radare2/libr/io/io.c

839 lines
21 KiB
C

/* radare2 - LGPL - Copyright 2008-2018 - condret, pancake, alvaro_fe */
#include <r_io.h>
#include <sdb.h>
#include <config.h>
R_LIB_VERSION (r_io);
static int fd_read_at_wrap (RIO *io, int fd, ut64 addr, ut8 *buf, int len, RIOMap *map, void *user) {
return r_io_fd_read_at (io, fd, addr, buf, len);
}
static int fd_write_at_wrap (RIO *io, int fd, ut64 addr, ut8 *buf, int len, RIOMap *map, void *user) {
return r_io_fd_write_at (io, fd, addr, buf, len);
}
static int al_fd_read_at_wrap (RIO *io, int fd, ut64 addr, ut8 *buf, int len, RIOMap *map, void *user) {
RIOAccessLog *log = (RIOAccessLog *)user;
RIOAccessLogElement *ale = R_NEW0 (RIOAccessLogElement);
int rlen = r_io_fd_read_at (io, fd, addr, buf, len);
if (ale) {
ale->expect_len = len;
ale->len = rlen;
ale->buf_idx = (int)(size_t)(buf - log->buf);
ale->flags = map->flags;
ale->fd = fd;
ale->mapid = map->id;
ale->paddr = addr;
ale->vaddr = map->itv.addr + (addr - map->delta);
r_list_append (log->log, ale);
} else {
log->allocation_failed = true;
}
return rlen;
}
static int al_fd_write_at_wrap (RIO *io, int fd, ut64 addr, ut8 *buf, int len, RIOMap *map, void *user) {
RIOAccessLog *log = (RIOAccessLog *)user;
RIOAccessLogElement *ale = R_NEW0 (RIOAccessLogElement);
int rlen = r_io_fd_write_at (io, fd, addr, buf, len);
if (ale) {
ale->expect_len = len;
ale->len = rlen;
ale->buf_idx = (int)(size_t)(buf - log->buf);
ale->flags = map->flags;
ale->fd = fd;
ale->mapid = map->id;
ale->paddr = addr;
ale->vaddr = map->itv.addr + (addr - map->delta);
r_list_append (log->log, ale);
} else {
log->allocation_failed = true;
}
return rlen;
}
typedef int (*cbOnIterMap)(RIO *io, int fd, ut64 addr, ut8 *buf, int len, RIOMap *map, void *user);
// If prefix_mode is true, returns the number of bytes of operated prefix; returns < 0 on error.
// If prefix_mode is false, operates in non-stop mode and returns true iff all IO operations on overlapped maps are complete.
static st64 on_map_skyline(RIO *io, ut64 vaddr, ut8 *buf, int len, int match_flg, cbOnIterMap op, bool prefix_mode) {
const RVector *skyline = &io->map_skyline;
ut64 addr = vaddr;
int i;
bool ret = true, wrap = !prefix_mode && vaddr + len < vaddr;
#define CMP(addr, part) (addr < r_itv_end (((RIOMapSkyline *)part)->itv) - 1 ? -1 : \
addr > r_itv_end (((RIOMapSkyline *)part)->itv) - 1 ? 1 : 0)
// Let i be the first skyline part whose right endpoint > addr
if (!len) {
i = skyline->len;
} else {
r_vector_lower_bound (skyline, addr, i, CMP);
if (i == skyline->len && wrap) {
wrap = false;
i = 0;
addr = 0;
}
}
#undef CMP
while (i < skyline->len) {
const RIOMapSkyline *part = skyline->a[i];
// Right endpoint <= addr
if (r_itv_end (part->itv) - 1 < addr) {
i++;
if (wrap && i == skyline->len) {
wrap = false;
i = 0;
addr = 0;
}
continue;
}
if (addr < part->itv.addr) {
// [addr, part->itv.addr) is a gap
if (prefix_mode || len <= part->itv.addr - vaddr) {
break;
}
addr = part->itv.addr;
}
// Now left endpoint <= addr < right endpoint
ut64 len1 = R_MIN (vaddr + len - addr, r_itv_end (part->itv) - addr);
// The map satisfies the permission requirement or p_cache is enabled
if (((part->map->flags & match_flg) == match_flg || io->p_cache)) {
st64 result = op (io, part->map->fd, part->map->delta + addr - part->map->itv.addr,
buf + addr - vaddr, len1, part->map, NULL);
if (prefix_mode) {
if (result < 0) {
return result;
}
addr += result;
if (result != len1) {
break;
}
} else {
if (result != len1) {
ret = false;
}
addr += len1;
}
} else if (prefix_mode) {
break;
} else {
addr += len1;
ret = false;
}
// Reaches the end
if (addr == vaddr + len) {
break;
}
// Wrap to the beginning of skyline if address wraps
if (!addr) {
i = 0;
}
}
return prefix_mode ? addr - vaddr : ret;
}
// Precondition: len > 0
// Non-stop IO
// Returns true iff all reads/writes on overlapped maps are complete.
static bool onIterMap(SdbListIter *iter, RIO *io, ut64 vaddr, ut8 *buf,
int len, int match_flg, cbOnIterMap op, void *user) {
// vendaddr may be 0 to denote 2**64
ut64 vendaddr = vaddr + len, len1;
int t;
bool ret = true;
for (; iter; iter = iter->p) {
RIOMap *map = (RIOMap *)iter->data;
ut64 to = r_itv_end (map->itv);
if (r_itv_overlap2 (map->itv, vaddr, len)) {
if ((map->flags & match_flg) == match_flg || io->p_cache) {
t = vaddr < map->itv.addr
? op (io, map->fd, map->delta, buf + map->itv.addr - vaddr,
len1 = R_MIN (vendaddr - map->itv.addr, map->itv.size), map, user)
: op (io, map->fd, map->delta + vaddr - map->itv.addr, buf,
len1 = R_MIN (to - vaddr, len), map, user);
if (t != len1) {
ret = false;
}
}
if (vaddr < map->itv.addr) {
t = onIterMap (iter->p, io, vaddr, buf, map->itv.addr - vaddr, match_flg, op, user);
if (!t) {
ret = false;
}
}
if (to - 1 < vendaddr - 1) {
t = onIterMap (iter->p, io, to, buf + to - vaddr, vendaddr - to, match_flg, op, user);
if (!t) {
ret = false;
}
}
}
}
return ret;
}
// Precondition: len >= 0
// Non-stop IO supporting address space wraparound
// Returns true iff all reads/writes on overlapped maps are complete.
static bool onIterMap_wrap(SdbListIter *iter, RIO *io, ut64 vaddr, ut8 *buf,
int len, int match_flg, cbOnIterMap op, void *user) {
#if 1
return on_map_skyline (io, vaddr, buf, len, match_flg, op, false);
#else
if (!len) {
return true;
}
bool ret = true;
// vaddr + len > 2**64
if (vaddr > UT64_MAX - len + 1) {
ret = onIterMap (iter, io, 0, buf - vaddr, len + vaddr, match_flg, op, user);
len = -vaddr;
}
if (ret) {
ret = onIterMap (iter, io, vaddr, buf, len, match_flg, op, user);
}
return ret;
#endif
}
R_API RIO* r_io_new() {
return r_io_init (R_NEW0 (RIO));
}
R_API RIO* r_io_init(RIO* io) {
if (!io) {
return NULL;
}
io->addrbytes = 1;
r_io_desc_init (io);
r_io_map_init (io);
r_io_section_init (io);
r_io_cache_init (io);
r_io_plugin_init (io);
r_io_undo_init (io);
return io;
}
R_API RBuffer *r_io_read_buf(RIO *io, ut64 addr, int len) {
RBuffer *b = R_NEW0 (RBuffer);
if (!b) return NULL;
b->buf = malloc (len);
if (!b->buf) {
free (b);
return NULL;
}
len = r_io_read_at (io, addr, b->buf, len);
b->length = (len < 0)? 0: len;
return b;
}
R_API int r_io_write_buf(RIO *io, struct r_buf_t *b) {
return r_io_write_at (io, b->base, b->buf, b->length);
}
R_API void r_io_free(RIO *io) {
if (!io) {
return;
}
r_io_fini (io);
r_cache_free (io->buffer);
free (io);
}
R_API RIODesc *r_io_open_buffer(RIO *io, RBuffer *b, int flags, int mode) {
const int bufSize = r_buf_size (b);
char *uri = r_str_newf ("malloc://%d", bufSize);
RIODesc *desc = r_io_open_nomap (io, uri, flags, mode);
if (desc) {
r_io_desc_write (desc, r_buf_get_at(b, 0, NULL), bufSize);
}
return desc;
}
R_API RIODesc *r_io_open_nomap(RIO *io, const char *uri, int flags, int mode) {
RIODesc *desc;
if (!io) {
return NULL;
}
desc = r_io_desc_open (io, uri, flags, mode);
if ((io->autofd || !io->desc) && desc) {
io->desc = desc;
}
//set desc as current if autofd or io->desc==NULL
return desc;
}
/* opens a file and maps it to 0x0 */
R_API RIODesc* r_io_open(RIO* io, const char* uri, int flags, int mode) {
RIODesc* desc;
if (!io || !io->maps) {
return NULL;
}
desc = r_io_open_nomap (io, uri, flags, mode);
if (!desc) {
return NULL;
}
r_io_map_new (io, desc->fd, desc->flags, 0LL, 0LL, r_io_desc_size (desc), true);
return desc;
}
/* opens a file and maps it to an offset specified by the "at"-parameter */
R_API RIODesc* r_io_open_at(RIO* io, const char* uri, int flags, int mode, ut64 at) {
RIODesc* desc;
ut64 size;
if (!io || !io->maps) {
return NULL;
}
desc = r_io_open_nomap (io, uri, flags, mode);
if (!desc) {
return NULL;
}
size = r_io_desc_size (desc);
// second map
if (size && ((UT64_MAX - size + 1) < at)) {
// split map into 2 maps if only 1 big map results into interger overflow
r_io_map_new (io, desc->fd, desc->flags, UT64_MAX - at + 1, 0LL, size - (UT64_MAX - at) - 1, false);
// someone pls take a look at this confusing stuff
size = UT64_MAX - at + 1;
}
// skyline not updated
r_io_map_new (io, desc->fd, desc->flags, 0LL, at, size, false);
return desc;
}
/* opens many files, without mapping them. This should be discussed */
R_API RList* r_io_open_many(RIO* io, const char* uri, int flags, int mode) {
RList* desc_list;
RListIter* iter;
RIODesc* desc;
if (!io || !io->files || !uri) {
return NULL;
}
RIOPlugin* plugin = r_io_plugin_resolve (io, uri, 1);
if (!plugin || !plugin->open_many || !plugin->close) {
return NULL;
}
if (!(desc_list = plugin->open_many (io, uri, flags, mode))) {
return NULL;
}
r_list_foreach (desc_list, iter, desc) {
if (desc) {
if (!desc->plugin) {
desc->plugin = plugin;
}
if (!desc->uri) {
desc->uri = strdup (uri);
}
//should autofd be honored here?
r_io_desc_add (io, desc);
if (!io->desc) {
io->desc = desc;
}
}
}
return desc_list;
}
R_API bool r_io_reopen(RIO* io, int fd, int flags, int mode) {
RIODesc *old, *new;
char *uri;
if (!(old = r_io_desc_get (io, fd))) {
return false;
}
//does this really work, or do we have to handler debuggers ugly
uri = old->referer? old->referer: old->uri;
#if __WINDOWS__ //TODO: workaround, see https://github.com/radare/radare2/issues/8840
if (!r_io_desc_close (old)) {
return false; // TODO: this is an unrecoverable scenario
}
if (!(new = r_io_open_nomap (io, uri, flags, mode))) {
return false;
}
r_io_desc_exchange (io, old->fd, new->fd);
return true;
#else
if (!(new = r_io_open_nomap (io, uri, flags, mode))) {
return false;
}
r_io_desc_exchange (io, old->fd, new->fd);
return r_io_desc_close (old); // magic
#endif // __WINDOWS__
}
R_API int r_io_close_all(RIO* io) { // what about undo?
if (!io) {
return false;
}
r_io_desc_fini (io);
r_io_map_fini (io);
r_io_section_fini (io);
ls_free (io->plugins);
r_list_free (io->cache);
r_io_desc_init (io);
r_io_map_init (io);
r_io_section_init (io);
r_io_cache_init (io);
r_io_plugin_init (io);
return true;
}
R_API int r_io_pread_at(RIO* io, ut64 paddr, ut8* buf, int len) {
if (!io || !buf || len < 1) {
return 0;
}
if (io->ff) {
memset (buf, io->Oxff, len);
}
return r_io_desc_read_at (io->desc, paddr, buf, len);
}
R_API int r_io_pwrite_at(RIO* io, ut64 paddr, const ut8* buf, int len) {
if (!io) {
return 0;
}
return r_io_desc_write_at (io->desc, paddr, buf, len);
}
// Returns true iff all reads on mapped regions are successful and complete.
R_API bool r_io_vread_at_mapped(RIO* io, ut64 vaddr, ut8* buf, int len) {
if (!io || !buf || (len < 1)) {
return false;
}
if (io->ff) {
memset (buf, io->Oxff, len);
}
r_io_map_cleanup (io);
if (!io->maps) {
return false;
}
return onIterMap_wrap (io->maps->tail, io, vaddr, buf, len, R_IO_READ, fd_read_at_wrap, NULL);
}
static bool r_io_vwrite_at(RIO* io, ut64 vaddr, const ut8* buf, int len) {
if (!io || !buf || (len < 1)) {
return false;
}
r_io_map_cleanup (io);
if (!io->maps) {
return false;
}
return onIterMap_wrap (io->maps->tail, io, vaddr, (ut8*)buf, len, R_IO_WRITE, fd_write_at_wrap, NULL);
}
R_API RIOAccessLog *r_io_al_vread_at(RIO* io, ut64 vaddr, ut8* buf, int len) {
RIOAccessLog *log;
if (!io || !buf || (len < 1)) {
return NULL;
}
r_io_map_cleanup (io);
if (!io->maps || !(log = r_io_accesslog_new ())) {
return NULL;
}
if (io->ff) {
memset (buf, io->Oxff, len);
}
log->buf = buf;
onIterMap (io->maps->tail, io, vaddr, buf, len, R_IO_READ, al_fd_read_at_wrap, log);
return log;
}
R_API RIOAccessLog *r_io_al_vwrite_at(RIO* io, ut64 vaddr, const ut8* buf, int len) {
RIOAccessLog *log;
if (!io || !buf || (len < 1)) {
return NULL;
}
r_io_map_cleanup (io);
if (!io->maps || !(log = r_io_accesslog_new ())) {
return NULL;
}
log->buf = (ut8*)buf;
(void)onIterMap (io->maps->tail, io, vaddr, (ut8*)buf, len, R_IO_WRITE, al_fd_write_at_wrap, log);
return log;
}
// Deprecated, use either r_io_read_at_mapped or r_io_nread_at instead.
// For virtual mode, returns true if all reads on mapped regions are successful
// and complete.
// For physical mode, the interface is broken because the actual read bytes are
// not available. This requires fixes in all call sites.
R_API bool r_io_read_at(RIO *io, ut64 addr, ut8 *buf, int len) {
bool ret;
if (!io || !buf || len < 1) {
return false;
}
if (io->buffer_enabled) {
return !!r_io_buffer_read(io, addr, buf, len);
}
if (io->va) {
ret = r_io_vread_at_mapped(io, addr, buf, len);
} else {
ret = r_io_pread_at(io, addr, buf, len) > 0;
}
if (io->cached & R_IO_READ) {
(void)r_io_cache_read(io, addr, buf, len);
}
return ret;
}
// Returns true iff all reads on mapped regions are successful and complete.
// Unmapped regions are filled with io->Oxff in both physical and virtual modes.
// Use this function if you want to ignore gaps or do not care about the number
// of read bytes.
R_API bool r_io_read_at_mapped(RIO *io, ut64 addr, ut8 *buf, int len) {
bool ret;
if (!io || !buf) {
return false;
}
if (io->buffer_enabled) {
return !!r_io_buffer_read(io, addr, buf, len);
}
if (io->ff) {
memset(buf, io->Oxff, len);
}
if (io->va) {
ret = on_map_skyline(io, addr, buf, len, R_IO_READ, fd_read_at_wrap, false);
} else {
ret = r_io_pread_at(io, addr, buf, len) > 0;
}
if (io->cached & R_IO_READ) {
(void)r_io_cache_read(io, addr, buf, len);
}
return ret;
}
// For both virtual and physical mode, returns the number of bytes of read
// prefix.
// Returns -1 on error.
R_API int r_io_nread_at(RIO *io, ut64 addr, ut8 *buf, int len) {
int ret;
if (!io || !buf) {
return -1;
}
if (io->buffer_enabled) {
return r_io_buffer_read(io, addr, buf, len);
}
if (io->va) {
if (io->ff) {
memset(buf, io->Oxff, len);
}
ret = on_map_skyline(io, addr, buf, len, R_IO_READ, fd_read_at_wrap, true);
} else {
ret = r_io_pread_at(io, addr, buf, len);
}
if (ret > 0 && io->cached & R_IO_READ) {
(void)r_io_cache_read(io, addr, buf, len);
}
return ret;
}
R_API RIOAccessLog *r_io_al_read_at(RIO* io, ut64 addr, ut8* buf, int len) {
RIOAccessLog *log;
RIOAccessLogElement *ale = NULL;
int rlen;
if (!io || !buf || (len < 1)) {
return NULL;
}
if (io->va) {
return r_io_al_vread_at (io, addr, buf, len);
}
if (!(log = r_io_accesslog_new ())) {
return NULL;
}
log->buf = buf;
if (io->ff) {
memset (buf, io->Oxff, len);
}
rlen = r_io_pread_at (io, addr, buf, len);
if (io->cached & R_IO_READ) {
(void)r_io_cache_read (io, addr, buf, len);
}
if (!(ale = R_NEW0 (RIOAccessLogElement))) {
log->allocation_failed = true;
} else {
ale->paddr = ale->vaddr = addr;
ale->len = rlen;
ale->expect_len = len;
ale->flags = io->desc ? io->desc->flags : 0;
ale->fd = io->desc ? io->desc->fd : 0; //xxx
r_list_append (log->log, ale);
}
return log;
}
R_API bool r_io_write_at(RIO* io, ut64 addr, const ut8* buf, int len) {
int i;
bool ret = false;
ut8 *mybuf = (ut8*)buf;
if (!io || !buf || len < 1) {
return false;
}
if (io->write_mask) {
mybuf = r_mem_dup ((void*)buf, len);
for (i = 0; i < len; i++) {
//this sucks
mybuf[i] &= io->write_mask[i % io->write_mask_len];
}
}
if (io->cached & R_IO_WRITE) {
ret = r_io_cache_write (io, addr, mybuf, len);
} else if (io->va) {
ret = r_io_vwrite_at (io, addr, mybuf, len);
} else {
ret = r_io_pwrite_at (io, addr, mybuf, len) > 0;
}
if (buf != mybuf) {
free (mybuf);
}
return ret;
}
R_API bool r_io_read(RIO* io, ut8* buf, int len) {
if (!io) {
return false;
}
if (r_io_read_at (io, io->off, buf, len)) {
io->off += len;
return true;
}
return false;
}
R_API bool r_io_write(RIO* io, ut8* buf, int len) {
if (!io || !buf || len < 1) {
return false;
}
if (r_io_write_at (io, io->off, buf, len)) {
io->off += len;
return true;
}
return false;
}
R_API ut64 r_io_size(RIO* io) {
// TODO: rethink this, maybe not needed
return io? r_io_desc_size (io->desc): 0LL;
}
R_API bool r_io_is_listener(RIO* io) {
if (io && io->desc && io->desc->plugin && io->desc->plugin->listener) {
return io->desc->plugin->listener (io->desc);
}
return false;
}
R_API char *r_io_system(RIO* io, const char* cmd) {
if (io && io->desc && io->desc->plugin && io->desc->plugin->system) {
return io->desc->plugin->system (io, io->desc, cmd);
}
return NULL;
}
R_API bool r_io_resize(RIO* io, ut64 newsize) {
if (io) {
RList *maps = r_io_map_get_for_fd (io, io->desc->fd);
RIOMap *current_map;
RListIter *iter;
ut64 fd_size = r_io_fd_size (io, io->desc->fd);
r_list_foreach (maps, iter, current_map) {
// we just resize map of the same size of its fd
if (current_map->itv.size == fd_size) {
r_io_map_resize (io, current_map->id, newsize);
}
}
r_list_free (maps);
return r_io_desc_resize (io->desc, newsize);
}
return false;
}
R_API bool r_io_close(RIO *io) {
return io ? r_io_desc_close (io->desc) : false;
}
R_API int r_io_extend_at(RIO* io, ut64 addr, ut64 size) {
ut64 cur_size, tmp_size;
ut8* buffer;
if (!io || !io->desc || !io->desc->plugin || !size) {
return false;
}
if (io->desc->plugin->extend) {
int ret;
ut64 cur_off = io->off;
r_io_seek (io, addr, R_IO_SEEK_SET);
ret = r_io_desc_extend (io->desc, size);
//no need to seek here
io->off = cur_off;
return ret;
}
if ((io->desc->flags & R_IO_RW) != R_IO_RW) {
return false;
}
cur_size = r_io_desc_size (io->desc);
if (addr > cur_size) {
return false;
}
if ((UT64_MAX - size) < cur_size) {
return false;
}
if (!r_io_resize (io, cur_size + size)) {
return false;
}
if ((tmp_size = cur_size - addr) == 0LL) {
return true;
}
if (!(buffer = calloc (1, (size_t) tmp_size + 1))) {
return false;
}
r_io_pread_at (io, addr, buffer, (int) tmp_size);
/* fill with null bytes */
ut8 *empty = calloc (1, size);
if (empty) {
r_io_pwrite_at (io, addr, empty, size);
free (empty);
}
r_io_pwrite_at (io, addr + size, buffer, (int) tmp_size);
free (buffer);
return true;
}
R_API bool r_io_set_write_mask(RIO* io, const ut8* mask, int len) {
if (!io || len < 1) {
return false;
}
free (io->write_mask);
if (!mask) {
io->write_mask = NULL;
io->write_mask_len = 0;
return true;
}
io->write_mask = (ut8*) malloc (len);
memcpy (io->write_mask, mask, len);
io->write_mask_len = len;
return true;
}
R_API int r_io_bind(RIO* io, RIOBind* bnd) {
if (!io || !bnd) {
return false;
}
bnd->io = io;
bnd->init = true;
bnd->desc_use = r_io_use_fd;
bnd->desc_get = r_io_desc_get;
bnd->desc_size = r_io_desc_size;
bnd->open = r_io_open_nomap;
bnd->open_at = r_io_open_at;
bnd->close = r_io_fd_close;
bnd->read_at = r_io_read_at;
bnd->al_read_at = r_io_al_read_at;
bnd->write_at = r_io_write_at;
bnd->system = r_io_system;
bnd->fd_open = r_io_fd_open;
bnd->fd_close = r_io_fd_close;
bnd->fd_seek = r_io_fd_seek;
bnd->fd_size = r_io_fd_size;
bnd->fd_read = r_io_fd_read;
bnd->fd_write = r_io_fd_write;
bnd->fd_read_at = r_io_fd_read_at;
bnd->fd_write_at = r_io_fd_write_at;
bnd->fd_is_dbg = r_io_fd_is_dbg;
bnd->fd_get_name = r_io_fd_get_name;
bnd->fd_get_map = r_io_map_get_for_fd;
bnd->fd_remap = r_io_map_remap_fd;
bnd->al_sort = r_io_accesslog_sort;
bnd->al_free = r_io_accesslog_free;
bnd->al_buf_byflags = r_io_accesslog_getf_buf_byflags;
bnd->is_valid_offset = r_io_is_valid_offset;
bnd->addr_is_mapped = r_io_addr_is_mapped;
bnd->sections_vget = r_io_sections_vget;
bnd->section_add = r_io_section_add;
bnd->sect_vget = r_io_section_vget;
return true;
}
/* moves bytes up (+) or down (-) within the specified range */
R_API int r_io_shift(RIO* io, ut64 start, ut64 end, st64 move) {
ut8* buf;
ut64 chunksize = 0x10000;
ut64 saved_off = io->off;
ut64 src, shiftsize = r_num_abs (move);
if (!shiftsize || (end - start) <= shiftsize) {
return false;
}
ut64 rest = (end - start) - shiftsize;
if (!(buf = calloc (1, chunksize + 1))) {
return false;
}
if (move > 0) {
src = end - shiftsize;
} else {
src = start + shiftsize;
}
while (rest > 0) {
if (chunksize > rest) {
chunksize = rest;
}
if (move > 0) {
src -= chunksize;
}
r_io_read_at (io, src, buf, chunksize);
r_io_write_at (io, src + move, buf, chunksize);
if (move < 0) {
src += chunksize;
}
rest -= chunksize;
}
free (buf);
io->off = r_io_desc_seek (io->desc, saved_off, R_IO_SEEK_SET);
return true;
}
R_API int r_io_create(RIO* io, const char* file, int mode, int type) {
if (io && io->desc && io->desc->plugin && io->desc->plugin->create) {
return io->desc->plugin->create (io, file, mode, type);
}
if (type == 'd' || type == 1) {
return r_sys_mkdir (file);
}
return r_sandbox_creat (file, mode);
}
R_API ut64 r_io_seek(RIO* io, ut64 offset, int whence) {
if (!io) {
return 0LL;
}
switch (whence) {
case R_IO_SEEK_SET:
io->off = offset;
break;
case R_IO_SEEK_CUR:
io->off += offset;
break;
case R_IO_SEEK_END:
default:
io->off = r_io_desc_seek (io->desc, offset, whence);
break;
}
return io->off;
}
//remove all descs and maps
R_API int r_io_fini(RIO* io) {
if (!io) {
return false;
}
r_io_desc_cache_fini_all (io);
r_io_desc_fini (io);
r_io_map_fini (io);
r_io_section_fini (io);
ls_free (io->plugins);
r_list_free (io->cache);
r_list_free (io->undo.w_list);
if (io->runprofile) {
R_FREE (io->runprofile);
}
return true;
}