lib/bup/hashsplit.py

   1
   2 from __future__ import absolute_import
   3 import io, math, os
   4
   5 from bup import _helpers, compat, helpers
   6 from bup.compat import buffer, join_bytes
   7 from bup.helpers import sc_page_size
   8
   9
  10 _fmincore = getattr(helpers, 'fmincore', None)
  11
  12 BLOB_MAX = 8192*4   # 8192 is the "typical" blob size for bupsplit
  13 BLOB_READ_SIZE = 1024*1024
  14 MAX_PER_TREE = 256
  15 progress_callback = None
  16 fanout = 16
  17
  18 GIT_MODE_FILE = 0o100644
  19 GIT_MODE_TREE = 0o40000
  20 GIT_MODE_SYMLINK = 0o120000
  21
  22 # The purpose of this type of buffer is to avoid copying on peek(), get(),
  23 # and eat().  We do copy the buffer contents on put(), but that should
  24 # be ok if we always only put() large amounts of data at a time.
  25 class Buf:
  26     def __init__(self):
  27         self.data = b''
  28         self.start = 0
  29
  30     def put(self, s):
  31         if s:
  32             self.data = join_bytes(buffer(self.data, self.start), s)
  33             self.start = 0
  34
  35     def peek(self, count):
  36         return buffer(self.data, self.start, count)
  37
  38     def eat(self, count):
  39         self.start += count
  40
  41     def get(self, count):
  42         v = buffer(self.data, self.start, count)
  43         self.start += count
  44         return v
  45
  46     def used(self):
  47         return len(self.data) - self.start
  48
  49
  50 def _fadvise_pages_done(fd, first_page, count):
  51     assert(first_page >= 0)
  52     assert(count >= 0)
  53     if count > 0:
  54         _helpers.fadvise_done(fd,
  55                               first_page * sc_page_size,
  56                               count * sc_page_size)
  57
  58
  59 def _nonresident_page_regions(status_bytes, incore_mask, max_region_len=None):
  60     """Return (start_page, count) pairs in ascending start_page order for
  61     each contiguous region of nonresident pages indicated by the
  62     mincore() status_bytes.  Limit the number of pages in each region
  63     to max_region_len."""
  64     assert(max_region_len is None or max_region_len > 0)
  65     start = None
  66     for i, x in enumerate(status_bytes):
  67         in_core = x & incore_mask
  68         if start is None:
  69             if not in_core:
  70                 start = i
  71         else:
  72             count = i - start
  73             if in_core:
  74                 yield (start, count)
  75                 start = None
  76             elif max_region_len and count >= max_region_len:
  77                 yield (start, count)
  78                 start = i
  79     if start is not None:
  80         yield (start, len(status_bytes) - start)
  81
  82
  83 def _uncache_ours_upto(fd, offset, first_region, remaining_regions):
  84     """Uncache the pages of fd indicated by first_region and
  85     remaining_regions that are before offset, where each region is a
  86     (start_page, count) pair.  The final region must have a start_page
  87     of None."""
  88     rstart, rlen = first_region
  89     while rstart is not None and (rstart + rlen) * sc_page_size <= offset:
  90         _fadvise_pages_done(fd, rstart, rlen)
  91         rstart, rlen = next(remaining_regions, (None, None))
  92     return (rstart, rlen)
  93
  94
  95 def readfile_iter(files, progress=None):
  96     for filenum,f in enumerate(files):
  97         ofs = 0
  98         b = ''
  99         fd = rpr = rstart = rlen = None
 100         if _fmincore and hasattr(f, 'fileno'):
 101             try:
 102                 fd = f.fileno()
 103             except io.UnsupportedOperation:
 104                 pass
 105             if fd:
 106                 mcore = _fmincore(fd)
 107                 if mcore:
 108                     max_chunk = max(1, (8 * 1024 * 1024) / sc_page_size)
 109                     rpr = _nonresident_page_regions(mcore, helpers.MINCORE_INCORE,
 110                                                     max_chunk)
 111                     rstart, rlen = next(rpr, (None, None))
 112         while 1:
 113             if progress:
 114                 progress(filenum, len(b))
 115             b = f.read(BLOB_READ_SIZE)
 116             ofs += len(b)
 117             if rpr:
 118                 rstart, rlen = _uncache_ours_upto(fd, ofs, (rstart, rlen), rpr)
 119             if not b:
 120                 break
 121             yield b
 122         if rpr:
 123             rstart, rlen = _uncache_ours_upto(fd, ofs, (rstart, rlen), rpr)
 124
 125
 126 def _splitbuf(buf, basebits, fanbits):
 127     while 1:
 128         b = buf.peek(buf.used())
 129         (ofs, bits) = _helpers.splitbuf(b)
 130         if ofs:
 131             if ofs > BLOB_MAX:
 132                 ofs = BLOB_MAX
 133                 level = 0
 134             else:
 135                 level = (bits-basebits)//fanbits  # integer division
 136             buf.eat(ofs)
 137             yield buffer(b, 0, ofs), level
 138         else:
 139             break
 140     while buf.used() >= BLOB_MAX:
 141         # limit max blob size
 142         yield buf.get(BLOB_MAX), 0
 143
 144
 145 def _hashsplit_iter(files, progress):
 146     assert(BLOB_READ_SIZE > BLOB_MAX)
 147     basebits = _helpers.blobbits()
 148     fanbits = int(math.log(fanout or 128, 2))
 149     buf = Buf()
 150     for inblock in readfile_iter(files, progress):
 151         buf.put(inblock)
 152         for buf_and_level in _splitbuf(buf, basebits, fanbits):
 153             yield buf_and_level
 154     if buf.used():
 155         yield buf.get(buf.used()), 0
 156
 157
 158 def _hashsplit_iter_keep_boundaries(files, progress):
 159     for real_filenum,f in enumerate(files):
 160         if progress:
 161             def prog(filenum, nbytes):
 162                 # the inner _hashsplit_iter doesn't know the real file count,
 163                 # so we'll replace it here.
 164                 return progress(real_filenum, nbytes)
 165         else:
 166             prog = None
 167         for buf_and_level in _hashsplit_iter([f], progress=prog):
 168             yield buf_and_level
 169
 170
 171 def hashsplit_iter(files, keep_boundaries, progress):
 172     if keep_boundaries:
 173         return _hashsplit_iter_keep_boundaries(files, progress)
 174     else:
 175         return _hashsplit_iter(files, progress)
 176
 177
 178 total_split = 0
 179 def split_to_blobs(makeblob, files, keep_boundaries, progress):
 180     global total_split
 181     for (blob, level) in hashsplit_iter(files, keep_boundaries, progress):
 182         sha = makeblob(blob)
 183         total_split += len(blob)
 184         if progress_callback:
 185             progress_callback(len(blob))
 186         yield (sha, len(blob), level)
 187
 188
 189 def _make_shalist(l):
 190     ofs = 0
 191     l = list(l)
 192     total = sum(size for mode,sha,size, in l)
 193     vlen = len('%x' % total)
 194     shalist = []
 195     for (mode, sha, size) in l:
 196         shalist.append((mode, '%0*x' % (vlen,ofs), sha))
 197         ofs += size
 198     assert(ofs == total)
 199     return (shalist, total)
 200
 201
 202 def _squish(maketree, stacks, n):
 203     i = 0
 204     while i < n or len(stacks[i]) >= MAX_PER_TREE:
 205         while len(stacks) <= i+1:
 206             stacks.append([])
 207         if len(stacks[i]) == 1:
 208             stacks[i+1] += stacks[i]
 209         elif stacks[i]:
 210             (shalist, size) = _make_shalist(stacks[i])
 211             tree = maketree(shalist)
 212             stacks[i+1].append((GIT_MODE_TREE, tree, size))
 213         stacks[i] = []
 214         i += 1
 215
 216
 217 def split_to_shalist(makeblob, maketree, files,
 218                      keep_boundaries, progress=None):
 219     sl = split_to_blobs(makeblob, files, keep_boundaries, progress)
 220     assert(fanout != 0)
 221     if not fanout:
 222         shal = []
 223         for (sha,size,level) in sl:
 224             shal.append((GIT_MODE_FILE, sha, size))
 225         return _make_shalist(shal)[0]
 226     else:
 227         stacks = [[]]
 228         for (sha,size,level) in sl:
 229             stacks[0].append((GIT_MODE_FILE, sha, size))
 230             _squish(maketree, stacks, level)
 231         #log('stacks: %r\n' % [len(i) for i in stacks])
 232         _squish(maketree, stacks, len(stacks)-1)
 233         #log('stacks: %r\n' % [len(i) for i in stacks])
 234         return _make_shalist(stacks[-1])[0]
 235
 236
 237 def split_to_blob_or_tree(makeblob, maketree, files,
 238                           keep_boundaries, progress=None):
 239     shalist = list(split_to_shalist(makeblob, maketree,
 240                                     files, keep_boundaries, progress))
 241     if len(shalist) == 1:
 242         return (shalist[0][0], shalist[0][2])
 243     elif len(shalist) == 0:
 244         return (GIT_MODE_FILE, makeblob(''))
 245     else:
 246         return (GIT_MODE_TREE, maketree(shalist))
 247
 248
 249 def open_noatime(name):
 250     fd = _helpers.open_noatime(name)
 251     try:
 252         return os.fdopen(fd, 'rb', 1024*1024)
 253     except:
 254         try:
 255             os.close(fd)
 256         except:
 257             pass
 258         raise