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