sapling/mercurial/osutil.c

998 lines
23 KiB
C
Raw Normal View History

/*
osutil.c - native operating system services
Copyright 2007 Matt Mackall and others
This software may be used and distributed according to the terms of
the GNU General Public License, incorporated herein by reference.
*/
2007-10-15 23:57:01 +04:00
#define _ATFILE_SOURCE
#include <Python.h>
#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <errno.h>
#ifdef _WIN32
2010-01-25 09:05:27 +03:00
#include <windows.h>
#include <io.h>
#else
2010-01-25 09:05:27 +03:00
#include <dirent.h>
#include <sys/socket.h>
2010-01-25 09:05:27 +03:00
#include <sys/stat.h>
#include <sys/types.h>
#include <unistd.h>
#endif
osutil: use getdirentriesattr on OS X if possible This is a significant win for large repositories on OS X, especially with a cold cache. Unfortunately we need to keep the lstat-based implementation around for two reasons: - Not all filesystems support this call. - There's an edge case in which it's best to fall back to avoid a retry loop. More about this in the comments. The below tests are all performed on a Mac with an SSD running OS X 10.9, on a repository with over 200k files. The results are best of 5 with simulated best-effort conditions. The gains with a hot cache are pretty impressive: 'hg status' goes from 5.18 seconds to 3.79 seconds. However, a repository that large will probably already be using something like hgwatchman [1], which helps much more (for this repo, 'hg status' with hgwatchman is approximately 1 second). Where this really helps is when the cache is cold [2]: hg status goes from 31.0 seconds to 9.66. See http://lists.apple.com/archives/filesystem-dev/2014/Dec/msg00002.html for some more discussion about this function. This is based on a patch by Sean Farley <sean@farley.io>. [1] https://bitbucket.org/facebook/hgwatchman [2] There appears to be no easy way to clear the file cache (aka "vnodes") on OS X short of rebooting. purge(8) purportedly does that but in my testing had little effect. The workaround I came up with was to assume that vnode eviction was LRU, make sure the kern.maxvnodes sysctl is smaller than the size of the repository, then make sure we'd always miss the cache by running 'hg status' in another clone of the repository before running it in the test repository.
2015-03-26 01:55:31 +03:00
#ifdef __APPLE__
#include <sys/attr.h>
#include <sys/vnode.h>
#endif
#include "util.h"
2009-08-14 13:18:23 +04:00
/* some platforms lack the PATH_MAX definition (eg. GNU/Hurd) */
#ifndef PATH_MAX
#define PATH_MAX 4096
#endif
#ifdef _WIN32
/*
stat struct compatible with hg expectations
Mercurial only uses st_mode, st_size and st_mtime
the rest is kept to minimize changes between implementations
*/
struct hg_stat {
int st_dev;
int st_mode;
int st_nlink;
__int64 st_size;
int st_mtime;
int st_ctime;
};
struct listdir_stat {
PyObject_HEAD
struct hg_stat st;
};
#else
struct listdir_stat {
PyObject_HEAD
struct stat st;
};
#endif
#ifdef IS_PY3K
#define listdir_slot(name) \
static PyObject *listdir_stat_##name(PyObject *self, void *x) \
{ \
return PyLong_FromLong(((struct listdir_stat *)self)->st.name); \
}
#else
#define listdir_slot(name) \
2008-10-20 17:19:05 +04:00
static PyObject *listdir_stat_##name(PyObject *self, void *x) \
{ \
return PyInt_FromLong(((struct listdir_stat *)self)->st.name); \
}
#endif
2007-10-09 09:45:48 +04:00
listdir_slot(st_dev)
listdir_slot(st_mode)
listdir_slot(st_nlink)
#ifdef _WIN32
static PyObject *listdir_stat_st_size(PyObject *self, void *x)
{
return PyLong_FromLongLong(
(PY_LONG_LONG)((struct listdir_stat *)self)->st.st_size);
}
#else
2007-10-09 09:45:48 +04:00
listdir_slot(st_size)
#endif
2007-10-09 09:45:48 +04:00
listdir_slot(st_mtime)
listdir_slot(st_ctime)
static struct PyGetSetDef listdir_stat_getsets[] = {
{"st_dev", listdir_stat_st_dev, 0, 0, 0},
{"st_mode", listdir_stat_st_mode, 0, 0, 0},
{"st_nlink", listdir_stat_st_nlink, 0, 0, 0},
{"st_size", listdir_stat_st_size, 0, 0, 0},
{"st_mtime", listdir_stat_st_mtime, 0, 0, 0},
{"st_ctime", listdir_stat_st_ctime, 0, 0, 0},
{0, 0, 0, 0, 0}
};
static PyObject *listdir_stat_new(PyTypeObject *t, PyObject *a, PyObject *k)
{
return t->tp_alloc(t, 0);
}
static void listdir_stat_dealloc(PyObject *o)
{
o->ob_type->tp_free(o);
}
static PyTypeObject listdir_stat_type = {
PyVarObject_HEAD_INIT(NULL, 0)
"osutil.stat", /*tp_name*/
sizeof(struct listdir_stat), /*tp_basicsize*/
0, /*tp_itemsize*/
(destructor)listdir_stat_dealloc, /*tp_dealloc*/
0, /*tp_print*/
0, /*tp_getattr*/
0, /*tp_setattr*/
0, /*tp_compare*/
0, /*tp_repr*/
0, /*tp_as_number*/
0, /*tp_as_sequence*/
0, /*tp_as_mapping*/
0, /*tp_hash */
0, /*tp_call*/
0, /*tp_str*/
0, /*tp_getattro*/
0, /*tp_setattro*/
0, /*tp_as_buffer*/
Py_TPFLAGS_DEFAULT | Py_TPFLAGS_BASETYPE, /*tp_flags*/
"stat objects", /* tp_doc */
0, /* tp_traverse */
0, /* tp_clear */
0, /* tp_richcompare */
0, /* tp_weaklistoffset */
0, /* tp_iter */
0, /* tp_iternext */
0, /* tp_methods */
0, /* tp_members */
listdir_stat_getsets, /* tp_getset */
0, /* tp_base */
0, /* tp_dict */
0, /* tp_descr_get */
0, /* tp_descr_set */
0, /* tp_dictoffset */
0, /* tp_init */
0, /* tp_alloc */
listdir_stat_new, /* tp_new */
};
#ifdef _WIN32
static int to_python_time(const FILETIME *tm)
{
/* number of seconds between epoch and January 1 1601 */
const __int64 a0 = (__int64)134774L * (__int64)24L * (__int64)3600L;
/* conversion factor from 100ns to 1s */
const __int64 a1 = 10000000;
/* explicit (int) cast to suspend compiler warnings */
return (int)((((__int64)tm->dwHighDateTime << 32)
+ tm->dwLowDateTime) / a1 - a0);
}
static PyObject *make_item(const WIN32_FIND_DATAA *fd, int wantstat)
{
PyObject *py_st;
struct hg_stat *stp;
int kind = (fd->dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY)
? _S_IFDIR : _S_IFREG;
if (!wantstat)
return Py_BuildValue("si", fd->cFileName, kind);
py_st = PyObject_CallObject((PyObject *)&listdir_stat_type, NULL);
if (!py_st)
return NULL;
stp = &((struct listdir_stat *)py_st)->st;
/*
use kind as st_mode
rwx bits on Win32 are meaningless
and Hg does not use them anyway
*/
stp->st_mode = kind;
stp->st_mtime = to_python_time(&fd->ftLastWriteTime);
stp->st_ctime = to_python_time(&fd->ftCreationTime);
if (kind == _S_IFREG)
2010-01-25 09:05:27 +03:00
stp->st_size = ((__int64)fd->nFileSizeHigh << 32)
+ fd->nFileSizeLow;
return Py_BuildValue("siN", fd->cFileName,
kind, py_st);
}
static PyObject *_listdir(char *path, int plen, int wantstat, char *skip)
{
PyObject *rval = NULL; /* initialize - return value */
PyObject *list;
HANDLE fh;
WIN32_FIND_DATAA fd;
char *pattern;
/* build the path + \* pattern string */
2010-01-25 09:05:27 +03:00
pattern = malloc(plen + 3); /* path + \* + \0 */
if (!pattern) {
PyErr_NoMemory();
goto error_nomem;
}
memcpy(pattern, path, plen);
if (plen > 0) {
char c = path[plen-1];
if (c != ':' && c != '/' && c != '\\')
pattern[plen++] = '\\';
}
pattern[plen++] = '*';
pattern[plen] = '\0';
fh = FindFirstFileA(pattern, &fd);
if (fh == INVALID_HANDLE_VALUE) {
PyErr_SetFromWindowsErrWithFilename(GetLastError(), path);
goto error_file;
}
list = PyList_New(0);
if (!list)
goto error_list;
do {
PyObject *item;
if (fd.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY) {
if (!strcmp(fd.cFileName, ".")
|| !strcmp(fd.cFileName, ".."))
continue;
if (skip && !strcmp(fd.cFileName, skip)) {
rval = PyList_New(0);
goto error;
}
}
item = make_item(&fd, wantstat);
if (!item)
goto error;
if (PyList_Append(list, item)) {
Py_XDECREF(item);
goto error;
}
Py_XDECREF(item);
} while (FindNextFileA(fh, &fd));
if (GetLastError() != ERROR_NO_MORE_FILES) {
PyErr_SetFromWindowsErrWithFilename(GetLastError(), path);
goto error;
}
rval = list;
Py_XINCREF(rval);
error:
Py_XDECREF(list);
error_list:
FindClose(fh);
error_file:
free(pattern);
error_nomem:
return rval;
}
#else
2008-09-13 00:11:02 +04:00
int entkind(struct dirent *ent)
{
#ifdef DT_REG
2008-09-13 00:11:02 +04:00
switch (ent->d_type) {
case DT_REG: return S_IFREG;
case DT_DIR: return S_IFDIR;
case DT_LNK: return S_IFLNK;
case DT_BLK: return S_IFBLK;
case DT_CHR: return S_IFCHR;
case DT_FIFO: return S_IFIFO;
case DT_SOCK: return S_IFSOCK;
}
2007-10-15 23:57:01 +04:00
#endif
return -1;
}
static PyObject *makestat(const struct stat *st)
{
2012-12-01 05:40:11 +04:00
PyObject *stat;
2012-12-01 05:40:11 +04:00
stat = PyObject_CallObject((PyObject *)&listdir_stat_type, NULL);
if (stat)
memcpy(&((struct listdir_stat *)stat)->st, st, sizeof(*st));
return stat;
}
static PyObject *_listdir_stat(char *path, int pathlen, int keepstat,
char *skip)
{
PyObject *list, *elem, *stat = NULL, *ret = NULL;
char fullpath[PATH_MAX + 10];
int kind, err;
2008-09-13 00:11:02 +04:00
struct stat st;
struct dirent *ent;
DIR *dir;
#ifdef AT_SYMLINK_NOFOLLOW
int dfd = -1;
#endif
if (pathlen >= PATH_MAX) {
errno = ENAMETOOLONG;
PyErr_SetFromErrnoWithFilename(PyExc_OSError, path);
goto error_value;
}
2008-09-13 00:11:02 +04:00
strncpy(fullpath, path, PATH_MAX);
fullpath[pathlen] = '/';
#ifdef AT_SYMLINK_NOFOLLOW
dfd = open(path, O_RDONLY);
if (dfd == -1) {
PyErr_SetFromErrnoWithFilename(PyExc_OSError, path);
goto error_value;
}
dir = fdopendir(dfd);
#else
dir = opendir(path);
#endif
if (!dir) {
2008-09-13 00:11:02 +04:00
PyErr_SetFromErrnoWithFilename(PyExc_OSError, path);
goto error_dir;
2015-01-27 18:12:55 +03:00
}
list = PyList_New(0);
2008-09-13 00:11:02 +04:00
if (!list)
goto error_list;
2008-09-13 00:11:02 +04:00
while ((ent = readdir(dir))) {
if (!strcmp(ent->d_name, ".") || !strcmp(ent->d_name, ".."))
continue;
2008-09-13 00:11:02 +04:00
kind = entkind(ent);
if (kind == -1 || keepstat) {
#ifdef AT_SYMLINK_NOFOLLOW
err = fstatat(dfd, ent->d_name, &st,
AT_SYMLINK_NOFOLLOW);
#else
strncpy(fullpath + pathlen + 1, ent->d_name,
PATH_MAX - pathlen);
fullpath[PATH_MAX] = '\0';
err = lstat(fullpath, &st);
#endif
2008-09-13 00:11:02 +04:00
if (err == -1) {
/* race with file deletion? */
if (errno == ENOENT)
continue;
2008-09-13 00:11:02 +04:00
strncpy(fullpath + pathlen + 1, ent->d_name,
PATH_MAX - pathlen);
fullpath[PATH_MAX] = 0;
PyErr_SetFromErrnoWithFilename(PyExc_OSError,
fullpath);
goto error;
}
kind = st.st_mode & S_IFMT;
}
/* quit early? */
if (skip && kind == S_IFDIR && !strcmp(ent->d_name, skip)) {
ret = PyList_New(0);
goto error;
}
2008-09-13 00:11:02 +04:00
if (keepstat) {
stat = makestat(&st);
2008-09-13 00:11:02 +04:00
if (!stat)
goto error;
elem = Py_BuildValue("siN", ent->d_name, kind, stat);
} else
elem = Py_BuildValue("si", ent->d_name, kind);
if (!elem)
goto error;
stat = NULL;
2008-09-13 00:11:02 +04:00
PyList_Append(list, elem);
Py_DECREF(elem);
}
2008-09-13 00:11:02 +04:00
ret = list;
Py_INCREF(ret);
error:
Py_DECREF(list);
Py_XDECREF(stat);
2008-09-13 00:11:02 +04:00
error_list:
closedir(dir);
error_dir:
#ifdef AT_SYMLINK_NOFOLLOW
close(dfd);
#endif
error_value:
2008-09-13 00:11:02 +04:00
return ret;
}
osutil: use getdirentriesattr on OS X if possible This is a significant win for large repositories on OS X, especially with a cold cache. Unfortunately we need to keep the lstat-based implementation around for two reasons: - Not all filesystems support this call. - There's an edge case in which it's best to fall back to avoid a retry loop. More about this in the comments. The below tests are all performed on a Mac with an SSD running OS X 10.9, on a repository with over 200k files. The results are best of 5 with simulated best-effort conditions. The gains with a hot cache are pretty impressive: 'hg status' goes from 5.18 seconds to 3.79 seconds. However, a repository that large will probably already be using something like hgwatchman [1], which helps much more (for this repo, 'hg status' with hgwatchman is approximately 1 second). Where this really helps is when the cache is cold [2]: hg status goes from 31.0 seconds to 9.66. See http://lists.apple.com/archives/filesystem-dev/2014/Dec/msg00002.html for some more discussion about this function. This is based on a patch by Sean Farley <sean@farley.io>. [1] https://bitbucket.org/facebook/hgwatchman [2] There appears to be no easy way to clear the file cache (aka "vnodes") on OS X short of rebooting. purge(8) purportedly does that but in my testing had little effect. The workaround I came up with was to assume that vnode eviction was LRU, make sure the kern.maxvnodes sysctl is smaller than the size of the repository, then make sure we'd always miss the cache by running 'hg status' in another clone of the repository before running it in the test repository.
2015-03-26 01:55:31 +03:00
#ifdef __APPLE__
typedef struct {
u_int32_t length;
attrreference_t name;
fsobj_type_t obj_type;
struct timespec mtime;
#if __LITTLE_ENDIAN__
mode_t access_mask;
uint16_t padding;
#else
uint16_t padding;
mode_t access_mask;
#endif
off_t size;
} __attribute__((packed)) attrbuf_entry;
int attrkind(attrbuf_entry *entry)
{
switch (entry->obj_type) {
case VREG: return S_IFREG;
case VDIR: return S_IFDIR;
case VLNK: return S_IFLNK;
case VBLK: return S_IFBLK;
case VCHR: return S_IFCHR;
case VFIFO: return S_IFIFO;
case VSOCK: return S_IFSOCK;
}
return -1;
}
/* get these many entries at a time */
#define LISTDIR_BATCH_SIZE 50
static PyObject *_listdir_batch(char *path, int pathlen, int keepstat,
char *skip, bool *fallback)
{
PyObject *list, *elem, *stat = NULL, *ret = NULL;
int kind, err;
unsigned long index;
unsigned int count, old_state, new_state;
bool state_seen = false;
attrbuf_entry *entry;
/* from the getattrlist(2) man page: a path can be no longer than
(NAME_MAX * 3 + 1) bytes. Also, "The getattrlist() function will
silently truncate attribute data if attrBufSize is too small." So
pass in a buffer big enough for the worst case. */
char attrbuf[LISTDIR_BATCH_SIZE * (sizeof(attrbuf_entry) + NAME_MAX * 3 + 1)];
unsigned int basep_unused;
struct stat st;
int dfd = -1;
/* these must match the attrbuf_entry struct, otherwise you'll end up
with garbage */
struct attrlist requested_attr = {0};
requested_attr.bitmapcount = ATTR_BIT_MAP_COUNT;
requested_attr.commonattr = (ATTR_CMN_NAME | ATTR_CMN_OBJTYPE |
ATTR_CMN_MODTIME | ATTR_CMN_ACCESSMASK);
requested_attr.fileattr = ATTR_FILE_DATALENGTH;
osutil: use getdirentriesattr on OS X if possible This is a significant win for large repositories on OS X, especially with a cold cache. Unfortunately we need to keep the lstat-based implementation around for two reasons: - Not all filesystems support this call. - There's an edge case in which it's best to fall back to avoid a retry loop. More about this in the comments. The below tests are all performed on a Mac with an SSD running OS X 10.9, on a repository with over 200k files. The results are best of 5 with simulated best-effort conditions. The gains with a hot cache are pretty impressive: 'hg status' goes from 5.18 seconds to 3.79 seconds. However, a repository that large will probably already be using something like hgwatchman [1], which helps much more (for this repo, 'hg status' with hgwatchman is approximately 1 second). Where this really helps is when the cache is cold [2]: hg status goes from 31.0 seconds to 9.66. See http://lists.apple.com/archives/filesystem-dev/2014/Dec/msg00002.html for some more discussion about this function. This is based on a patch by Sean Farley <sean@farley.io>. [1] https://bitbucket.org/facebook/hgwatchman [2] There appears to be no easy way to clear the file cache (aka "vnodes") on OS X short of rebooting. purge(8) purportedly does that but in my testing had little effect. The workaround I came up with was to assume that vnode eviction was LRU, make sure the kern.maxvnodes sysctl is smaller than the size of the repository, then make sure we'd always miss the cache by running 'hg status' in another clone of the repository before running it in the test repository.
2015-03-26 01:55:31 +03:00
*fallback = false;
if (pathlen >= PATH_MAX) {
errno = ENAMETOOLONG;
PyErr_SetFromErrnoWithFilename(PyExc_OSError, path);
goto error_value;
}
dfd = open(path, O_RDONLY);
if (dfd == -1) {
PyErr_SetFromErrnoWithFilename(PyExc_OSError, path);
goto error_value;
}
list = PyList_New(0);
if (!list)
goto error_dir;
do {
count = LISTDIR_BATCH_SIZE;
err = getdirentriesattr(dfd, &requested_attr, &attrbuf,
sizeof(attrbuf), &count, &basep_unused,
&new_state, 0);
if (err < 0) {
if (errno == ENOTSUP) {
/* We're on a filesystem that doesn't support
getdirentriesattr. Fall back to the
stat-based implementation. */
*fallback = true;
} else
PyErr_SetFromErrnoWithFilename(PyExc_OSError, path);
goto error;
}
if (!state_seen) {
old_state = new_state;
state_seen = true;
} else if (old_state != new_state) {
/* There's an edge case with getdirentriesattr. Consider
the following initial list of files:
a
b
<--
c
d
If the iteration is paused at the arrow, and b is
deleted before it is resumed, getdirentriesattr will
not return d at all! Ordinarily we're expected to
restart the iteration from the beginning. To avoid
getting stuck in a retry loop here, fall back to
stat. */
*fallback = true;
goto error;
}
entry = (attrbuf_entry *)attrbuf;
for (index = 0; index < count; index++) {
char *filename = ((char *)&entry->name) +
entry->name.attr_dataoffset;
if (!strcmp(filename, ".") || !strcmp(filename, ".."))
continue;
kind = attrkind(entry);
if (kind == -1) {
PyErr_Format(PyExc_OSError,
"unknown object type %u for file "
"%s%s!",
entry->obj_type, path, filename);
goto error;
}
/* quit early? */
if (skip && kind == S_IFDIR && !strcmp(filename, skip)) {
ret = PyList_New(0);
goto error;
}
if (keepstat) {
/* from the getattrlist(2) man page: "Only the
permission bits ... are valid". */
st.st_mode = (entry->access_mask & ~S_IFMT) | kind;
st.st_mtime = entry->mtime.tv_sec;
st.st_size = entry->size;
stat = makestat(&st);
if (!stat)
goto error;
elem = Py_BuildValue("siN", filename, kind, stat);
} else
elem = Py_BuildValue("si", filename, kind);
if (!elem)
goto error;
stat = NULL;
PyList_Append(list, elem);
Py_DECREF(elem);
entry = (attrbuf_entry *)((char *)entry + entry->length);
}
} while (err == 0);
ret = list;
Py_INCREF(ret);
error:
Py_DECREF(list);
Py_XDECREF(stat);
error_dir:
close(dfd);
error_value:
return ret;
}
#endif /* __APPLE__ */
static PyObject *_listdir(char *path, int pathlen, int keepstat, char *skip)
{
osutil: use getdirentriesattr on OS X if possible This is a significant win for large repositories on OS X, especially with a cold cache. Unfortunately we need to keep the lstat-based implementation around for two reasons: - Not all filesystems support this call. - There's an edge case in which it's best to fall back to avoid a retry loop. More about this in the comments. The below tests are all performed on a Mac with an SSD running OS X 10.9, on a repository with over 200k files. The results are best of 5 with simulated best-effort conditions. The gains with a hot cache are pretty impressive: 'hg status' goes from 5.18 seconds to 3.79 seconds. However, a repository that large will probably already be using something like hgwatchman [1], which helps much more (for this repo, 'hg status' with hgwatchman is approximately 1 second). Where this really helps is when the cache is cold [2]: hg status goes from 31.0 seconds to 9.66. See http://lists.apple.com/archives/filesystem-dev/2014/Dec/msg00002.html for some more discussion about this function. This is based on a patch by Sean Farley <sean@farley.io>. [1] https://bitbucket.org/facebook/hgwatchman [2] There appears to be no easy way to clear the file cache (aka "vnodes") on OS X short of rebooting. purge(8) purportedly does that but in my testing had little effect. The workaround I came up with was to assume that vnode eviction was LRU, make sure the kern.maxvnodes sysctl is smaller than the size of the repository, then make sure we'd always miss the cache by running 'hg status' in another clone of the repository before running it in the test repository.
2015-03-26 01:55:31 +03:00
#ifdef __APPLE__
PyObject *ret;
bool fallback = false;
ret = _listdir_batch(path, pathlen, keepstat, skip, &fallback);
if (ret != NULL || !fallback)
return ret;
#endif
return _listdir_stat(path, pathlen, keepstat, skip);
}
static PyObject *statfiles(PyObject *self, PyObject *args)
{
2012-12-04 01:17:01 +04:00
PyObject *names, *stats;
Py_ssize_t i, count;
if (!PyArg_ParseTuple(args, "O:statfiles", &names))
return NULL;
count = PySequence_Length(names);
if (count == -1) {
PyErr_SetString(PyExc_TypeError, "not a sequence");
return NULL;
}
stats = PyList_New(count);
if (stats == NULL)
return NULL;
for (i = 0; i < count; i++) {
PyObject *stat, *pypath;
2012-12-04 01:17:01 +04:00
struct stat st;
int ret, kind;
char *path;
/* With a large file count or on a slow filesystem,
don't block signals for long (issue4878). */
if ((i % 1000) == 999 && PyErr_CheckSignals() == -1)
goto bail;
pypath = PySequence_GetItem(names, i);
if (!pypath)
goto bail;
path = PyBytes_AsString(pypath);
2012-12-04 01:17:01 +04:00
if (path == NULL) {
Py_DECREF(pypath);
2012-12-04 01:17:01 +04:00
PyErr_SetString(PyExc_TypeError, "not a string");
goto bail;
}
ret = lstat(path, &st);
Py_DECREF(pypath);
2012-12-04 01:17:01 +04:00
kind = st.st_mode & S_IFMT;
if (ret != -1 && (kind == S_IFREG || kind == S_IFLNK)) {
stat = makestat(&st);
if (stat == NULL)
goto bail;
PyList_SET_ITEM(stats, i, stat);
} else {
Py_INCREF(Py_None);
PyList_SET_ITEM(stats, i, Py_None);
}
}
return stats;
bail:
2012-12-04 01:17:01 +04:00
Py_DECREF(stats);
return NULL;
}
/*
* recvfds() simply does not release GIL during blocking io operation because
* command server is known to be single-threaded.
*
* Old systems such as Solaris don't provide CMSG_LEN, msg_control, etc.
* Currently, recvfds() is not supported on these platforms.
*/
#ifdef CMSG_LEN
static ssize_t recvfdstobuf(int sockfd, int **rfds, void *cbuf, size_t cbufsize)
{
char dummy[1];
struct iovec iov = {dummy, sizeof(dummy)};
struct msghdr msgh = {0};
struct cmsghdr *cmsg;
msgh.msg_iov = &iov;
msgh.msg_iovlen = 1;
msgh.msg_control = cbuf;
msgh.msg_controllen = (socklen_t)cbufsize;
if (recvmsg(sockfd, &msgh, 0) < 0)
return -1;
for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg;
cmsg = CMSG_NXTHDR(&msgh, cmsg)) {
if (cmsg->cmsg_level != SOL_SOCKET ||
cmsg->cmsg_type != SCM_RIGHTS)
continue;
*rfds = (int *)CMSG_DATA(cmsg);
return (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
}
*rfds = cbuf;
return 0;
}
static PyObject *recvfds(PyObject *self, PyObject *args)
{
int sockfd;
int *rfds = NULL;
ssize_t rfdscount, i;
char cbuf[256];
PyObject *rfdslist = NULL;
if (!PyArg_ParseTuple(args, "i", &sockfd))
return NULL;
rfdscount = recvfdstobuf(sockfd, &rfds, cbuf, sizeof(cbuf));
if (rfdscount < 0)
return PyErr_SetFromErrno(PyExc_OSError);
rfdslist = PyList_New(rfdscount);
if (!rfdslist)
goto bail;
for (i = 0; i < rfdscount; i++) {
PyObject *obj = PyLong_FromLong(rfds[i]);
if (!obj)
goto bail;
PyList_SET_ITEM(rfdslist, i, obj);
}
return rfdslist;
bail:
Py_XDECREF(rfdslist);
return NULL;
}
#endif /* CMSG_LEN */
osutil: implement setprocname to set process title for some platforms This patch adds a simple setprocname method to osutil. The operation is not defined by any standard and is platform-specific, the current implementation tries to cover some major platforms (ex. Linux, OS X, FreeBSD) that is relatively easy to support. Other platforms (Windows [4], other BSDs, ...) can be added in the future. The current implementation supports two methods to change process title: a. setproctitle if available (works in FreeBSD). b. rewrite argv in place (works in Linux [1] and Mac OS X). [2] [3] [1]: Linux has "prctl(PR_SET_NAME, ...)" but 1) it has 16-byte limit, which is too small; 2) it is not quite equivalent to what we want - it changes "/proc/self/comm", not "/proc/self/cmdline" - "comm" change won't show up in "ps" output unless "-o comm" is used. [2]: The implementation does not rewrite the **environ buffer like some other implementations do, just to make the code simpler and safer. However, this also means the buffer size we can rewrite is significantly shorter. If we are really greedy and want the "environ" space, we can change the implementation later. [3]: It requires a CPython private API: Py_GetArgcArgv to get the original argv. Unfortunately Python 3 makes a copy of argv and returns the wchar_t version, so it is not supported for now. (if we really want to, we could count backwards from "char **environ", given known argc and argv, not sure if that's a good idea - probably not) [4]: The feature is aimed to make it easier for forked command server processes to show what they are doing. Since Windows does not support fork(), despite it's a major platform, its support is not added in this patch.
2016-11-12 00:11:17 +03:00
#if defined(HAVE_SETPROCTITLE)
/* setproctitle is the first choice - available in FreeBSD */
#define SETPROCNAME_USE_SETPROCTITLE
#elif (defined(__linux__) || defined(__APPLE__)) && PY_MAJOR_VERSION == 2
/* rewrite the argv buffer in place - works in Linux and OS X. Py_GetArgcArgv
* in Python 3 returns the copied wchar_t **argv, thus unsupported. */
#define SETPROCNAME_USE_ARGVREWRITE
#else
#define SETPROCNAME_USE_NONE
#endif
#ifndef SETPROCNAME_USE_NONE
static PyObject *setprocname(PyObject *self, PyObject *args)
{
const char *name = NULL;
if (!PyArg_ParseTuple(args, "s", &name))
return NULL;
#if defined(SETPROCNAME_USE_SETPROCTITLE)
setproctitle("%s", name);
#elif defined(SETPROCNAME_USE_ARGVREWRITE)
{
static char *argvstart = NULL;
static size_t argvsize = 0;
if (argvstart == NULL) {
int argc = 0, i;
char **argv = NULL;
char *argvend;
extern void Py_GetArgcArgv(int *argc, char ***argv);
Py_GetArgcArgv(&argc, &argv);
/* Check the memory we can use. Typically, argv[i] and
* argv[i + 1] are continuous. */
argvend = argvstart = argv[0];
for (i = 0; i < argc; ++i) {
if (argv[i] > argvend || argv[i] < argvstart)
break; /* not continuous */
size_t len = strlen(argv[i]);
argvend = argv[i] + len + 1 /* '\0' */;
}
if (argvend > argvstart) /* sanity check */
argvsize = argvend - argvstart;
}
if (argvstart && argvsize > 1) {
int n = snprintf(argvstart, argvsize, "%s", name);
if (n >= 0 && (size_t)n < argvsize)
memset(argvstart + n, 0, argvsize - n);
}
}
#endif
Py_RETURN_NONE;
}
#endif /* ndef SETPROCNAME_USE_NONE */
#endif /* ndef _WIN32 */
static PyObject *listdir(PyObject *self, PyObject *args, PyObject *kwargs)
{
PyObject *statobj = NULL; /* initialize - optional arg */
PyObject *skipobj = NULL; /* initialize - optional arg */
char *path, *skip = NULL;
int wantstat, plen;
static char *kwlist[] = {"path", "stat", "skip", NULL};
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "s#|OO:listdir",
kwlist, &path, &plen, &statobj, &skipobj))
return NULL;
wantstat = statobj && PyObject_IsTrue(statobj);
if (skipobj && skipobj != Py_None) {
skip = PyBytes_AsString(skipobj);
if (!skip)
return NULL;
}
return _listdir(path, plen, wantstat, skip);
}
#ifdef _WIN32
static PyObject *posixfile(PyObject *self, PyObject *args, PyObject *kwds)
{
static char *kwlist[] = {"name", "mode", "buffering", NULL};
PyObject *file_obj = NULL;
char *name = NULL;
char *mode = "rb";
DWORD access = 0;
DWORD creation;
HANDLE handle;
int fd, flags = 0;
int bufsize = -1;
char m0, m1, m2;
char fpmode[4];
int fppos = 0;
int plus;
FILE *fp;
if (!PyArg_ParseTupleAndKeywords(args, kwds, "et|si:posixfile", kwlist,
Py_FileSystemDefaultEncoding,
&name, &mode, &bufsize))
return NULL;
m0 = mode[0];
m1 = m0 ? mode[1] : '\0';
m2 = m1 ? mode[2] : '\0';
plus = m1 == '+' || m2 == '+';
fpmode[fppos++] = m0;
if (m1 == 'b' || m2 == 'b') {
flags = _O_BINARY;
fpmode[fppos++] = 'b';
}
else
flags = _O_TEXT;
if (m0 == 'r' && !plus) {
flags |= _O_RDONLY;
access = GENERIC_READ;
} else {
/*
work around http://support.microsoft.com/kb/899149 and
set _O_RDWR for 'w' and 'a', even if mode has no '+'
*/
flags |= _O_RDWR;
access = GENERIC_READ | GENERIC_WRITE;
fpmode[fppos++] = '+';
}
fpmode[fppos++] = '\0';
switch (m0) {
case 'r':
creation = OPEN_EXISTING;
break;
case 'w':
creation = CREATE_ALWAYS;
break;
case 'a':
creation = OPEN_ALWAYS;
flags |= _O_APPEND;
break;
default:
PyErr_Format(PyExc_ValueError,
"mode string must begin with one of 'r', 'w', "
"or 'a', not '%c'", m0);
goto bail;
}
handle = CreateFile(name, access,
FILE_SHARE_READ | FILE_SHARE_WRITE |
FILE_SHARE_DELETE,
NULL,
creation,
FILE_ATTRIBUTE_NORMAL,
0);
if (handle == INVALID_HANDLE_VALUE) {
PyErr_SetFromWindowsErrWithFilename(GetLastError(), name);
goto bail;
}
2010-01-25 09:05:27 +03:00
fd = _open_osfhandle((intptr_t)handle, flags);
if (fd == -1) {
CloseHandle(handle);
PyErr_SetFromErrnoWithFilename(PyExc_IOError, name);
goto bail;
}
#ifndef IS_PY3K
fp = _fdopen(fd, fpmode);
if (fp == NULL) {
_close(fd);
PyErr_SetFromErrnoWithFilename(PyExc_IOError, name);
goto bail;
}
file_obj = PyFile_FromFile(fp, name, mode, fclose);
if (file_obj == NULL) {
fclose(fp);
goto bail;
}
PyFile_SetBufSize(file_obj, bufsize);
#else
file_obj = PyFile_FromFd(fd, name, mode, bufsize, NULL, NULL, NULL, 1);
if (file_obj == NULL)
goto bail;
#endif
bail:
PyMem_Free(name);
return file_obj;
}
#endif
#ifdef __APPLE__
#include <ApplicationServices/ApplicationServices.h>
static PyObject *isgui(PyObject *self)
{
2011-03-23 17:41:58 +03:00
CFDictionaryRef dict = CGSessionCopyCurrentDictionary();
if (dict != NULL) {
CFRelease(dict);
Py_RETURN_TRUE;
2011-03-23 17:41:58 +03:00
} else {
Py_RETURN_FALSE;
2011-03-23 17:41:58 +03:00
}
}
#endif
static char osutil_doc[] = "Native operating system services.";
static PyMethodDef methods[] = {
{"listdir", (PyCFunction)listdir, METH_VARARGS | METH_KEYWORDS,
"list a directory\n"},
#ifdef _WIN32
{"posixfile", (PyCFunction)posixfile, METH_VARARGS | METH_KEYWORDS,
"Open a file with POSIX-like semantics.\n"
"On error, this function may raise either a WindowsError or an IOError."},
#else
{"statfiles", (PyCFunction)statfiles, METH_VARARGS | METH_KEYWORDS,
"stat a series of files or symlinks\n"
"Returns None for non-existent entries and entries of other types.\n"},
#ifdef CMSG_LEN
{"recvfds", (PyCFunction)recvfds, METH_VARARGS,
"receive list of file descriptors via socket\n"},
#endif
osutil: implement setprocname to set process title for some platforms This patch adds a simple setprocname method to osutil. The operation is not defined by any standard and is platform-specific, the current implementation tries to cover some major platforms (ex. Linux, OS X, FreeBSD) that is relatively easy to support. Other platforms (Windows [4], other BSDs, ...) can be added in the future. The current implementation supports two methods to change process title: a. setproctitle if available (works in FreeBSD). b. rewrite argv in place (works in Linux [1] and Mac OS X). [2] [3] [1]: Linux has "prctl(PR_SET_NAME, ...)" but 1) it has 16-byte limit, which is too small; 2) it is not quite equivalent to what we want - it changes "/proc/self/comm", not "/proc/self/cmdline" - "comm" change won't show up in "ps" output unless "-o comm" is used. [2]: The implementation does not rewrite the **environ buffer like some other implementations do, just to make the code simpler and safer. However, this also means the buffer size we can rewrite is significantly shorter. If we are really greedy and want the "environ" space, we can change the implementation later. [3]: It requires a CPython private API: Py_GetArgcArgv to get the original argv. Unfortunately Python 3 makes a copy of argv and returns the wchar_t version, so it is not supported for now. (if we really want to, we could count backwards from "char **environ", given known argc and argv, not sure if that's a good idea - probably not) [4]: The feature is aimed to make it easier for forked command server processes to show what they are doing. Since Windows does not support fork(), despite it's a major platform, its support is not added in this patch.
2016-11-12 00:11:17 +03:00
#ifndef SETPROCNAME_USE_NONE
{"setprocname", (PyCFunction)setprocname, METH_VARARGS,
"set process title (best-effort)\n"},
#endif
osutil: implement setprocname to set process title for some platforms This patch adds a simple setprocname method to osutil. The operation is not defined by any standard and is platform-specific, the current implementation tries to cover some major platforms (ex. Linux, OS X, FreeBSD) that is relatively easy to support. Other platforms (Windows [4], other BSDs, ...) can be added in the future. The current implementation supports two methods to change process title: a. setproctitle if available (works in FreeBSD). b. rewrite argv in place (works in Linux [1] and Mac OS X). [2] [3] [1]: Linux has "prctl(PR_SET_NAME, ...)" but 1) it has 16-byte limit, which is too small; 2) it is not quite equivalent to what we want - it changes "/proc/self/comm", not "/proc/self/cmdline" - "comm" change won't show up in "ps" output unless "-o comm" is used. [2]: The implementation does not rewrite the **environ buffer like some other implementations do, just to make the code simpler and safer. However, this also means the buffer size we can rewrite is significantly shorter. If we are really greedy and want the "environ" space, we can change the implementation later. [3]: It requires a CPython private API: Py_GetArgcArgv to get the original argv. Unfortunately Python 3 makes a copy of argv and returns the wchar_t version, so it is not supported for now. (if we really want to, we could count backwards from "char **environ", given known argc and argv, not sure if that's a good idea - probably not) [4]: The feature is aimed to make it easier for forked command server processes to show what they are doing. Since Windows does not support fork(), despite it's a major platform, its support is not added in this patch.
2016-11-12 00:11:17 +03:00
#endif /* ndef _WIN32 */
#ifdef __APPLE__
2011-03-23 17:41:58 +03:00
{
"isgui", (PyCFunction)isgui, METH_NOARGS,
"Is a CoreGraphics session available?"
},
#endif
{NULL, NULL}
};
#ifdef IS_PY3K
static struct PyModuleDef osutil_module = {
PyModuleDef_HEAD_INIT,
"osutil",
osutil_doc,
-1,
methods
};
PyMODINIT_FUNC PyInit_osutil(void)
{
if (PyType_Ready(&listdir_stat_type) < 0)
return NULL;
return PyModule_Create(&osutil_module);
}
#else
PyMODINIT_FUNC initosutil(void)
{
if (PyType_Ready(&listdir_stat_type) == -1)
return;
Py_InitModule3("osutil", methods, osutil_doc);
}
#endif