#define _LARGEFILE64_SOURCE 1
+#define PY_SSIZE_T_CLEAN 1
#undef NDEBUG
-#include "bupsplit.h"
+#include "../../config/config.h"
+
+// According to Python, its header has to go first:
+// http://docs.python.org/2/c-api/intro.html#include-files
#include <Python.h>
+
#include <assert.h>
#include <errno.h>
#include <fcntl.h>
#include <arpa/inet.h>
#include <stdint.h>
-#include <unistd.h>
#include <stdlib.h>
#include <stdio.h>
+#include <sys/mman.h>
-#ifdef linux
+#ifdef HAVE_SYS_TYPES_H
+#include <sys/types.h>
+#endif
+#ifdef HAVE_SYS_STAT_H
+#include <sys/stat.h>
+#endif
+#ifdef HAVE_UNISTD_H
+#include <unistd.h>
+#endif
+
+#ifdef HAVE_LINUX_FS_H
#include <linux/fs.h>
+#endif
+#ifdef HAVE_SYS_IOCTL_H
#include <sys/ioctl.h>
-#include <sys/stat.h>
-#include <sys/time.h>
+#endif
+
+#include "bupsplit.h"
+
+#if defined(FS_IOC_GETFLAGS) && defined(FS_IOC_SETFLAGS)
+#define BUP_HAVE_FILE_ATTRS 1
+#endif
+
+/*
+ * Check for incomplete UTIMENSAT support (NetBSD 6), and if so,
+ * pretend we don't have it.
+ */
+#if !defined(AT_FDCWD) || !defined(AT_SYMLINK_NOFOLLOW)
+#undef HAVE_UTIMENSAT
+#endif
+
+#ifndef FS_NOCOW_FL
+// Of course, this assumes it's a bitfield value.
+#define FS_NOCOW_FL 0
#endif
static int istty2 = 0;
+
+#ifndef htonll
+// This function should technically be macro'd out if it's going to be used
+// more than ocasionally. As of this writing, it'll actually never be called
+// in real world bup scenarios (because our packs are < MAX_INT bytes).
+static uint64_t htonll(uint64_t value)
+{
+ static const int endian_test = 42;
+
+ if (*(char *)&endian_test == endian_test) // LSB-MSB
+ return ((uint64_t)htonl(value & 0xFFFFFFFF) << 32) | htonl(value >> 32);
+ return value; // already in network byte order MSB-LSB
+}
+#endif
+
+
+#define INTEGRAL_ASSIGNMENT_FITS(dest, src) \
+ ({ \
+ *(dest) = (src); \
+ *(dest) == (src) && (*(dest) < 1) == ((src) < 1); \
+ })
+
+
+// At the moment any code that calls INTGER_TO_PY() will have to
+// disable -Wtautological-compare for clang. See below.
+
+#define INTEGER_TO_PY(x) \
+ (((x) >= 0) ? PyLong_FromUnsignedLongLong(x) : PyLong_FromLongLong(x))
+
+
+static int bup_ulong_from_pyint(unsigned long *x, PyObject *py,
+ const char *name)
+{
+ const long tmp = PyInt_AsLong(py);
+ if (tmp == -1 && PyErr_Occurred())
+ {
+ if (PyErr_ExceptionMatches(PyExc_OverflowError))
+ PyErr_Format(PyExc_OverflowError, "%s too big for unsigned long",
+ name);
+ return 0;
+ }
+ if (tmp < 0)
+ {
+ PyErr_Format(PyExc_OverflowError,
+ "negative %s cannot be converted to unsigned long", name);
+ return 0;
+ }
+ *x = tmp;
+ return 1;
+}
+
+
+static int bup_ulong_from_py(unsigned long *x, PyObject *py, const char *name)
+{
+ if (PyInt_Check(py))
+ return bup_ulong_from_pyint(x, py, name);
+
+ if (!PyLong_Check(py))
+ {
+ PyErr_Format(PyExc_TypeError, "expected integer %s", name);
+ return 0;
+ }
+
+ const unsigned long tmp = PyLong_AsUnsignedLong(py);
+ if (PyErr_Occurred())
+ {
+ if (PyErr_ExceptionMatches(PyExc_OverflowError))
+ PyErr_Format(PyExc_OverflowError, "%s too big for unsigned long",
+ name);
+ return 0;
+ }
+ *x = tmp;
+ return 1;
+}
+
+
+static int bup_uint_from_py(unsigned int *x, PyObject *py, const char *name)
+{
+ unsigned long tmp;
+ if (!bup_ulong_from_py(&tmp, py, name))
+ return 0;
+
+ if (tmp > UINT_MAX)
+ {
+ PyErr_Format(PyExc_OverflowError, "%s too big for unsigned int", name);
+ return 0;
+ }
+ *x = tmp;
+ return 1;
+}
+
+static int bup_ullong_from_py(unsigned PY_LONG_LONG *x, PyObject *py,
+ const char *name)
+{
+ if (PyInt_Check(py))
+ {
+ unsigned long tmp;
+ if (bup_ulong_from_pyint(&tmp, py, name))
+ {
+ *x = tmp;
+ return 1;
+ }
+ return 0;
+ }
+
+ if (!PyLong_Check(py))
+ {
+ PyErr_Format(PyExc_TypeError, "integer argument expected for %s", name);
+ return 0;
+ }
+
+ const unsigned PY_LONG_LONG tmp = PyLong_AsUnsignedLongLong(py);
+ if (tmp == (unsigned long long) -1 && PyErr_Occurred())
+ {
+ if (PyErr_ExceptionMatches(PyExc_OverflowError))
+ PyErr_Format(PyExc_OverflowError,
+ "%s too big for unsigned long long", name);
+ return 0;
+ }
+ *x = tmp;
+ return 1;
+}
+
+
// Probably we should use autoconf or something and set HAVE_PY_GETARGCARGV...
#if __WIN32__ || __CYGWIN__
#endif // not __WIN32__ or __CYGWIN__
+static unsigned long long count_leading_zeros(const unsigned char * const buf,
+ unsigned long long len)
+{
+ const unsigned char *cur = buf;
+ while(len-- && *cur == 0)
+ cur++;
+ return cur - buf;
+}
+
+
+static int write_all(int fd, const void *buf, const size_t count)
+{
+ size_t written = 0;
+ while (written < count)
+ {
+ const ssize_t rc = write(fd, buf + written, count - written);
+ if (rc == -1)
+ return -1;
+ written += rc;
+ }
+ return 0;
+}
+
+
+static int uadd(unsigned long long *dest,
+ const unsigned long long x,
+ const unsigned long long y)
+{
+ const unsigned long long result = x + y;
+ if (result < x || result < y)
+ return 0;
+ *dest = result;
+ return 1;
+}
+
+
+static PyObject *bup_write_sparsely(PyObject *self, PyObject *args)
+{
+ int fd;
+ unsigned char *buf = NULL;
+ Py_ssize_t sbuf_len;
+ PyObject *py_min_sparse_len, *py_prev_sparse_len;
+ if (!PyArg_ParseTuple(args, "it#OO",
+ &fd, &buf, &sbuf_len,
+ &py_min_sparse_len, &py_prev_sparse_len))
+ return NULL;
+ unsigned long long min_sparse_len, prev_sparse_len, buf_len;
+ if (!bup_ullong_from_py(&min_sparse_len, py_min_sparse_len, "min_sparse_len"))
+ return NULL;
+ if (!bup_ullong_from_py(&prev_sparse_len, py_prev_sparse_len, "prev_sparse_len"))
+ return NULL;
+ if (sbuf_len < 0)
+ return PyErr_Format(PyExc_ValueError, "negative bufer length");
+ if (!INTEGRAL_ASSIGNMENT_FITS(&buf_len, sbuf_len))
+ return PyErr_Format(PyExc_OverflowError, "buffer length too large");
+
+ // For now, there are some cases where we just give up if the
+ // values are too large, but we could try to break up the relevant
+ // operations into chunks.
+
+ // Deal with preceding zeros. Just make them sparse, along with
+ // any leading zeros in buf, even if the region's not >= min,
+ // since the alternative is a potentially extra small write.
+ if (prev_sparse_len)
+ {
+ const unsigned long long zeros = count_leading_zeros(buf, buf_len);
+ unsigned long long new_sparse_len = 0;
+ if (!uadd(&new_sparse_len, prev_sparse_len, zeros))
+ return PyErr_Format (PyExc_OverflowError, "sparse region too large");
+ if (zeros == buf_len)
+ return PyLong_FromUnsignedLongLong(new_sparse_len);
+
+ off_t new_off;
+ if (!INTEGRAL_ASSIGNMENT_FITS(&new_off, new_sparse_len))
+ return PyErr_Format(PyExc_OverflowError,
+ "sparse region too large for seek");
+ const off_t off = lseek(fd, new_off, SEEK_CUR);
+ if (off == -1)
+ return PyErr_SetFromErrno(PyExc_IOError);
+ buf += zeros;
+ buf_len -= zeros;
+ }
+
+ int rc;
+ unsigned long long unexamined = buf_len;
+ unsigned char *block_start = buf, *cur = buf;
+ while(unexamined)
+ {
+ const unsigned long long zeros = count_leading_zeros(cur, unexamined);
+ assert(zeros <= unexamined);
+ unexamined -= zeros;
+ if (unexamined == 0) // Runs off the end.
+ {
+ rc = write_all(fd, block_start, cur - block_start);
+ if (rc)
+ return PyErr_SetFromErrno(PyExc_IOError);
+ return PyLong_FromUnsignedLongLong(zeros);
+ }
+ cur += zeros;
+ if (zeros >= min_sparse_len)
+ {
+ off_t new_off;
+ if (!INTEGRAL_ASSIGNMENT_FITS(&new_off, zeros))
+ return PyErr_Format(PyExc_ValueError,
+ "zero count overflows off_t");
+ off_t off = lseek(fd, new_off, SEEK_CUR);
+ if (off == -1)
+ return PyErr_SetFromErrno(PyExc_IOError);
+ block_start = cur;
+ }
+ while (unexamined && *cur != 0)
+ {
+ cur++; unexamined--;
+ }
+ }
+ rc = write_all(fd, block_start, cur - block_start);
+ if (rc)
+ return PyErr_SetFromErrno(PyExc_IOError);
+ return PyInt_FromLong(0);
+}
+
+
static PyObject *selftest(PyObject *self, PyObject *args)
{
if (!PyArg_ParseTuple(args, ""))
static PyObject *splitbuf(PyObject *self, PyObject *args)
{
unsigned char *buf = NULL;
- int len = 0, out = 0, bits = -1;
+ Py_ssize_t len = 0;
+ int out = 0, bits = -1;
if (!PyArg_ParseTuple(args, "t#", &buf, &len))
return NULL;
+ assert(len <= INT_MAX);
out = bupsplit_find_ofs(buf, len, &bits);
if (out) assert(bits >= BUP_BLOBBITS);
return Py_BuildValue("ii", out, bits);
static PyObject *bitmatch(PyObject *self, PyObject *args)
{
unsigned char *buf1 = NULL, *buf2 = NULL;
- int len1 = 0, len2 = 0;
- int byte, bit;
+ Py_ssize_t len1 = 0, len2 = 0;
+ Py_ssize_t byte;
+ int bit;
if (!PyArg_ParseTuple(args, "t#t#", &buf1, &len1, &buf2, &len2))
return NULL;
}
}
+ assert(byte <= (INT_MAX >> 3));
return Py_BuildValue("i", byte*8 + bit);
}
static PyObject *firstword(PyObject *self, PyObject *args)
{
unsigned char *buf = NULL;
- int len = 0;
+ Py_ssize_t len = 0;
uint32_t v;
if (!PyArg_ParseTuple(args, "t#", &buf, &len))
#define BLOOM2_HEADERLEN 16
-typedef struct {
- uint32_t high;
- unsigned char low;
-} bits40_t;
-
-static void to_bloom_address_bitmask4(const bits40_t *buf,
+static void to_bloom_address_bitmask4(const unsigned char *buf,
const int nbits, uint64_t *v, unsigned char *bitmask)
{
int bit;
+ uint32_t high;
uint64_t raw, mask;
+ memcpy(&high, buf, 4);
mask = (1<<nbits) - 1;
- raw = (((uint64_t)ntohl(buf->high)) << 8) | buf->low;
+ raw = (((uint64_t)ntohl(high) << 8) | buf[4]);
bit = (raw >> (37-nbits)) & 0x7;
*v = (raw >> (40-nbits)) & mask;
*bitmask = 1 << bit;
}
-static void to_bloom_address_bitmask5(const uint32_t *buf,
+static void to_bloom_address_bitmask5(const unsigned char *buf,
const int nbits, uint32_t *v, unsigned char *bitmask)
{
int bit;
+ uint32_t high;
uint32_t raw, mask;
+ memcpy(&high, buf, 4);
mask = (1<<nbits) - 1;
- raw = ntohl(*buf);
+ raw = ntohl(high);
bit = (raw >> (29-nbits)) & 0x7;
*v = (raw >> (32-nbits)) & mask;
*bitmask = 1 << bit;
}
-#define BLOOM_SET_BIT(name, address, itype, otype) \
-static void name(unsigned char *bloom, const void *buf, const int nbits)\
+#define BLOOM_SET_BIT(name, address, otype) \
+static void name(unsigned char *bloom, const unsigned char *buf, const int nbits)\
{\
unsigned char bitmask;\
otype v;\
- address((itype *)buf, nbits, &v, &bitmask);\
+ address(buf, nbits, &v, &bitmask);\
bloom[BLOOM2_HEADERLEN+v] |= bitmask;\
}
-BLOOM_SET_BIT(bloom_set_bit4, to_bloom_address_bitmask4, bits40_t, uint64_t)
-BLOOM_SET_BIT(bloom_set_bit5, to_bloom_address_bitmask5, uint32_t, uint32_t)
+BLOOM_SET_BIT(bloom_set_bit4, to_bloom_address_bitmask4, uint64_t)
+BLOOM_SET_BIT(bloom_set_bit5, to_bloom_address_bitmask5, uint32_t)
-#define BLOOM_GET_BIT(name, address, itype, otype) \
-static int name(const unsigned char *bloom, const void *buf, const int nbits)\
+#define BLOOM_GET_BIT(name, address, otype) \
+static int name(const unsigned char *bloom, const unsigned char *buf, const int nbits)\
{\
unsigned char bitmask;\
otype v;\
- address((itype *)buf, nbits, &v, &bitmask);\
+ address(buf, nbits, &v, &bitmask);\
return bloom[BLOOM2_HEADERLEN+v] & bitmask;\
}
-BLOOM_GET_BIT(bloom_get_bit4, to_bloom_address_bitmask4, bits40_t, uint64_t)
-BLOOM_GET_BIT(bloom_get_bit5, to_bloom_address_bitmask5, uint32_t, uint32_t)
+BLOOM_GET_BIT(bloom_get_bit4, to_bloom_address_bitmask4, uint64_t)
+BLOOM_GET_BIT(bloom_get_bit5, to_bloom_address_bitmask5, uint32_t)
static PyObject *bloom_add(PyObject *self, PyObject *args)
{
unsigned char *sha = NULL, *bloom = NULL;
unsigned char *end;
- int len = 0, blen = 0, nbits = 0, k = 0;
+ Py_ssize_t len = 0, blen = 0;
+ int nbits = 0, k = 0;
if (!PyArg_ParseTuple(args, "w#s#ii", &bloom, &blen, &sha, &len, &nbits, &k))
return NULL;
return NULL;
- return Py_BuildValue("i", len/20);
+ return Py_BuildValue("n", len/20);
}
static PyObject *bloom_contains(PyObject *self, PyObject *args)
{
unsigned char *sha = NULL, *bloom = NULL;
- int len = 0, blen = 0, nbits = 0, k = 0;
+ Py_ssize_t len = 0, blen = 0;
+ int nbits = 0, k = 0;
unsigned char *end;
int steps;
v = (v >> (32-nbits)) & mask;
return v;
}
+
+
static PyObject *extract_bits(PyObject *self, PyObject *args)
{
unsigned char *buf = NULL;
- int len = 0, nbits = 0;
+ Py_ssize_t len = 0;
+ int nbits = 0;
if (!PyArg_ParseTuple(args, "t#i", &buf, &len, &nbits))
return NULL;
struct sha {
unsigned char bytes[20];
};
+
+
struct idx {
unsigned char *map;
struct sha *cur;
struct sha *end;
uint32_t *cur_name;
- long bytes;
+ Py_ssize_t bytes;
int name_base;
};
static PyObject *merge_into(PyObject *self, PyObject *args)
{
- PyObject *ilist = NULL;
+ PyObject *py_total, *ilist = NULL;
unsigned char *fmap = NULL;
struct sha *sha_ptr, *sha_start = NULL;
uint32_t *table_ptr, *name_ptr, *name_start;
struct idx **idxs = NULL;
- int flen = 0, bits = 0, i;
- uint32_t total, count, prefix;
+ Py_ssize_t flen = 0;
+ int bits = 0, i;
+ unsigned int total;
+ uint32_t count, prefix;
int num_i;
int last_i;
- if (!PyArg_ParseTuple(args, "w#iIO", &fmap, &flen, &bits, &total, &ilist))
+ if (!PyArg_ParseTuple(args, "w#iOO",
+ &fmap, &flen, &bits, &py_total, &ilist))
return NULL;
+ if (!bup_uint_from_py(&total, py_total, "total"))
+ return NULL;
+
num_i = PyList_Size(ilist);
idxs = (struct idx **)PyMem_Malloc(num_i * sizeof(struct idx *));
return PyLong_FromUnsignedLong(count);
}
-// This function should technically be macro'd out if it's going to be used
-// more than ocasionally. As of this writing, it'll actually never be called
-// in real world bup scenarios (because our packs are < MAX_INT bytes).
-static uint64_t htonll(uint64_t value)
-{
- static const int endian_test = 42;
-
- if (*(char *)&endian_test == endian_test) // LSB-MSB
- return ((uint64_t)htonl(value & 0xFFFFFFFF) << 32) | htonl(value >> 32);
- return value; // already in network byte order MSB-LSB
-}
-
-#define PACK_IDX_V2_HEADERLEN 8
#define FAN_ENTRIES 256
static PyObject *write_idx(PyObject *self, PyObject *args)
{
- PyObject *pf = NULL, *idx = NULL;
+ char *filename = NULL;
+ PyObject *py_total, *idx = NULL;
PyObject *part;
- FILE *f;
unsigned char *fmap = NULL;
- int flen = 0;
- uint32_t total = 0;
+ Py_ssize_t flen = 0;
+ unsigned int total = 0;
uint32_t count;
int i, j, ofs64_count;
uint32_t *fan_ptr, *crc_ptr, *ofs_ptr;
+ uint64_t *ofs64_ptr;
struct sha *sha_ptr;
- if (!PyArg_ParseTuple(args, "Ow#OI", &pf, &fmap, &flen, &idx, &total))
+ if (!PyArg_ParseTuple(args, "sw#OO",
+ &filename, &fmap, &flen, &idx, &py_total))
return NULL;
- fan_ptr = (uint32_t *)&fmap[PACK_IDX_V2_HEADERLEN];
+ if (!bup_uint_from_py(&total, py_total, "total"))
+ return NULL;
+
+ if (PyList_Size (idx) != FAN_ENTRIES) // Check for list of the right length.
+ return PyErr_Format (PyExc_TypeError, "idx must contain %d entries",
+ FAN_ENTRIES);
+
+ const char idx_header[] = "\377tOc\0\0\0\002";
+ memcpy (fmap, idx_header, sizeof(idx_header) - 1);
+
+ fan_ptr = (uint32_t *)&fmap[sizeof(idx_header) - 1];
sha_ptr = (struct sha *)&fan_ptr[FAN_ENTRIES];
crc_ptr = (uint32_t *)&sha_ptr[total];
ofs_ptr = (uint32_t *)&crc_ptr[total];
- f = PyFile_AsFile(pf);
+ ofs64_ptr = (uint64_t *)&ofs_ptr[total];
count = 0;
ofs64_count = 0;
for (j = 0; j < plen; ++j)
{
unsigned char *sha = NULL;
- int sha_len = 0;
- uint32_t crc = 0;
- uint64_t ofs = 0;
- if (!PyArg_ParseTuple(PyList_GET_ITEM(part, j), "t#IK",
- &sha, &sha_len, &crc, &ofs))
+ Py_ssize_t sha_len = 0;
+ PyObject *crc_py, *ofs_py;
+ unsigned int crc;
+ unsigned PY_LONG_LONG ofs_ull;
+ uint64_t ofs;
+ if (!PyArg_ParseTuple(PyList_GET_ITEM(part, j), "t#OO",
+ &sha, &sha_len, &crc_py, &ofs_py))
return NULL;
+ if(!bup_uint_from_py(&crc, crc_py, "crc"))
+ return NULL;
+ if(!bup_ullong_from_py(&ofs_ull, ofs_py, "ofs"))
+ return NULL;
+ assert(crc <= UINT32_MAX);
+ assert(ofs_ull <= UINT64_MAX);
+ ofs = ofs_ull;
if (sha_len != sizeof(struct sha))
return NULL;
memcpy(sha_ptr++, sha, sizeof(struct sha));
*crc_ptr++ = htonl(crc);
if (ofs > 0x7fffffff)
{
- uint64_t nofs = htonll(ofs);
- if (fwrite(&nofs, sizeof(uint64_t), 1, f) != 1)
- return PyErr_SetFromErrno(PyExc_OSError);
+ *ofs64_ptr++ = htonll(ofs);
ofs = 0x80000000 | ofs64_count++;
}
*ofs_ptr++ = htonl((uint32_t)ofs);
}
}
+
+ int rc = msync(fmap, flen, MS_ASYNC);
+ if (rc != 0)
+ return PyErr_SetFromErrnoWithFilename(PyExc_IOError, filename);
+
return PyLong_FromUnsignedLong(count);
}
}
-static PyObject *open_noatime(PyObject *self, PyObject *args)
+static int _open_noatime(const char *filename, int attrs)
{
- char *filename = NULL;
- int attrs, attrs_noatime, fd;
- if (!PyArg_ParseTuple(args, "s", &filename))
- return NULL;
- attrs = O_RDONLY;
+ int attrs_noatime, fd;
+ attrs |= O_RDONLY;
#ifdef O_NOFOLLOW
attrs |= O_NOFOLLOW;
#endif
// just harmlessly ignore it, so this branch won't trigger)
fd = open(filename, attrs);
}
+ return fd;
+}
+
+
+static PyObject *open_noatime(PyObject *self, PyObject *args)
+{
+ char *filename = NULL;
+ int fd;
+ if (!PyArg_ParseTuple(args, "s", &filename))
+ return NULL;
+ fd = _open_noatime(filename, 0);
if (fd < 0)
return PyErr_SetFromErrnoWithFilename(PyExc_OSError, filename);
return Py_BuildValue("i", fd);
}
-#if defined(linux) && defined(FS_IOC_GETFLAGS)
+// Currently the Linux kernel and FUSE disagree over the type for
+// FS_IOC_GETFLAGS and FS_IOC_SETFLAGS. The kernel actually uses int,
+// but FUSE chose long (matching the declaration in linux/fs.h). So
+// if you use int, and then traverse a FUSE filesystem, you may
+// corrupt the stack. But if you use long, then you may get invalid
+// results on big-endian systems.
+//
+// For now, we just use long, and then disable Linux attrs entirely
+// (with a warning) in helpers.py on systems that are affected.
+
+#ifdef BUP_HAVE_FILE_ATTRS
static PyObject *bup_get_linux_file_attr(PyObject *self, PyObject *args)
{
int rc;
if (!PyArg_ParseTuple(args, "s", &path))
return NULL;
- fd = open(path, O_RDONLY | O_NONBLOCK | O_LARGEFILE | O_NOFOLLOW);
+ fd = _open_noatime(path, O_NONBLOCK);
if (fd == -1)
return PyErr_SetFromErrnoWithFilename(PyExc_OSError, path);
- attr = 0;
+ attr = 0; // Handle int/long mismatch (see above)
rc = ioctl(fd, FS_IOC_GETFLAGS, &attr);
if (rc == -1)
{
close(fd);
return PyErr_SetFromErrnoWithFilename(PyExc_OSError, path);
}
-
close(fd);
- return Py_BuildValue("k", attr);
+ assert(attr <= UINT_MAX); // Kernel type is actually int
+ return PyLong_FromUnsignedLong(attr);
}
+#endif /* def BUP_HAVE_FILE_ATTRS */
+
+#ifdef BUP_HAVE_FILE_ATTRS
static PyObject *bup_set_linux_file_attr(PyObject *self, PyObject *args)
{
int rc;
- unsigned long attr;
+ unsigned long orig_attr;
+ unsigned int attr;
char *path;
+ PyObject *py_attr;
int fd;
- if (!PyArg_ParseTuple(args, "sk", &path, &attr))
+ if (!PyArg_ParseTuple(args, "sO", &path, &py_attr))
+ return NULL;
+
+ if (!bup_uint_from_py(&attr, py_attr, "attr"))
return NULL;
fd = open(path, O_RDONLY | O_NONBLOCK | O_LARGEFILE | O_NOFOLLOW);
if (fd == -1)
return PyErr_SetFromErrnoWithFilename(PyExc_OSError, path);
+ // Restrict attr to modifiable flags acdeijstuADST -- see
+ // chattr(1) and the e2fsprogs source. Letter to flag mapping is
+ // in pf.c flags_array[].
+ attr &= FS_APPEND_FL | FS_COMPR_FL | FS_NODUMP_FL | FS_EXTENT_FL
+ | FS_IMMUTABLE_FL | FS_JOURNAL_DATA_FL | FS_SECRM_FL | FS_NOTAIL_FL
+ | FS_UNRM_FL | FS_NOATIME_FL | FS_DIRSYNC_FL | FS_SYNC_FL
+ | FS_TOPDIR_FL | FS_NOCOW_FL;
+
+ // The extents flag can't be removed, so don't (see chattr(1) and chattr.c).
+ orig_attr = 0; // Handle int/long mismatch (see above)
+ rc = ioctl(fd, FS_IOC_GETFLAGS, &orig_attr);
+ assert(orig_attr <= UINT_MAX); // Kernel type is actually int
+ if (rc == -1)
+ {
+ close(fd);
+ return PyErr_SetFromErrnoWithFilename(PyExc_OSError, path);
+ }
+ attr |= ((unsigned int) orig_attr) & FS_EXTENT_FL;
+
rc = ioctl(fd, FS_IOC_SETFLAGS, &attr);
if (rc == -1)
{
}
close(fd);
- return Py_BuildValue("i", 1);
+ return Py_BuildValue("O", Py_None);
}
-#endif /* def linux */
+#endif /* def BUP_HAVE_FILE_ATTRS */
+
+#ifndef HAVE_UTIMENSAT
+#ifndef HAVE_UTIMES
+#error "cannot find utimensat or utimes()"
+#endif
+#ifndef HAVE_LUTIMES
+#error "cannot find utimensat or lutimes()"
+#endif
+#endif
-#if defined(_ATFILE_SOURCE) \
- || _XOPEN_SOURCE >= 700 || _POSIX_C_SOURCE >= 200809L
-#define HAVE_BUP_UTIMENSAT 1
+#define ASSIGN_PYLONG_TO_INTEGRAL(dest, pylong, overflow) \
+ ({ \
+ int result = 0; \
+ *(overflow) = 0; \
+ const long long lltmp = PyLong_AsLongLong(pylong); \
+ if (lltmp == -1 && PyErr_Occurred()) \
+ { \
+ if (PyErr_ExceptionMatches(PyExc_OverflowError)) \
+ { \
+ const unsigned long long ulltmp = PyLong_AsUnsignedLongLong(pylong); \
+ if (ulltmp == (unsigned long long) -1 && PyErr_Occurred()) \
+ { \
+ if (PyErr_ExceptionMatches(PyExc_OverflowError)) \
+ { \
+ PyErr_Clear(); \
+ *(overflow) = 1; \
+ } \
+ } \
+ if (INTEGRAL_ASSIGNMENT_FITS((dest), ulltmp)) \
+ result = 1; \
+ else \
+ *(overflow) = 1; \
+ } \
+ } \
+ else \
+ { \
+ if (INTEGRAL_ASSIGNMENT_FITS((dest), lltmp)) \
+ result = 1; \
+ else \
+ *(overflow) = 1; \
+ } \
+ result; \
+ })
+
+
+#ifdef HAVE_UTIMENSAT
static PyObject *bup_utimensat(PyObject *self, PyObject *args)
{
- int rc, dirfd, flags;
+ int rc;
+ int fd, flag;
char *path;
- long access, access_ns, modification, modification_ns;
+ PyObject *access_py, *modification_py;
struct timespec ts[2];
- if (!PyArg_ParseTuple(args, "is((ll)(ll))i",
- &dirfd,
+ if (!PyArg_ParseTuple(args, "is((Ol)(Ol))i",
+ &fd,
&path,
- &access, &access_ns,
- &modification, &modification_ns,
- &flags))
+ &access_py, &(ts[0].tv_nsec),
+ &modification_py, &(ts[1].tv_nsec),
+ &flag))
return NULL;
- if (isnan(access))
- {
- PyErr_SetString(PyExc_ValueError, "access time is NaN");
- return NULL;
- }
- else if (isinf(access))
- {
- PyErr_SetString(PyExc_ValueError, "access time is infinite");
- return NULL;
- }
- else if (isnan(modification))
+ int overflow;
+ if (!ASSIGN_PYLONG_TO_INTEGRAL(&(ts[0].tv_sec), access_py, &overflow))
{
- PyErr_SetString(PyExc_ValueError, "modification time is NaN");
+ if (overflow)
+ PyErr_SetString(PyExc_ValueError,
+ "unable to convert access time seconds for utimensat");
return NULL;
}
- else if (isinf(modification))
+ if (!ASSIGN_PYLONG_TO_INTEGRAL(&(ts[1].tv_sec), modification_py, &overflow))
{
- PyErr_SetString(PyExc_ValueError, "modification time is infinite");
+ if (overflow)
+ PyErr_SetString(PyExc_ValueError,
+ "unable to convert modification time seconds for utimensat");
return NULL;
}
+ rc = utimensat(fd, path, ts, flag);
+ if (rc != 0)
+ return PyErr_SetFromErrnoWithFilename(PyExc_OSError, path);
+
+ return Py_BuildValue("O", Py_None);
+}
+
+#endif /* def HAVE_UTIMENSAT */
+
- if (isnan(access_ns))
+#if defined(HAVE_UTIMES) || defined(HAVE_LUTIMES)
+
+static int bup_parse_xutimes_args(char **path,
+ struct timeval tv[2],
+ PyObject *args)
+{
+ PyObject *access_py, *modification_py;
+ long long access_us, modification_us; // POSIX guarantees tv_usec is signed.
+
+ if (!PyArg_ParseTuple(args, "s((OL)(OL))",
+ path,
+ &access_py, &access_us,
+ &modification_py, &modification_us))
+ return 0;
+
+ int overflow;
+ if (!ASSIGN_PYLONG_TO_INTEGRAL(&(tv[0].tv_sec), access_py, &overflow))
{
- PyErr_SetString(PyExc_ValueError, "access time ns is NaN");
- return NULL;
+ if (overflow)
+ PyErr_SetString(PyExc_ValueError, "unable to convert access time seconds to timeval");
+ return 0;
}
- else if (isinf(access_ns))
+ if (!INTEGRAL_ASSIGNMENT_FITS(&(tv[0].tv_usec), access_us))
{
- PyErr_SetString(PyExc_ValueError, "access time ns is infinite");
- return NULL;
+ PyErr_SetString(PyExc_ValueError, "unable to convert access time nanoseconds to timeval");
+ return 0;
}
- else if (isnan(modification_ns))
+ if (!ASSIGN_PYLONG_TO_INTEGRAL(&(tv[1].tv_sec), modification_py, &overflow))
{
- PyErr_SetString(PyExc_ValueError, "modification time ns is NaN");
- return NULL;
+ if (overflow)
+ PyErr_SetString(PyExc_ValueError, "unable to convert modification time seconds to timeval");
+ return 0;
}
- else if (isinf(modification_ns))
+ if (!INTEGRAL_ASSIGNMENT_FITS(&(tv[1].tv_usec), modification_us))
{
- PyErr_SetString(PyExc_ValueError, "modification time ns is infinite");
- return NULL;
+ PyErr_SetString(PyExc_ValueError, "unable to convert modification time nanoseconds to timeval");
+ return 0;
}
+ return 1;
+}
+
+#endif /* defined(HAVE_UTIMES) || defined(HAVE_LUTIMES) */
+
+
+#ifdef HAVE_UTIMES
+static PyObject *bup_utimes(PyObject *self, PyObject *args)
+{
+ char *path;
+ struct timeval tv[2];
+ if (!bup_parse_xutimes_args(&path, tv, args))
+ return NULL;
+ int rc = utimes(path, tv);
+ if (rc != 0)
+ return PyErr_SetFromErrnoWithFilename(PyExc_OSError, path);
+ return Py_BuildValue("O", Py_None);
+}
+#endif /* def HAVE_UTIMES */
- ts[0].tv_sec = access;
- ts[0].tv_nsec = access_ns;
- ts[1].tv_sec = modification;
- ts[1].tv_nsec = modification_ns;
- rc = utimensat(dirfd, path, ts, flags);
+#ifdef HAVE_LUTIMES
+static PyObject *bup_lutimes(PyObject *self, PyObject *args)
+{
+ char *path;
+ struct timeval tv[2];
+ if (!bup_parse_xutimes_args(&path, tv, args))
+ return NULL;
+ int rc = lutimes(path, tv);
if (rc != 0)
return PyErr_SetFromErrnoWithFilename(PyExc_OSError, path);
- return Py_BuildValue("i", 1);
+ return Py_BuildValue("O", Py_None);
}
+#endif /* def HAVE_LUTIMES */
-#endif /* defined(_ATFILE_SOURCE)
- || _XOPEN_SOURCE >= 700 || _POSIX_C_SOURCE >= 200809L */
+#ifdef HAVE_STAT_ST_ATIM
+# define BUP_STAT_ATIME_NS(st) (st)->st_atim.tv_nsec
+# define BUP_STAT_MTIME_NS(st) (st)->st_mtim.tv_nsec
+# define BUP_STAT_CTIME_NS(st) (st)->st_ctim.tv_nsec
+#elif defined HAVE_STAT_ST_ATIMENSEC
+# define BUP_STAT_ATIME_NS(st) (st)->st_atimespec.tv_nsec
+# define BUP_STAT_MTIME_NS(st) (st)->st_mtimespec.tv_nsec
+# define BUP_STAT_CTIME_NS(st) (st)->st_ctimespec.tv_nsec
+#else
+# define BUP_STAT_ATIME_NS(st) 0
+# define BUP_STAT_MTIME_NS(st) 0
+# define BUP_STAT_CTIME_NS(st) 0
+#endif
+
+
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wtautological-compare" // For INTEGER_TO_PY().
+
+static PyObject *stat_struct_to_py(const struct stat *st,
+ const char *filename,
+ int fd)
+{
+ // We can check the known (via POSIX) signed and unsigned types at
+ // compile time, but not (easily) the unspecified types, so handle
+ // those via INTEGER_TO_PY(). Assumes ns values will fit in a
+ // long.
+ return Py_BuildValue("OKOOOOOL(Ol)(Ol)(Ol)",
+ INTEGER_TO_PY(st->st_mode),
+ (unsigned PY_LONG_LONG) st->st_ino,
+ INTEGER_TO_PY(st->st_dev),
+ INTEGER_TO_PY(st->st_nlink),
+ INTEGER_TO_PY(st->st_uid),
+ INTEGER_TO_PY(st->st_gid),
+ INTEGER_TO_PY(st->st_rdev),
+ (PY_LONG_LONG) st->st_size,
+ INTEGER_TO_PY(st->st_atime),
+ (long) BUP_STAT_ATIME_NS(st),
+ INTEGER_TO_PY(st->st_mtime),
+ (long) BUP_STAT_MTIME_NS(st),
+ INTEGER_TO_PY(st->st_ctime),
+ (long) BUP_STAT_CTIME_NS(st));
+}
-#ifdef linux /* and likely others */
+#pragma clang diagnostic pop // ignored "-Wtautological-compare"
-#define HAVE_BUP_STAT 1
static PyObject *bup_stat(PyObject *self, PyObject *args)
{
int rc;
rc = stat(filename, &st);
if (rc != 0)
return PyErr_SetFromErrnoWithFilename(PyExc_OSError, filename);
+ return stat_struct_to_py(&st, filename, 0);
+}
+
- return Py_BuildValue("kkkkkkkk"
- "(ll)"
- "(ll)"
- "(ll)",
- (unsigned long) st.st_mode,
- (unsigned long) st.st_ino,
- (unsigned long) st.st_dev,
- (unsigned long) st.st_nlink,
- (unsigned long) st.st_uid,
- (unsigned long) st.st_gid,
- (unsigned long) st.st_rdev,
- (unsigned long) st.st_size,
- (long) st.st_atime,
- (long) st.st_atim.tv_nsec,
- (long) st.st_mtime,
- (long) st.st_mtim.tv_nsec,
- (long) st.st_ctime,
- (long) st.st_ctim.tv_nsec);
-}
-
-
-#define HAVE_BUP_LSTAT 1
static PyObject *bup_lstat(PyObject *self, PyObject *args)
{
int rc;
rc = lstat(filename, &st);
if (rc != 0)
return PyErr_SetFromErrnoWithFilename(PyExc_OSError, filename);
+ return stat_struct_to_py(&st, filename, 0);
+}
+
- return Py_BuildValue("kkkkkkkk"
- "(ll)"
- "(ll)"
- "(ll)",
- (unsigned long) st.st_mode,
- (unsigned long) st.st_ino,
- (unsigned long) st.st_dev,
- (unsigned long) st.st_nlink,
- (unsigned long) st.st_uid,
- (unsigned long) st.st_gid,
- (unsigned long) st.st_rdev,
- (unsigned long) st.st_size,
- (long) st.st_atime,
- (long) st.st_atim.tv_nsec,
- (long) st.st_mtime,
- (long) st.st_mtim.tv_nsec,
- (long) st.st_ctime,
- (long) st.st_ctim.tv_nsec);
-}
-
-
-#define HAVE_BUP_FSTAT 1
static PyObject *bup_fstat(PyObject *self, PyObject *args)
{
int rc, fd;
rc = fstat(fd, &st);
if (rc != 0)
return PyErr_SetFromErrno(PyExc_OSError);
-
- return Py_BuildValue("kkkkkkkk"
- "(ll)"
- "(ll)"
- "(ll)",
- (unsigned long) st.st_mode,
- (unsigned long) st.st_ino,
- (unsigned long) st.st_dev,
- (unsigned long) st.st_nlink,
- (unsigned long) st.st_uid,
- (unsigned long) st.st_gid,
- (unsigned long) st.st_rdev,
- (unsigned long) st.st_size,
- (long) st.st_atime,
- (long) st.st_atim.tv_nsec,
- (long) st.st_mtime,
- (long) st.st_mtim.tv_nsec,
- (long) st.st_ctime,
- (long) st.st_ctim.tv_nsec);
-}
-
-#endif /* def linux */
+ return stat_struct_to_py(&st, NULL, fd);
+}
static PyMethodDef helper_methods[] = {
+ { "write_sparsely", bup_write_sparsely, METH_VARARGS,
+ "Write buf excepting zeros at the end. Return trailing zero count." },
{ "selftest", selftest, METH_VARARGS,
"Check that the rolling checksum rolls correctly (for unit tests)." },
{ "blobbits", blobbits, METH_VARARGS,
"open() the given filename for read with O_NOATIME if possible" },
{ "fadvise_done", fadvise_done, METH_VARARGS,
"Inform the kernel that we're finished with earlier parts of a file" },
-#if defined(linux) && defined(FS_IOC_GETFLAGS)
+#ifdef BUP_HAVE_FILE_ATTRS
{ "get_linux_file_attr", bup_get_linux_file_attr, METH_VARARGS,
"Return the Linux attributes for the given file." },
+#endif
+#ifdef BUP_HAVE_FILE_ATTRS
{ "set_linux_file_attr", bup_set_linux_file_attr, METH_VARARGS,
"Set the Linux attributes for the given file." },
#endif
-#ifdef HAVE_BUP_UTIMENSAT
- { "utimensat", bup_utimensat, METH_VARARGS,
- "Change file timestamps with nanosecond precision." },
+#ifdef HAVE_UTIMENSAT
+ { "bup_utimensat", bup_utimensat, METH_VARARGS,
+ "Change path timestamps with nanosecond precision (POSIX)." },
+#endif
+#ifdef HAVE_UTIMES
+ { "bup_utimes", bup_utimes, METH_VARARGS,
+ "Change path timestamps with microsecond precision." },
+#endif
+#ifdef HAVE_LUTIMES
+ { "bup_lutimes", bup_lutimes, METH_VARARGS,
+ "Change path timestamps with microsecond precision;"
+ " don't follow symlinks." },
#endif
-#ifdef HAVE_BUP_STAT
{ "stat", bup_stat, METH_VARARGS,
"Extended version of stat." },
-#endif
-#ifdef HAVE_BUP_LSTAT
{ "lstat", bup_lstat, METH_VARARGS,
"Extended version of lstat." },
-#endif
-#ifdef HAVE_BUP_FSTAT
{ "fstat", bup_fstat, METH_VARARGS,
"Extended version of fstat." },
-#endif
{ NULL, NULL, 0, NULL }, // sentinel
};
PyMODINIT_FUNC init_helpers(void)
{
+ // FIXME: migrate these tests to configure. Check against the
+ // type we're going to use when passing to python. Other stat
+ // types are tested at runtime.
+ assert(sizeof(ino_t) <= sizeof(unsigned PY_LONG_LONG));
+ assert(sizeof(off_t) <= sizeof(PY_LONG_LONG));
+ assert(sizeof(blksize_t) <= sizeof(PY_LONG_LONG));
+ assert(sizeof(blkcnt_t) <= sizeof(PY_LONG_LONG));
+ // Just be sure (relevant when passing timestamps back to Python above).
+ assert(sizeof(PY_LONG_LONG) <= sizeof(long long));
+ assert(sizeof(unsigned PY_LONG_LONG) <= sizeof(unsigned long long));
+
char *e;
PyObject *m = Py_InitModule("_helpers", helper_methods);
if (m == NULL)
return;
-#ifdef HAVE_BUP_UTIMENSAT
- PyModule_AddObject(m, "AT_FDCWD", Py_BuildValue("i", AT_FDCWD));
- PyModule_AddObject(m, "AT_SYMLINK_NOFOLLOW",
- Py_BuildValue("i", AT_SYMLINK_NOFOLLOW));
-#endif
-#ifdef HAVE_BUP_STAT
- PyModule_AddIntConstant(m, "_have_ns_fs_timestamps", 1);
-#else
- PyModule_AddIntConstant(m, "_have_ns_fs_timestamps", 0);
+
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wtautological-compare" // For INTEGER_TO_PY().
+#ifdef HAVE_UTIMENSAT
+ {
+ PyObject *value;
+ value = INTEGER_TO_PY(AT_FDCWD);
+ PyObject_SetAttrString(m, "AT_FDCWD", value);
+ Py_DECREF(value);
+ value = INTEGER_TO_PY(AT_SYMLINK_NOFOLLOW);
+ PyObject_SetAttrString(m, "AT_SYMLINK_NOFOLLOW", value);
+ Py_DECREF(value);
+ value = INTEGER_TO_PY(UTIME_NOW);
+ PyObject_SetAttrString(m, "UTIME_NOW", value);
+ Py_DECREF(value);
+ }
#endif
+ {
+ PyObject *value;
+ const long arg_max = sysconf(_SC_ARG_MAX);
+ if (arg_max == -1)
+ {
+ fprintf(stderr, "Cannot find SC_ARG_MAX, please report a bug.\n");
+ exit(1);
+ }
+ value = INTEGER_TO_PY(arg_max);
+ PyObject_SetAttrString(m, "SC_ARG_MAX", value);
+ Py_DECREF(value);
+ }
+#pragma clang diagnostic pop // ignored "-Wtautological-compare"
+
e = getenv("BUP_FORCE_TTY");
istty2 = isatty(2) || (atoi(e ? e : "0") & 2);
unpythonize_argv();