]> arthur.barton.de Git - bup.git/blobdiff - lib/bup/_helpers.c
index: raise children_n limit to struct 'I' max
[bup.git] / lib / bup / _helpers.c
index 65a8b6bafa8ef51efeab07c5d8cf7febb7b0787e..808c728642ae2d0669ecebdb256ec85a294f7592 100644 (file)
 #include <stdint.h>
 #include <stdlib.h>
 #include <stdio.h>
-#include <sys/mman.h>
 
+#ifdef HAVE_SYS_MMAN_H
+#include <sys/mman.h>
+#endif
 #ifdef HAVE_SYS_TYPES_H
 #include <sys/types.h>
 #endif
 #include <sys/ioctl.h>
 #endif
 
+#ifdef HAVE_TM_TM_GMTOFF
+#include <time.h>
+#endif
+
 #include "bupsplit.h"
 
 #if defined(FS_IOC_GETFLAGS) && defined(FS_IOC_SETFLAGS)
@@ -70,6 +76,13 @@ static uint64_t htonll(uint64_t value)
 #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.
 
@@ -222,6 +235,135 @@ static void unpythonize_argv(void)
 #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 *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 *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");
+
+    // The value of zeros_read indicates the number of zeros read from
+    // buf that haven't been accounted for yet (with respect to cur),
+    // while zeros indicates the total number of pending zeros, which
+    // could be larger in the first iteration if prev_sparse_len
+    // wasn't zero.
+    int rc;
+    unsigned long long unexamined = buf_len;
+    unsigned char *block_start = buf, *cur = buf;
+    unsigned long long zeros, zeros_read = count_leading_zeros(cur, unexamined);
+    assert(zeros_read <= unexamined);
+    unexamined -= zeros_read;
+    if (!uadd(&zeros, prev_sparse_len, zeros_read))
+    {
+        PyObject *err = append_sparse_region(fd, prev_sparse_len);
+        if (err != NULL)
+            return err;
+        zeros = zeros_read;
+    }
+
+    while(unexamined)
+    {
+        if (zeros < min_sparse_len)
+            cur += zeros_read;
+        else
+        {
+            rc = write_all(fd, block_start, cur - block_start);
+            if (rc)
+                return PyErr_SetFromErrno(PyExc_IOError);
+            PyObject *err = append_sparse_region(fd, zeros);
+            if (err != NULL)
+                return err;
+            cur += zeros_read;
+            block_start = cur;
+        }
+        // Pending zeros have ether been made sparse, or are going to
+        // be rolled into the next non-sparse block since we know we
+        // now have at least one unexamined non-zero byte.
+        assert(unexamined && *cur != 0);
+        zeros = zeros_read = 0;
+        while (unexamined && *cur != 0)
+        {
+            cur++; unexamined--;
+        }
+        if (unexamined)
+        {
+            zeros_read = count_leading_zeros(cur, unexamined);
+            assert(zeros_read <= unexamined);
+            unexamined -= zeros_read;
+            zeros = zeros_read;
+        }
+    }
+    rc = write_all(fd, block_start, cur - block_start);
+    if (rc)
+        return PyErr_SetFromErrno(PyExc_IOError);
+    return PyLong_FromUnsignedLongLong(zeros);
+}
+
+
 static PyObject *selftest(PyObject *self, PyObject *args)
 {
     if (!PyArg_ParseTuple(args, ""))
@@ -805,11 +947,18 @@ 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("");
 }
@@ -842,14 +991,13 @@ static PyObject *bup_get_linux_file_attr(PyObject *self, PyObject *args)
 
     attr = 0;  // Handle int/long mismatch (see above)
     rc = ioctl(fd, FS_IOC_GETFLAGS, &attr);
-    assert(attr <= UINT_MAX);  // Kernel type is actually int
     if (rc == -1)
     {
         close(fd);
         return PyErr_SetFromErrnoWithFilename(PyExc_OSError, path);
     }
-
     close(fd);
+    assert(attr <= UINT_MAX);  // Kernel type is actually int
     return PyLong_FromUnsignedLong(attr);
 }
 #endif /* def BUP_HAVE_FILE_ATTRS */
@@ -917,14 +1065,6 @@ static PyObject *bup_set_linux_file_attr(PyObject *self, PyObject *args)
 #endif
 #endif
 
-
-#define INTEGRAL_ASSIGNMENT_FITS(dest, src)                             \
-    ({                                                                  \
-        *(dest) = (src);                                                \
-        *(dest) == (src) && (*(dest) < 1) == ((src) < 1);               \
-    })
-
-
 #define ASSIGN_PYLONG_TO_INTEGRAL(dest, pylong, overflow) \
     ({                                                     \
         int result = 0;                                                 \
@@ -1170,7 +1310,78 @@ 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)
+{
+    const char *src;
+    Py_ssize_t src_ssize;
+    Py_buffer dest;
+    PyObject *py_src_n, *py_src_off, *py_dest_off;
+    if (!PyArg_ParseTuple(args, "s#OOw*O",
+                          &src, &src_ssize, &py_src_n, &py_src_off,
+                          &dest, &py_dest_off))
+       return NULL;
+
+    unsigned long long src_size, src_n, src_off, dest_size, 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")))
+        return NULL;
+
+    if (!INTEGRAL_ASSIGNMENT_FITS(&src_size, src_ssize))
+        return PyErr_Format(PyExc_OverflowError, "invalid src size");
+    unsigned long long src_region_end;
+
+    if (!uadd(&src_region_end, src_off, src_n))
+        return PyErr_Format(PyExc_OverflowError, "(src_off + src_n) too large");
+    if (src_region_end > src_size)
+        return PyErr_Format(PyExc_OverflowError, "region runs off end of src");
+
+    if (!INTEGRAL_ASSIGNMENT_FITS(&dest_size, dest.len))
+        return PyErr_Format(PyExc_OverflowError, "invalid dest size");
+    if (dest_off > dest_size)
+        return PyErr_Format(PyExc_OverflowError, "region runs off end of dest");
+
+    size_t length;
+    if (!INTEGRAL_ASSIGNMENT_FITS(&length, src_n))
+        return PyErr_Format(PyExc_OverflowError, "src_n overflows size_t");
+    int rc = mincore((void *)(src + src_off), src_n,
+                     (BUP_MINCORE_BUF_TYPE *) (dest.buf + dest_off));
+    if (rc != 0)
+        return PyErr_SetFromErrno(PyExc_OSError);
+    return Py_BuildValue("O", Py_None);
+}
+#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,
@@ -1226,6 +1437,15 @@ 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
+#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
 };
 
@@ -1243,6 +1463,13 @@ PyMODINIT_FUNC init_helpers(void)
     assert(sizeof(PY_LONG_LONG) <= sizeof(long long));
     assert(sizeof(unsigned PY_LONG_LONG) <= sizeof(unsigned long long));
 
+    if (sizeof(off_t) < sizeof(int))
+    {
+        // Originally required by append_sparse_region().
+        fprintf(stderr, "sizeof(off_t) < sizeof(int); please report.\n");
+        exit(1);
+    }
+
     char *e;
     PyObject *m = Py_InitModule("_helpers", helper_methods);
     if (m == NULL)
@@ -1250,6 +1477,15 @@ PyMODINIT_FUNC init_helpers(void)
 
 #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;
@@ -1264,18 +1500,14 @@ PyMODINIT_FUNC init_helpers(void)
         Py_DECREF(value);
     }
 #endif
+#ifdef BUP_HAVE_MINCORE_INCORE
     {
         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);
+        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");