X-Git-Url: https://arthur.barton.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=lib%2Fbup%2F_helpers.c;h=9078be3da723d8e7757662950a7521bdc275754f;hb=544208fae44dfeaccb626c16f3a5c387f318d08f;hp=9634b1e6d0643fcec76ab128023c75a5626f0f5b;hpb=449d0097b35c6e516809bfa6452465089aa19038;p=bup.git diff --git a/lib/bup/_helpers.c b/lib/bup/_helpers.c index 9634b1e..9078be3 100644 --- a/lib/bup/_helpers.c +++ b/lib/bup/_helpers.c @@ -11,11 +11,15 @@ #include #include #include +#include #include #include #include -#include +#include +#ifdef HAVE_SYS_MMAN_H +#include +#endif #ifdef HAVE_SYS_TYPES_H #include #endif @@ -25,6 +29,9 @@ #ifdef HAVE_UNISTD_H #include #endif +#ifdef HAVE_SYS_TIME_H +#include +#endif #ifdef HAVE_LINUX_FS_H #include @@ -33,18 +40,197 @@ #include #endif +#ifdef HAVE_TM_TM_GMTOFF +#include +#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; + +typedef unsigned char byte; + + +typedef struct { + int istty2; +} state_t; + +#if PY_MAJOR_VERSION < 3 +static state_t state; +# define get_state(x) (&state) +# define cstr_argf "s" +#else +# define get_state(x) ((state_t *) PyModule_GetState(x)) +# define cstr_argf "y" +#endif // PY_MAJOR_VERSION >= 3 + + +#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)) + + + +#if PY_MAJOR_VERSION < 3 +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; +} +#endif + + +static int bup_ulong_from_py(unsigned long *x, PyObject *py, const char *name) +{ +#if PY_MAJOR_VERSION < 3 + if (PyInt_Check(py)) + return bup_ulong_from_pyint(x, py, name); +#endif + + 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 PY_MAJOR_VERSION < 3 + if (PyInt_Check(py)) + { + unsigned long tmp; + if (bup_ulong_from_pyint(&tmp, py, name)) + { + *x = tmp; + return 1; + } + return 0; + } +#endif + + 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; +} + + +static PyObject *bup_bytescmp(PyObject *self, PyObject *args) +{ + PyObject *py_s1, *py_s2; // This is really a PyBytes/PyString + if (!PyArg_ParseTuple(args, "SS", &py_s1, &py_s2)) + return NULL; + char *s1, *s2; + Py_ssize_t s1_len, s2_len; + if (PyBytes_AsStringAndSize(py_s1, &s1, &s1_len) == -1) + return NULL; + if (PyBytes_AsStringAndSize(py_s2, &s2, &s2_len) == -1) + return NULL; + const Py_ssize_t n = (s1_len < s2_len) ? s1_len : s2_len; + const int cmp = memcmp(s1, s2, n); + if (cmp != 0) + return PyLong_FromLong(cmp); + if (s1_len == s2_len) + return PyLong_FromLong(0);; + return PyLong_FromLong((s1_len < s2_len) ? -1 : 1); +} + // Probably we should use autoconf or something and set HAVE_PY_GETARGCARGV... #if __WIN32__ || __CYGWIN__ @@ -97,6 +283,233 @@ static void unpythonize_argv(void) #endif // not __WIN32__ or __CYGWIN__ +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 *append_sparse_region(const int fd, unsigned long long n) +{ + while (n) + { + off_t new_off; + if (!INTEGRAL_ASSIGNMENT_FITS(&new_off, n)) + new_off = INT_MAX; + const off_t off = lseek(fd, new_off, SEEK_CUR); + if (off == (off_t) -1) + return PyErr_SetFromErrno(PyExc_IOError); + n -= new_off; + } + return NULL; +} + + +static PyObject *record_sparse_zeros(unsigned long long *new_pending, + const int fd, + unsigned long long prev_pending, + const unsigned long long count) +{ + // Add count additional sparse zeros to prev_pending and store the + // result in new_pending, or if the total won't fit in + // new_pending, write some of the zeros to fd sparsely, and store + // the remaining sum in new_pending. + if (!uadd(new_pending, prev_pending, count)) + { + PyObject *err = append_sparse_region(fd, prev_pending); + if (err != NULL) + return err; + *new_pending = count; + } + return NULL; +} + + +static byte* find_not_zero(const byte * const start, const byte * const end) +{ + // Return a pointer to first non-zero byte between start and end, + // or end if there isn't one. + assert(start <= end); + const unsigned char *cur = start; + while (cur < end && *cur == 0) + cur++; + return (byte *) cur; +} + + +static byte* find_trailing_zeros(const byte * const start, + const byte * const end) +{ + // Return a pointer to the start of any trailing run of zeros, or + // end if there isn't one. + assert(start <= end); + if (start == end) + return (byte *) end; + const byte * cur = end; + while (cur > start && *--cur == 0) {} + if (*cur == 0) + return (byte *) cur; + else + return (byte *) (cur + 1); +} + + +static byte *find_non_sparse_end(const byte * const start, + const byte * const end, + const unsigned long long min_len) +{ + // Return the first pointer to a min_len sparse block in [start, + // end) if there is one, otherwise a pointer to the start of any + // trailing run of zeros. If there are no trailing zeros, return + // end. + if (start == end) + return (byte *) end; + assert(start < end); + assert(min_len); + // Probe in min_len jumps, searching backward from the jump + // destination for a non-zero byte. If such a byte is found, move + // just past it and try again. + const byte *candidate = start; + // End of any run of zeros, starting at candidate, that we've already seen + const byte *end_of_known_zeros = candidate; + while (end - candidate >= min_len) // Handle all min_len candidate blocks + { + const byte * const probe_end = candidate + min_len; + const byte * const trailing_zeros = + find_trailing_zeros(end_of_known_zeros, probe_end); + if (trailing_zeros == probe_end) + end_of_known_zeros = candidate = probe_end; + else if (trailing_zeros == end_of_known_zeros) + { + assert(candidate >= start); + assert(candidate <= end); + assert(*candidate == 0); + return (byte *) candidate; + } + else + { + candidate = trailing_zeros; + end_of_known_zeros = probe_end; + } + } + + if (candidate == end) + return (byte *) end; + + // No min_len sparse run found, search backward from end + const byte * const trailing_zeros = find_trailing_zeros(end_of_known_zeros, + end); + + if (trailing_zeros == end_of_known_zeros) + { + assert(candidate >= start); + assert(candidate < end); + assert(*candidate == 0); + assert(end - candidate < min_len); + return (byte *) candidate; + } + + if (trailing_zeros == end) + { + assert(*(end - 1) != 0); + return (byte *) end; + } + + assert(end - trailing_zeros < min_len); + assert(trailing_zeros >= start); + assert(trailing_zeros < end); + assert(*trailing_zeros == 0); + return (byte *) trailing_zeros; +} + + +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"); + + const byte * block = buf; // Start of pending block + const byte * const end = buf + buf_len; + unsigned long long zeros = prev_sparse_len; + while (1) + { + assert(block <= end); + if (block == end) + return PyLong_FromUnsignedLongLong(zeros); + + if (*block != 0) + { + // Look for the end of block, i.e. the next sparse run of + // at least min_sparse_len zeros, or the end of the + // buffer. + const byte * const probe = find_non_sparse_end(block + 1, end, + min_sparse_len); + // Either at end of block, or end of non-sparse; write pending data + PyObject *err = append_sparse_region(fd, zeros); + if (err != NULL) + return err; + int rc = write_all(fd, block, probe - block); + if (rc) + return PyErr_SetFromErrno(PyExc_IOError); + + if (end - probe < min_sparse_len) + zeros = end - probe; + else + zeros = min_sparse_len; + block = probe + zeros; + } + else // *block == 0 + { + // Should be in the first loop iteration, a sparse run of + // zeros, or nearly at the end of the block (within + // min_sparse_len). + const byte * const zeros_end = find_not_zero(block, end); + PyObject *err = record_sparse_zeros(&zeros, fd, + zeros, zeros_end - block); + if (err != NULL) + return err; + assert(block <= zeros_end); + block = zeros_end; + } + } +} + + static PyObject *selftest(PyObject *self, PyObject *args) { if (!PyArg_ParseTuple(args, "")) @@ -332,6 +745,11 @@ struct sha { unsigned char bytes[20]; }; +static inline int _cmp_sha(const struct sha *sha1, const struct sha *sha2) +{ + return memcmp(sha1->bytes, sha2->bytes, sizeof(sha1->bytes)); +} + struct idx { unsigned char *map; @@ -342,17 +760,6 @@ struct idx { int name_base; }; - -static int _cmp_sha(const struct sha *sha1, const struct sha *sha2) -{ - int i; - for (i = 0; i < sizeof(struct sha); i++) - if (sha1->bytes[i] != sha2->bytes[i]) - return sha1->bytes[i] - sha2->bytes[i]; - return 0; -} - - static void _fix_idx_order(struct idx **idxs, int *last_i) { struct idx *idx; @@ -403,7 +810,7 @@ static uint32_t _get_idx_i(struct idx *idx) 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; @@ -415,9 +822,13 @@ static PyObject *merge_into(PyObject *self, PyObject *args) 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 *)); @@ -447,7 +858,7 @@ static PyObject *merge_into(PyObject *self, PyObject *args) { struct idx *idx; uint32_t new_prefix; - if (count % 102424 == 0 && istty2) + if (count % 102424 == 0 && get_state(self)->istty2) fprintf(stderr, "midx: writing %.2f%% (%d/%d)\r", count*100.0/total, count, total); idx = idxs[last_i]; @@ -473,24 +884,12 @@ static PyObject *merge_into(PyObject *self, PyObject *args) 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 FAN_ENTRIES 256 static PyObject *write_idx(PyObject *self, PyObject *args) { char *filename = NULL; - PyObject *idx = NULL; + PyObject *py_total, *idx = NULL; PyObject *part; unsigned char *fmap = NULL; Py_ssize_t flen = 0; @@ -501,9 +900,13 @@ static PyObject *write_idx(PyObject *self, PyObject *args) uint64_t *ofs64_ptr; struct sha *sha_ptr; - if (!PyArg_ParseTuple(args, "sw#OI", &filename, &fmap, &flen, &idx, &total)) + if (!PyArg_ParseTuple(args, "sw#OO", + &filename, &fmap, &flen, &idx, &py_total)) return NULL; + 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); @@ -531,15 +934,20 @@ static PyObject *write_idx(PyObject *self, PyObject *args) { unsigned char *sha = NULL; Py_ssize_t sha_len = 0; - unsigned int crc = 0; - unsigned PY_LONG_LONG ofs_py = 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#IK", - &sha, &sha_len, &crc, &ofs_py)) + 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_py <= UINT64_MAX); - ofs = ofs_py; + assert(ofs_ull <= UINT64_MAX); + ofs = ofs_ull; if (sha_len != sizeof(struct sha)) return NULL; memcpy(sha_ptr++, sha, sizeof(struct sha)); @@ -679,21 +1087,38 @@ static PyObject *open_noatime(PyObject *self, PyObject *args) static PyObject *fadvise_done(PyObject *self, PyObject *args) { int fd = -1; - long long ofs = 0; - if (!PyArg_ParseTuple(args, "iL", &fd, &ofs)) + long long llofs, lllen = 0; + if (!PyArg_ParseTuple(args, "iLL", &fd, &llofs, &lllen)) return NULL; + off_t ofs, len; + if (!INTEGRAL_ASSIGNMENT_FITS(&ofs, llofs)) + return PyErr_Format(PyExc_OverflowError, + "fadvise offset overflows off_t"); + if (!INTEGRAL_ASSIGNMENT_FITS(&len, lllen)) + return PyErr_Format(PyExc_OverflowError, + "fadvise length overflows off_t"); #ifdef POSIX_FADV_DONTNEED - posix_fadvise(fd, 0, ofs, POSIX_FADV_DONTNEED); + posix_fadvise(fd, ofs, len, POSIX_FADV_DONTNEED); #endif return Py_BuildValue(""); } +// 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; - unsigned int attr; + unsigned long attr; char *path; int fd; @@ -704,29 +1129,35 @@ static PyObject *bup_get_linux_file_attr(PyObject *self, PyObject *args) 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("I", 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 int orig_attr, attr; + unsigned long orig_attr; + unsigned int attr; char *path; + PyObject *py_attr; int fd; - if (!PyArg_ParseTuple(args, "sI", &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); @@ -742,13 +1173,15 @@ static PyObject *bup_set_linux_file_attr(PyObject *self, PyObject *args) | 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); if (rc == -1) { close(fd); return PyErr_SetFromErrnoWithFilename(PyExc_OSError, path); } - attr |= (orig_attr & FS_EXTENT_FL); + assert(orig_attr <= UINT_MAX); // Kernel type is actually int + attr |= ((unsigned int) orig_attr) & FS_EXTENT_FL; rc = ioctl(fd, FS_IOC_SETFLAGS, &attr); if (rc == -1) @@ -772,162 +1205,151 @@ static PyObject *bup_set_linux_file_attr(PyObject *self, PyObject *args) #endif #endif +#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; \ + }) + -#if defined(HAVE_UTIMENSAT) || defined(HAVE_FUTIMES) || defined(HAVE_LUTIMES) +#ifdef HAVE_UTIMENSAT -static int bup_parse_xutime_args(char **path, - long *access, - long *access_ns, - long *modification, - long *modification_ns, - PyObject *self, PyObject *args) +static PyObject *bup_utimensat(PyObject *self, PyObject *args) { - if (!PyArg_ParseTuple(args, "s((ll)(ll))", - path, - access, access_ns, - modification, modification_ns)) - return 0; + int rc; + int fd, flag; + char *path; + PyObject *access_py, *modification_py; + struct timespec ts[2]; - if (isnan(*access)) - { - PyErr_SetString(PyExc_ValueError, "access time is NaN"); - return 0; - } - else if (isinf(*access)) + if (!PyArg_ParseTuple(args, "i" cstr_argf "((Ol)(Ol))i", + &fd, + &path, + &access_py, &(ts[0].tv_nsec), + &modification_py, &(ts[1].tv_nsec), + &flag)) + return NULL; + + int overflow; + if (!ASSIGN_PYLONG_TO_INTEGRAL(&(ts[0].tv_sec), access_py, &overflow)) { - PyErr_SetString(PyExc_ValueError, "access time is infinite"); - return 0; + if (overflow) + PyErr_SetString(PyExc_ValueError, + "unable to convert access time seconds for utimensat"); + return NULL; } - else if (isnan(*modification)) + if (!ASSIGN_PYLONG_TO_INTEGRAL(&(ts[1].tv_sec), modification_py, &overflow)) { - PyErr_SetString(PyExc_ValueError, "modification time is NaN"); - return 0; + if (overflow) + PyErr_SetString(PyExc_ValueError, + "unable to convert modification time seconds for utimensat"); + return NULL; } - else if (isinf(*modification)) - { - PyErr_SetString(PyExc_ValueError, "modification time is infinite"); + 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 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, cstr_argf "((OL)(OL))", + path, + &access_py, &access_us, + &modification_py, &modification_us)) return 0; - } - if (isnan(*access_ns)) + int overflow; + if (!ASSIGN_PYLONG_TO_INTEGRAL(&(tv[0].tv_sec), access_py, &overflow)) { - PyErr_SetString(PyExc_ValueError, "access time ns is NaN"); + 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"); + 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"); + 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"); + PyErr_SetString(PyExc_ValueError, "unable to convert modification time nanoseconds to timeval"); return 0; } - return 1; } -#endif /* defined(HAVE_UTIMENSAT) || defined(HAVE_FUTIMES) - || defined(HAVE_LUTIMES) */ - - -#ifdef HAVE_UTIMENSAT - -static PyObject *bup_xutime_ns(PyObject *self, PyObject *args, - int follow_symlinks) -{ - int rc; - char *path; - long access, access_ns, modification, modification_ns; - struct timespec ts[2]; - - if (!bup_parse_xutime_args(&path, &access, &access_ns, - &modification, &modification_ns, - self, args)) - return NULL; - - ts[0].tv_sec = access; - ts[0].tv_nsec = access_ns; - ts[1].tv_sec = modification; - ts[1].tv_nsec = modification_ns; - rc = utimensat(AT_FDCWD, path, ts, - follow_symlinks ? 0 : AT_SYMLINK_NOFOLLOW); - if (rc != 0) - return PyErr_SetFromErrnoWithFilename(PyExc_OSError, path); - - return Py_BuildValue("O", Py_None); -} - - -#define BUP_HAVE_BUP_UTIME_NS 1 -static PyObject *bup_utime_ns(PyObject *self, PyObject *args) -{ - return bup_xutime_ns(self, args, 1); -} - - -#define BUP_HAVE_BUP_LUTIME_NS 1 -static PyObject *bup_lutime_ns(PyObject *self, PyObject *args) -{ - return bup_xutime_ns(self, args, 0); -} - - -#else /* not defined(HAVE_UTIMENSAT) */ +#endif /* defined(HAVE_UTIMES) || defined(HAVE_LUTIMES) */ #ifdef HAVE_UTIMES -#define BUP_HAVE_BUP_UTIME_NS 1 -static PyObject *bup_utime_ns(PyObject *self, PyObject *args) +static PyObject *bup_utimes(PyObject *self, PyObject *args) { - int rc; char *path; - long access, access_ns, modification, modification_ns; struct timeval tv[2]; - - if (!bup_parse_xutime_args(&path, &access, &access_ns, - &modification, &modification_ns, - self, args)) - return NULL; - - tv[0].tv_sec = access; - tv[0].tv_usec = access_ns / 1000; - tv[1].tv_sec = modification; - tv[1].tv_usec = modification_ns / 1000; - rc = utimes(path, tv); + 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 */ #ifdef HAVE_LUTIMES -#define BUP_HAVE_BUP_LUTIME_NS 1 -static PyObject *bup_lutime_ns(PyObject *self, PyObject *args) +static PyObject *bup_lutimes(PyObject *self, PyObject *args) { - int rc; char *path; - long access, access_ns, modification, modification_ns; struct timeval tv[2]; - - if (!bup_parse_xutime_args(&path, &access, &access_ns, - &modification, &modification_ns, - self, args)) - return NULL; - - tv[0].tv_sec = access; - tv[0].tv_usec = access_ns / 1000; - tv[1].tv_sec = modification; - tv[1].tv_usec = modification_ns / 1000; - rc = lutimes(path, tv); + if (!bup_parse_xutimes_args(&path, tv, args)) + return NULL; + int rc = lutimes(path, tv); if (rc != 0) return PyErr_SetFromErrnoWithFilename(PyExc_OSError, path); @@ -936,9 +1358,6 @@ static PyObject *bup_lutime_ns(PyObject *self, PyObject *args) #endif /* def HAVE_LUTIMES */ -#endif /* not defined(HAVE_UTIMENSAT) */ - - #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 @@ -954,74 +1373,18 @@ static PyObject *bup_lutime_ns(PyObject *self, PyObject *args) #endif -static void set_invalid_timespec_msg(const char *field, - const long long sec, - const long nsec, - const char *filename, - int fd) -{ - if (filename != NULL) - PyErr_Format(PyExc_ValueError, - "invalid %s timespec (%lld %ld) for file \"%s\"", - field, sec, nsec, filename); - else - PyErr_Format(PyExc_ValueError, - "invalid %s timespec (%lld %ld) for file descriptor %d", - field, sec, nsec, fd); -} - - -static int normalize_timespec_values(const char *name, - long long *sec, - long *nsec, - const char *filename, - int fd) -{ - if (*nsec < -999999999 || *nsec > 999999999) - { - set_invalid_timespec_msg(name, *sec, *nsec, filename, fd); - return 0; - } - if (*nsec < 0) - { - if (*sec == LONG_MIN) - { - set_invalid_timespec_msg(name, *sec, *nsec, filename, fd); - return 0; - } - *nsec += 1000000000; - *sec -= 1; - } - return 1; -} - - -#define INTEGER_TO_PY(x) \ - (((x) >= 0) ? PyLong_FromUnsignedLongLong(x) : PyLong_FromLongLong(x)) - +#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) { - long long atime = st->st_atime; - long long mtime = st->st_mtime; - long long ctime = st->st_ctime; - long atime_ns = BUP_STAT_ATIME_NS(st); - long mtime_ns = BUP_STAT_MTIME_NS(st); - long ctime_ns = BUP_STAT_CTIME_NS(st); - - if (!normalize_timespec_values("atime", &atime, &atime_ns, filename, fd)) - return NULL; - if (!normalize_timespec_values("mtime", &mtime, &mtime_ns, filename, fd)) - return NULL; - if (!normalize_timespec_values("ctime", &ctime, &ctime_ns, filename, fd)) - return NULL; - // 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(). - return Py_BuildValue("OKOOOOOL(Ll)(Ll)(Ll)", + // 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), @@ -1030,21 +1393,22 @@ static PyObject *stat_struct_to_py(const struct stat *st, INTEGER_TO_PY(st->st_gid), INTEGER_TO_PY(st->st_rdev), (PY_LONG_LONG) st->st_size, - (PY_LONG_LONG) atime, - (long) atime_ns, - (PY_LONG_LONG) mtime, - (long) mtime_ns, - (PY_LONG_LONG) ctime, - (long) ctime_ns); + 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)); } +#pragma clang diagnostic pop // ignored "-Wtautological-compare" static PyObject *bup_stat(PyObject *self, PyObject *args) { int rc; char *filename; - if (!PyArg_ParseTuple(args, "s", &filename)) + if (!PyArg_ParseTuple(args, cstr_argf, &filename)) return NULL; struct stat st; @@ -1060,7 +1424,7 @@ static PyObject *bup_lstat(PyObject *self, PyObject *args) int rc; char *filename; - if (!PyArg_ParseTuple(args, "s", &filename)) + if (!PyArg_ParseTuple(args, cstr_argf, &filename)) return NULL; struct stat st; @@ -1086,7 +1450,94 @@ static PyObject *bup_fstat(PyObject *self, PyObject *args) } +#ifdef HAVE_TM_TM_GMTOFF +static PyObject *bup_localtime(PyObject *self, PyObject *args) +{ + long long lltime; + time_t ttime; + if (!PyArg_ParseTuple(args, "L", &lltime)) + return NULL; + if (!INTEGRAL_ASSIGNMENT_FITS(&ttime, lltime)) + return PyErr_Format(PyExc_OverflowError, "time value too large"); + + struct tm tm; + tzset(); + if(localtime_r(&ttime, &tm) == NULL) + return PyErr_SetFromErrno(PyExc_OSError); + + // Match the Python struct_time values. + return Py_BuildValue("[i,i,i,i,i,i,i,i,i,i,s]", + 1900 + tm.tm_year, tm.tm_mon + 1, tm.tm_mday, + tm.tm_hour, tm.tm_min, tm.tm_sec, + tm.tm_wday, tm.tm_yday + 1, + tm.tm_isdst, tm.tm_gmtoff, tm.tm_zone); +} +#endif /* def HAVE_TM_TM_GMTOFF */ + + +#ifdef BUP_MINCORE_BUF_TYPE +static PyObject *bup_mincore(PyObject *self, PyObject *args) +{ + Py_buffer src, dest; + PyObject *py_src_n, *py_src_off, *py_dest_off; + + if (!PyArg_ParseTuple(args, cstr_argf "*OOw*O", + &src, &py_src_n, &py_src_off, + &dest, &py_dest_off)) + return NULL; + + PyObject *result = NULL; + + unsigned long long src_n, src_off, dest_off; + if (!(bup_ullong_from_py(&src_n, py_src_n, "src_n") + && bup_ullong_from_py(&src_off, py_src_off, "src_off") + && bup_ullong_from_py(&dest_off, py_dest_off, "dest_off"))) + goto clean_and_return; + + unsigned long long src_region_end; + if (!uadd(&src_region_end, src_off, src_n)) { + result = PyErr_Format(PyExc_OverflowError, "(src_off + src_n) too large"); + goto clean_and_return; + } + if (src_region_end > src.len) { + result = PyErr_Format(PyExc_OverflowError, "region runs off end of src"); + goto clean_and_return; + } + + unsigned long long dest_size; + if (!INTEGRAL_ASSIGNMENT_FITS(&dest_size, dest.len)) { + result = PyErr_Format(PyExc_OverflowError, "invalid dest size"); + goto clean_and_return; + } + if (dest_off > dest_size) { + result = PyErr_Format(PyExc_OverflowError, "region runs off end of dest"); + goto clean_and_return; + } + + size_t length; + if (!INTEGRAL_ASSIGNMENT_FITS(&length, src_n)) { + result = PyErr_Format(PyExc_OverflowError, "src_n overflows size_t"); + goto clean_and_return; + } + int rc = mincore((void *)(src.buf + src_off), src_n, + (BUP_MINCORE_BUF_TYPE *) (dest.buf + dest_off)); + if (rc != 0) { + result = PyErr_SetFromErrno(PyExc_OSError); + goto clean_and_return; + } + result = Py_BuildValue("O", Py_None); + + clean_and_return: + PyBuffer_Release(&src); + PyBuffer_Release(&dest); + return result; +} +#endif /* def BUP_MINCORE_BUF_TYPE */ + + 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, @@ -1123,13 +1574,17 @@ static PyMethodDef helper_methods[] = { { "set_linux_file_attr", bup_set_linux_file_attr, METH_VARARGS, "Set the Linux attributes for the given file." }, #endif -#ifdef BUP_HAVE_BUP_UTIME_NS - { "bup_utime_ns", bup_utime_ns, METH_VARARGS, - "Change path timestamps with up to 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 BUP_HAVE_BUP_LUTIME_NS - { "bup_lutime_ns", bup_lutime_ns, METH_VARARGS, - "Change path timestamps with up to nanosecond precision;" +#ifdef HAVE_LUTIMES + { "bup_lutimes", bup_lutimes, METH_VARARGS, + "Change path timestamps with microsecond precision;" " don't follow symlinks." }, #endif { "stat", bup_stat, METH_VARARGS, @@ -1138,27 +1593,127 @@ static PyMethodDef helper_methods[] = { "Extended version of lstat." }, { "fstat", bup_fstat, METH_VARARGS, "Extended version of fstat." }, +#ifdef HAVE_TM_TM_GMTOFF + { "localtime", bup_localtime, METH_VARARGS, + "Return struct_time elements plus the timezone offset and name." }, +#endif + { "bytescmp", bup_bytescmp, METH_VARARGS, + "Return a negative value if x < y, zero if equal, positive otherwise."}, +#ifdef BUP_MINCORE_BUF_TYPE + { "mincore", bup_mincore, METH_VARARGS, + "For mincore(src, src_n, src_off, dest, dest_off)" + " call the system mincore(src + src_off, src_n, &dest[dest_off])." }, +#endif { NULL, NULL, 0, NULL }, // sentinel }; - -PyMODINIT_FUNC init_helpers(void) +static int setup_module(PyObject *m) { - // 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. + // FIXME: migrate these tests to configure, or at least don't + // possibly crash the whole application. 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)); + + // Originally required by append_sparse_region() + { + off_t probe; + if (!INTEGRAL_ASSIGNMENT_FITS(&probe, INT_MAX)) + { + fprintf(stderr, "off_t can't hold INT_MAX; please report.\n"); + exit(1); + } + } char *e; +#pragma clang diagnostic push +#pragma clang diagnostic ignored "-Wtautological-compare" // For INTEGER_TO_PY(). + { + PyObject *value; + value = INTEGER_TO_PY(INT_MAX); + PyObject_SetAttrString(m, "INT_MAX", value); + Py_DECREF(value); + value = INTEGER_TO_PY(UINT_MAX); + PyObject_SetAttrString(m, "UINT_MAX", value); + Py_DECREF(value); + } +#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 +#ifdef BUP_HAVE_MINCORE_INCORE + { + PyObject *value; + value = INTEGER_TO_PY(MINCORE_INCORE); + PyObject_SetAttrString(m, "MINCORE_INCORE", value); + Py_DECREF(value); + } +#endif +#pragma clang diagnostic pop // ignored "-Wtautological-compare" + + e = getenv("BUP_FORCE_TTY"); + get_state(m)->istty2 = isatty(2) || (atoi(e ? e : "0") & 2); + unpythonize_argv(); + return 1; +} + + +#if PY_MAJOR_VERSION < 3 + +PyMODINIT_FUNC init_helpers(void) +{ PyObject *m = Py_InitModule("_helpers", helper_methods); if (m == NULL) return; - e = getenv("BUP_FORCE_TTY"); - istty2 = isatty(2) || (atoi(e ? e : "0") & 2); - unpythonize_argv(); + + if (!setup_module(m)) + { + Py_DECREF(m); + return; + } } + +# else // PY_MAJOR_VERSION >= 3 + +static struct PyModuleDef helpers_def = { + PyModuleDef_HEAD_INIT, + "_helpers", + NULL, + sizeof(state_t), + helper_methods, + NULL, + NULL, // helpers_traverse, + NULL, // helpers_clear, + NULL +}; + +PyMODINIT_FUNC PyInit__helpers(void) +{ + PyObject *module = PyModule_Create(&helpers_def); + if (module == NULL) + return NULL; + if (!setup_module(module)) + { + Py_DECREF(module); + return NULL; + } + return module; +} + +#endif // PY_MAJOR_VERSION >= 3