subsequent save it has 'hashvalid'
# but is recorded as the file type from the index, when
# the content is something else ...
# Avoid all of these consistency issues by just skipping such
# things - it really ought to not happen anyway.
add_error("%s: mode changed since indexing, skipping." %
path_msg(ent.name))
lastskip_name = ent.name
continue
if stat.S_ISREG(ent.mode):
try:
# If the file changes while we're reading it, then our reading
# may stop at some point, but the stat() above may have gotten
# a different size already. Recalculate the meta size so that
# the repository records the accurate size in the metadata, even
# if the other stat() data might be slightly older than the file
# content (which we can't fix, this is inherently racy, but we
# can prevent the size mismatch.)
meta.size = 0
def new_blob(data):
meta.size += len(data)
return w.new_blob(data)
before_saving_regular_file(ent.name)
with hashsplit.open_noatime(ent.name) as f:
(mode, id) = hashsplit.split_to_blob_or_tree(
new_blob, w.new_tree, [f], keep_boundaries=False)
except (IOError, OSError) as e:
add_error('%s: %s' % (ent.name, e))
lastskip_name = ent.name
elif stat.S_ISDIR(ent.mode):
assert (0) # handled above
elif stat.S_ISLNK(ent.mode):
mode, id = (GIT_MODE_SYMLINK, w.new_blob(meta.symlink_target))
else:
# Everything else should be fully described by its
# metadata, so just record an empty blob, so the paths
# in the tree and .bupm will match up.
(mode, id) = (GIT_MODE_FILE, w.new_blob(b''))
if id:
ent.validate(mode, id)
ent.repack()
git_name = git.mangle_name(file, ent.mode, ent.gitmode)
git_info = (mode, git_name, id)
shalists[-1].append(git_info)
sort_key = git.shalist_item_sort_key((ent.mode, file, id))
metalists[-1].append((sort_key, meta))
if exists and wasmissing:
_nonlocal['count'] += oldsize
_nonlocal['subcount'] = 0
if opt.progress:
pct = total and _nonlocal['count'] * 100.0 / total or 100
progress(
'Saving: %.2f%% (%d/%dk, %d/%d files), done. \n' %
(pct, _nonlocal['count'] / 1024, total / 1024, fcount, ftotal))
while len(parts) > 1: # _pop() all the parts above the root
_pop(force_tree=None)
assert (len(shalists) == 1)
assert (len(metalists) == 1)
# Finish the root directory.
tree = _pop(
force_tree=None,
# When there's a collision, use empty metadata for the root.
dir_metadata=metadata.Metadata() if root_collision else None)
sys.stdout.flush()
out = byte_stream(sys.stdout)
if opt.tree:
out.write(hexlify(tree))
out.write(b'\n')
if opt.commit or name:
if compat.py_maj > 2:
# Strip b prefix from python 3 bytes reprs to preserve previous format
msgcmd = b'[%s]' % b', '.join(
[repr(argv_bytes(x))[1:].encode('ascii') for x in argv])
else:
msgcmd = repr(argv)
msg = b'bup save\n\nGenerated by command:\n%s\n' % msgcmd
userline = (b'%s <%s@%s>' % (userfullname(), username(), hostname()))
commit = w.new_commit(tree, oldref, userline, date, None, userline,
date, None, msg)
if opt.commit:
out.write(hexlify(commit))
out.write(b'\n')
msr.close()
w.close() # must close before we can update the ref
if opt.name:
if cli:
cli.update_ref(refname, commit, oldref)
else:
git.update_ref(refname, commit, oldref)
if cli:
cli.close()
if saved_errors:
log('WARNING: %d errors encountered while saving.\n' %
len(saved_errors))
sys.exit(1)
dirp = stripped_path_components(dir, [opt.strip_path])
elif graft_points:
dirp = grafted_path_components(graft_points, dir)
else:
dirp = path_components(dir)
while parts > [x[0] for x in dirp]:
_pop(force_tree = None)
if dir != '/':
for path_component in dirp[len(parts):]:
dir_name, fs_path = path_component
if fs_path:
meta = metadata.from_path(fs_path)
else:
meta = metadata.Metadata()
_push(dir_name, meta)
if not file:
# no filename portion means this is a subdir. But
# sub/parentdirectories already handled in the pop/push() part above.
oldtree = already_saved(ent) # may be None
newtree = _pop(force_tree = oldtree)
if not oldtree:
if lastskip_name and lastskip_name.startswith(ent.name):
ent.invalidate()
else:
ent.validate(GIT_MODE_TREE, newtree)
ent.repack()
if exists and wasmissing:
count += oldsize
def test_index_dirty(tmpdir):
orig_cwd = os.getcwd()
try:
os.chdir(tmpdir)
default_meta = metadata.Metadata()
with index.MetaStoreWriter(b'index.meta.tmp') as ms1, \
index.MetaStoreWriter(b'index2.meta.tmp') as ms2, \
index.MetaStoreWriter(b'index3.meta.tmp') as ms3:
meta_ofs1 = ms1.store(default_meta)
meta_ofs2 = ms2.store(default_meta)
meta_ofs3 = ms3.store(default_meta)
ds = xstat.stat(lib_t_dir)
fs = xstat.stat(lib_t_dir + b'/test_index.py')
tmax = (time.time() - 1) * 10**9
with index.Writer(b'index.tmp', ms1, tmax) as w1, \
index.Writer(b'index2.tmp', ms2, tmax) as w2, \
index.Writer(b'index3.tmp', ms3, tmax) as w3:
w1.add(b'/a/b/x', fs, meta_ofs1)
w1.add(b'/a/b/c', fs, meta_ofs1)
w1.add(b'/a/b/', ds, meta_ofs1)
w1.add(b'/a/', ds, meta_ofs1)
#w1.close()
WVPASS()
w2.add(b'/a/b/n/2', fs, meta_ofs2)
#w2.close()
WVPASS()
w3.add(b'/a/c/n/3', fs, meta_ofs3)
#w3.close()
WVPASS()
with w1.new_reader() as r1, \
w2.new_reader() as r2, \
w3.new_reader() as r3:
WVPASS()
r1all = [e.name for e in r1]
WVPASSEQ(r1all,
[b'/a/b/x', b'/a/b/c', b'/a/b/', b'/a/', b'/'])
r2all = [e.name for e in r2]
WVPASSEQ(r2all,
[b'/a/b/n/2', b'/a/b/n/', b'/a/b/', b'/a/', b'/'])
r3all = [e.name for e in r3]
WVPASSEQ(r3all,
[b'/a/c/n/3', b'/a/c/n/', b'/a/c/', b'/a/', b'/'])
all = [e.name for e in index.merge(r2, r1, r3)]
WVPASSEQ(all, [
b'/a/c/n/3', b'/a/c/n/', b'/a/c/', b'/a/b/x',
b'/a/b/n/2', b'/a/b/n/', b'/a/b/c', b'/a/b/', b'/a/',
b'/'
])
fake_validate(r1)
dump(r1)
print([hex(e.flags) for e in r1])
WVPASSEQ([e.name for e in r1 if e.is_valid()], r1all)
WVPASSEQ([e.name for e in r1 if not e.is_valid()], [])
WVPASSEQ([
e.name
for e in index.merge(r2, r1, r3) if not e.is_valid()
], [
b'/a/c/n/3', b'/a/c/n/', b'/a/c/', b'/a/b/n/2',
b'/a/b/n/', b'/a/b/', b'/a/', b'/'
])
expect_invalid = [b'/'] + r2all + r3all
expect_real = (set(r1all) - set(r2all) - set(r3all)) \
| set([b'/a/b/n/2', b'/a/c/n/3'])
dump(index.merge(r2, r1, r3))
for e in index.merge(r2, r1, r3):
print(e.name, hex(e.flags), e.ctime)
eiv = e.name in expect_invalid
er = e.name in expect_real
WVPASSEQ(eiv, not e.is_valid())
WVPASSEQ(er, e.is_real())
fake_validate(r2, r3)
dump(index.merge(r2, r1, r3))
WVPASSEQ([
e.name
for e in index.merge(r2, r1, r3) if not e.is_valid()
], [])
e = eget(index.merge(r2, r1, r3), b'/a/b/c')
e.invalidate()
e.repack()
dump(index.merge(r2, r1, r3))
WVPASSEQ([
e.name
for e in index.merge(r2, r1, r3) if not e.is_valid()
], [b'/a/b/c', b'/a/b/', b'/a/', b'/'])
finally:
os.chdir(orig_cwd)
def save_tree(opt, reader, hlink_db, msr, w):
# Metadata is stored in a file named .bupm in each directory. The
# first metadata entry will be the metadata for the current directory.
# The remaining entries will be for each of the other directory
# elements, in the order they're listed in the index.
#
# Since the git tree elements are sorted according to
# git.shalist_item_sort_key, the metalist items are accumulated as
# (sort_key, metadata) tuples, and then sorted when the .bupm file is
# created. The sort_key should have been computed using the element's
# mangled name and git mode (after hashsplitting), but the code isn't
# actually doing that but rather uses the element's real name and mode.
# This makes things a bit more difficult when reading it back, see
# vfs.ordered_tree_entries().
# Maintain a stack of information representing the current location in
# the archive being constructed. The current path is recorded in
# parts, which will be something like
# [StackDir(name=''), StackDir(name='home'), StackDir(name='someuser')],
# and the accumulated content and metadata for files in the dirs is stored
# in the .items member of the StackDir.
stack = []
def _push(part, metadata):
# Enter a new archive directory -- make it the current directory.
item = StackDir(part, metadata)
stack.append(item)
def _pop(force_tree=None, dir_metadata=None):
# Leave the current archive directory and add its tree to its parent.
item = stack.pop()
# FIXME: only test if collision is possible (i.e. given --strip, etc.)?
if force_tree:
tree = force_tree
else:
names_seen = set()
clean_list = []
for x in item.items:
name = x.name
if name in names_seen:
parent_path = b'/'.join(x.name for x in stack) + b'/'
add_error('error: ignoring duplicate path %s in %s'
% (path_msg(name), path_msg(parent_path)))
else:
names_seen.add(name)
clean_list.append(x)
# if set, overrides the original metadata pushed for this dir.
if dir_metadata is None:
dir_metadata = item.meta
metalist = [(b'', dir_metadata)]
metalist += [(git.shalist_item_sort_key((entry.mode, entry.name, None)),
entry.meta)
for entry in clean_list if entry.mode != GIT_MODE_TREE]
metalist.sort(key = lambda x: x[0])
metadata = BytesIO(b''.join(m[1].encode() for m in metalist))
mode, id = hashsplit.split_to_blob_or_tree(w.new_blob, w.new_tree,
[metadata],
keep_boundaries=False)
shalist = [(mode, b'.bupm', id)]
shalist += [(entry.gitmode,
git.mangle_name(entry.name, entry.mode, entry.gitmode),
entry.oid)
for entry in clean_list]
tree = w.new_tree(shalist)
if stack:
stack[-1].append(item.name, GIT_MODE_TREE, GIT_MODE_TREE, tree, None)
return tree
# Hack around lack of nonlocal vars in python 2
_nonlocal = {}
_nonlocal['count'] = 0
_nonlocal['subcount'] = 0
_nonlocal['lastremain'] = None
def progress_report(n):
_nonlocal['subcount'] += n
cc = _nonlocal['count'] + _nonlocal['subcount']
pct = total and (cc*100.0/total) or 0
now = time.time()
elapsed = now - tstart
kps = elapsed and int(cc/1024./elapsed)
kps_frac = 10 ** int(math.log(kps+1, 10) - 1)
kps = int(kps/kps_frac)*kps_frac
if cc:
remain = elapsed*1.0/cc * (total-cc)
else:
remain = 0.0
if (_nonlocal['lastremain'] and (remain > _nonlocal['lastremain'])
and ((remain - _nonlocal['lastremain'])/_nonlocal['lastremain'] < 0.05)):
remain = _nonlocal['lastremain']
else:
_nonlocal['lastremain'] = remain
hours = int(remain/60/60)
mins = int(remain/60 - hours*60)
secs = int(remain - hours*60*60 - mins*60)
if elapsed < 30:
remainstr = ''
kpsstr = ''
else:
kpsstr = '%dk/s' % kps
if hours:
remainstr = '%dh%dm' % (hours, mins)
elif mins:
remainstr = '%dm%d' % (mins, secs)
else:
remainstr = '%ds' % secs
qprogress('Saving: %.2f%% (%d/%dk, %d/%d files) %s %s\r'
% (pct, cc/1024, total/1024, fcount, ftotal,
remainstr, kpsstr))
def already_saved(ent):
return ent.is_valid() and w.exists(ent.sha) and ent.sha
def wantrecurse_pre(ent):
return not already_saved(ent)
def wantrecurse_during(ent):
return not already_saved(ent) or ent.sha_missing()
def find_hardlink_target(hlink_db, ent):
if hlink_db and not stat.S_ISDIR(ent.mode) and ent.nlink > 1:
link_paths = hlink_db.node_paths(ent.dev, ent.ino)
if link_paths:
return link_paths[0]
return None
total = ftotal = 0
if opt.progress:
for transname, ent in reader.filter(opt.sources,
wantrecurse=wantrecurse_pre):
if not (ftotal % 10024):
qprogress('Reading index: %d\r' % ftotal)
exists = ent.exists()
hashvalid = already_saved(ent)
ent.set_sha_missing(not hashvalid)
if not opt.smaller or ent.size < opt.smaller:
if exists and not hashvalid:
total += ent.size
ftotal += 1
progress('Reading index: %d, done.\n' % ftotal)
hashsplit.progress_callback = progress_report
# Root collisions occur when strip or graft options map more than one
# path to the same directory (paths which originally had separate
# parents). When that situation is detected, use empty metadata for
# the parent. Otherwise, use the metadata for the common parent.
# Collision example: "bup save ... --strip /foo /foo/bar /bar".
# FIXME: Add collision tests, or handle collisions some other way.
# FIXME: Detect/handle strip/graft name collisions (other than root),
# i.e. if '/foo/bar' and '/bar' both map to '/'.
first_root = None
root_collision = None
tstart = time.time()
fcount = 0
lastskip_name = None
lastdir = b''
for transname, ent in reader.filter(opt.sources,
wantrecurse=wantrecurse_during):
(dir, file) = os.path.split(ent.name)
exists = (ent.flags & index.IX_EXISTS)
hashvalid = already_saved(ent)
wasmissing = ent.sha_missing()
oldsize = ent.size
if opt.verbose:
if not exists:
status = 'D'
elif not hashvalid:
if ent.sha == index.EMPTY_SHA:
status = 'A'
else:
status = 'M'
else:
status = ' '
if opt.verbose >= 2:
log('%s %-70s\n' % (status, path_msg(ent.name)))
elif not stat.S_ISDIR(ent.mode) and lastdir != dir:
if not lastdir.startswith(dir):
log('%s %-70s\n' % (status, path_msg(os.path.join(dir, b''))))
lastdir = dir
if opt.progress:
progress_report(0)
fcount += 1
if not exists:
continue
if opt.smaller and ent.size >= opt.smaller:
if exists and not hashvalid:
if opt.verbose:
log('skipping large file "%s"\n' % path_msg(ent.name))
lastskip_name = ent.name
continue
assert(dir.startswith(b'/'))
if opt.strip:
dirp = stripped_path_components(dir, opt.sources)
elif opt.strip_path:
dirp = stripped_path_components(dir, [opt.strip_path])
elif opt.grafts:
dirp = grafted_path_components(opt.grafts, dir)
else:
dirp = path_components(dir)
# At this point, dirp contains a representation of the archive
# path that looks like [(archive_dir_name, real_fs_path), ...].
# So given "bup save ... --strip /foo/bar /foo/bar/baz", dirp
# might look like this at some point:
# [('', '/foo/bar'), ('baz', '/foo/bar/baz'), ...].
# This dual representation supports stripping/grafting, where the
# archive path may not have a direct correspondence with the
# filesystem. The root directory is represented by an initial
# component named '', and any component that doesn't have a
# corresponding filesystem directory (due to grafting, for
# example) will have a real_fs_path of None, i.e. [('', None),
# ...].
if first_root == None:
first_root = dirp[0]
elif first_root != dirp[0]:
root_collision = True
# If switching to a new sub-tree, finish the current sub-tree.
while [x.name for x in stack] > [x[0] for x in dirp]:
_pop()
# If switching to a new sub-tree, start a new sub-tree.
for path_component in dirp[len(stack):]:
dir_name, fs_path = path_component
# Not indexed, so just grab the FS metadata or use empty metadata.
try:
meta = metadata.from_path(fs_path, normalized=True) \
if fs_path else metadata.Metadata()
except (OSError, IOError) as e:
add_error(e)
lastskip_name = dir_name
meta = metadata.Metadata()
_push(dir_name, meta)
if not file:
if len(stack) == 1:
continue # We're at the top level -- keep the current root dir
# Since there's no filename, this is a subdir -- finish it.
oldtree = already_saved(ent) # may be None
newtree = _pop(force_tree = oldtree)
if not oldtree:
if lastskip_name and lastskip_name.startswith(ent.name):
ent.invalidate()
else:
ent.validate(GIT_MODE_TREE, newtree)
ent.repack()
if exists and wasmissing:
_nonlocal['count'] += oldsize
continue
# it's not a directory
if hashvalid:
meta = msr.metadata_at(ent.meta_ofs)
meta.hardlink_target = find_hardlink_target(hlink_db, ent)
# Restore the times that were cleared to 0 in the metastore.
(meta.atime, meta.mtime, meta.ctime) = (ent.atime, ent.mtime, ent.ctime)
stack[-1].append(file, ent.mode, ent.gitmode, ent.sha, meta)
else:
id = None
hlink = find_hardlink_target(hlink_db, ent)
try:
meta = metadata.from_path(ent.name, hardlink_target=hlink,
normalized=True,
after_stat=after_nondir_metadata_stat)
except (OSError, IOError) as e:
add_error(e)
lastskip_name = ent.name
continue
if stat.S_IFMT(ent.mode) != stat.S_IFMT(meta.mode):
# The mode changed since we indexed the file, this is bad.
# This can cause two issues:
# 1) We e.g. think the file is a regular file, but now it's
# something else (a device, socket, FIFO or symlink, etc.)
# and _read_ from it when we shouldn't.
# 2) We then record it as valid, but don't update the index
# metadata, and on a subsequent save it has 'hashvalid'
# but is recorded as the file type from the index, when
# the content is something else ...
# Avoid all of these consistency issues by just skipping such
# things - it really ought to not happen anyway.
add_error("%s: mode changed since indexing, skipping." % path_msg(ent.name))
lastskip_name = ent.name
continue
if stat.S_ISREG(ent.mode):
try:
# If the file changes while we're reading it, then our reading
# may stop at some point, but the stat() above may have gotten
# a different size already. Recalculate the meta size so that
# the repository records the accurate size in the metadata, even
# if the other stat() data might be slightly older than the file
# content (which we can't fix, this is inherently racy, but we
# can prevent the size mismatch.)
meta.size = 0
def new_blob(data):
meta.size += len(data)
return w.new_blob(data)
before_saving_regular_file(ent.name)
with hashsplit.open_noatime(ent.name) as f:
(mode, id) = hashsplit.split_to_blob_or_tree(
new_blob, w.new_tree, [f],
keep_boundaries=False)
except (IOError, OSError) as e:
add_error('%s: %s' % (ent.name, e))
lastskip_name = ent.name
elif stat.S_ISDIR(ent.mode):
assert(0) # handled above
elif stat.S_ISLNK(ent.mode):
mode, id = (GIT_MODE_SYMLINK, w.new_blob(meta.symlink_target))
else:
# Everything else should be fully described by its
# metadata, so just record an empty blob, so the paths
# in the tree and .bupm will match up.
(mode, id) = (GIT_MODE_FILE, w.new_blob(b''))
if id:
ent.validate(mode, id)
ent.repack()
stack[-1].append(file, ent.mode, ent.gitmode, id, meta)
if exists and wasmissing:
_nonlocal['count'] += oldsize
_nonlocal['subcount'] = 0
if opt.progress:
pct = total and _nonlocal['count']*100.0/total or 100
progress('Saving: %.2f%% (%d/%dk, %d/%d files), done. \n'
% (pct, _nonlocal['count']/1024, total/1024, fcount, ftotal))
while len(stack) > 1: # _pop() all the parts above the root
_pop()
# Finish the root directory.
# When there's a collision, use empty metadata for the root.
tree = _pop(dir_metadata = metadata.Metadata() if root_collision else None)
return tree