subsurface/uemis-downloader.c
Dirk Hohndel bd3df859bc Restructure the Uemis native download buffer code
Running under Valgrind showed a couple of silly bugs.

Worse, intentionally running into various error scenarios showed that we
could get the buffer handling in the raw parsing code to break down - we
would fail to process the correctly downloaded files.

To make it easier to get this right I restructured the code to collect the
XML buffer in a different way - this works much better and has stood up
well under testing so far.

Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
2012-09-27 00:29:31 -07:00

645 lines
18 KiB
C

/*
* uemis-downloader.c
*
* Copyright (c) Dirk Hohndel <dirk@hohndel.org>
* released under GPL2
*
* very (VERY) loosely based on the algorithms found in Java code by Fabian Gast <fgast@only640k.net>
* which was released under the BSD-STYLE BEER WARE LICENSE
* I believe that I only used the information about HOW to do this (download data from the Uemis
* Zurich) but did not actually use any of his copyrighted code, therefore the license under which
* he released his code does not apply to this new implementation in C
*/
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <stdio.h>
#include <pthread.h>
#include <unistd.h>
#include <string.h>
#include "uemis.h"
#include "dive.h"
#include "divelist.h"
#include "display.h"
#include "display-gtk.h"
#define ERR_FS_ALMOST_FULL "Uemis Zurich: File System is almost full\nDisconnect/reconnect the dive computer\nand try again"
#define ERR_FS_FULL "Uemis Zurich: File System is full\nDisconnect/reconnect the dive computer\nand try again"
#define ERR_FS_SHORT_WRITE "Short write to req.txt file\nIs the Uemis Zurich plugged in correctly?"
#define BUFLEN 2048
#define NUM_PARAM_BUFS 6
#define UEMIS_TIMEOUT 100000
static char *param_buff[NUM_PARAM_BUFS];
static char *reqtxt_path;
static int reqtxt_file;
static int filenr;
static int number_of_files;
static char *mbuf = NULL;
static int mbuf_size = 0;
struct argument_block {
const char *mountpath;
char **max_dive_data;
char **xml_buffer;
progressbar_t *progress;
};
static int import_thread_done = 0, import_thread_cancelled;
static const char *progress_bar_text = "";
static double progress_bar_fraction = 0.0;
static GError *error(const char *fmt, ...)
{
va_list args;
GError *error;
va_start(args, fmt);
error = g_error_new_valist(
g_quark_from_string("subsurface"),
DIVE_ERROR_PARSE, fmt, args);
va_end(args);
return error;
}
/* send text to the importer progress bar */
static void uemis_info(const char *fmt, ...)
{
static char buffer[32];
va_list ap;
va_start(ap, fmt);
vsnprintf(buffer, sizeof(buffer), fmt, ap);
va_end(ap);
progress_bar_text = buffer;
}
static long bytes_available(int file)
{
long result;
long now = lseek(file, 0, SEEK_CUR);
result = lseek(file, 0, SEEK_END);
lseek(file, now, SEEK_SET);
return result;
}
static int number_of_file(char *path)
{
int count = 0;
GDir *dir = g_dir_open(path, 0, NULL);
while (g_dir_read_name(dir))
count++;
return count;
}
/* Check if there's a req.txt file and get the starting filenr from it.
* Test for the maximum number of ANS files (I believe this is always
* 4000 but in case there are differences depending on firmware, this
* code is easy enough */
static gboolean uemis_init(const char *path)
{
char *ans_path;
int i;
if (!path)
return FALSE;
/* let's check if this is indeed a Uemis DC */
reqtxt_path = g_build_filename(path, "/req.txt", NULL);
reqtxt_file = g_open(reqtxt_path, O_RDONLY, 0666);
if (!reqtxt_file)
return FALSE;
if (bytes_available(reqtxt_file) > 5) {
char tmp[6];
read(reqtxt_file, tmp, 5);
tmp[5] = '\0';
#if UEMIS_DEBUG > 2
fprintf(debugfile, "::r req.txt \"%s\"\n", tmp);
#endif
if (sscanf(tmp + 1, "%d", &filenr) != 1)
return FALSE;
}
#if UEMIS_DEBUG > 2
else {
fprintf(debugfile, "::r req.txt skipped as there were fewer than 5 bytes\n");
}
#endif
close (reqtxt_file);
/* It would be nice if we could simply go back to the first set of
* ANS files. Something like the below - unfortunately this is known
* to fail - more information from Uemis is needed here.
* Without code like this it is very easy when downloading large amounts
* of dives to run out of space on the dive computer - which can only
* be fixed by unmounting and rebooting the DC
* reqtxt_file = g_open(reqtxt_path, O_RDWR | O_CREAT | O_TRUNC, 0666);
* write(reqtxt_file, "n0001", 5);
* close(reqtxt_file);
* filenr = 1;
*/
ans_path = g_build_filename(path, "ANS", NULL);
number_of_files = number_of_file(ans_path);
g_free(ans_path);
/* initialize the array in which we collect the answers */
for (i = 0; i < NUM_PARAM_BUFS; i++)
param_buff[i] = "";
return TRUE;
}
static void str_append_with_delim(char *s, char *t)
{
int len = strlen(s);
snprintf(s + len, BUFLEN - len, "%s{", t);
}
/* The communication protocoll with the DC is truly funky.
* After you write your request to the req.txt file you call this function.
* It writes the number of the next ANS file at the beginning of the req.txt
* file (prefixed by 'n' or 'r') and then again at the very end of it, after
* the full request (this time without the prefix).
* Then it syncs (not needed on Windows) and closes the file. */
static void trigger_response(int file, char *command, int nr, long tailpos)
{
char fl[10];
snprintf(fl, 8, "%s%04d", command, nr);
#if UEMIS_DEBUG > 2
fprintf(debugfile,"::: %s (after seeks)\n", fl);
#endif
lseek(file, 0, SEEK_SET);
write(file, fl, strlen(fl));
lseek(file, tailpos, SEEK_SET);
write(file, fl + 1, strlen(fl + 1));
#ifndef WIN32
fsync(file);
#endif
close(file);
}
static char *next_token(char **buf)
{
char *q, *p = strchr(*buf, '{');
if (p)
*p = '\0';
else
p = *buf + strlen(*buf) - 1;
q = *buf;
*buf = p + 1;
return q;
}
/* poor man's tokenizer that understands a quoted delimter ('{') */
static char *next_segment(char *buf, int *offset, int size)
{
int i = *offset;
int seg_size;
gboolean done = FALSE;
char *segment;
while (!done) {
if (i < size) {
if (buf[i] == '\\' && i < size - 1 &&
(buf[i+1] == '\\' || buf[i+1] == '{'))
memcpy(buf + i, buf + i + 1, size - i - 1);
else if (buf[i] == '{')
done = TRUE;
i++;
} else {
done = TRUE;
}
}
seg_size = i - *offset - 1;
if (seg_size < 0)
seg_size = 0;
segment = malloc(seg_size + 1);
memcpy(segment, buf + *offset, seg_size);
segment[seg_size] = '\0';
*offset = i;
return segment;
}
/* a dynamically growing buffer to store the potentially massive responses.
* The binary data block can be more than 100k in size (base64 encoded) */
static void buffer_add(char **buffer, int *buffer_size, char *buf)
{
if (buf) {
if (! *buffer) {
*buffer = strdup(buf);
*buffer_size = strlen(*buffer) + 1;
} else {
*buffer_size += strlen(buf);
*buffer = realloc(*buffer, *buffer_size);
strcat(*buffer, buf);
}
#if UEMIS_DEBUG > 5
fprintf(debugfile,"added \"%s\" to buffer - new length %d\n", buf, *buffer_size);
#endif
}
}
/* are there more ANS files we can check? */
static gboolean next_file(int max)
{
if (filenr >= max)
return FALSE;
filenr++;
return TRUE;
}
/* ultra-simplistic; it doesn't deal with the case when the object_id is
* split across two chunks. It also doesn't deal with the discrepancy between
* object_id and dive number as understood by the dive computer */
static void show_progress(char *buf)
{
char *object;
object = strstr(buf, "object_id");
if (object) {
/* let the user know which dive we are working on */
char tmp[10];
char *p = object + 14;
char *t = tmp;
if (p < buf + strlen(buf)) {
while (*p != '{' && t < tmp + 9)
*t++ = *p++;
*t = '\0';
uemis_info("Reading dive %s", tmp);
}
}
}
/* send a request to the dive computer and collect the answer */
static gboolean uemis_get_answer(const char *path, char *request, int n_param_in,
int n_param_out, char **error_text)
{
int i = 0, file_length;
char sb[BUFLEN];
char fl[13];
char tmp[101];
gboolean searching = TRUE;
gboolean assembling_mbuf = FALSE;
gboolean ismulti = FALSE;
gboolean found_answer = FALSE;
gboolean more_files = TRUE;
gboolean answer_in_mbuf = FALSE;
char *ans_path;
int ans_file;
reqtxt_file = g_open(reqtxt_path, O_RDWR | O_CREAT, 0666);
snprintf(sb, BUFLEN, "n%04d12345678", filenr);
str_append_with_delim(sb, request);
for (i = 0; i < n_param_in; i++)
str_append_with_delim(sb, param_buff[i]);
if (! strcmp(request, "getDivelogs") || ! strcmp(request, "getDeviceData")) {
answer_in_mbuf = TRUE;
str_append_with_delim(sb, "");
}
str_append_with_delim(sb, "");
file_length = strlen(sb);
snprintf(fl, 10, "%08d", file_length - 13);
memcpy(sb + 5, fl, strlen(fl));
#ifdef UEMIS_DEBUG
fprintf(debugfile,"::w req.txt \"%s\"\n", sb);
#endif
if (write(reqtxt_file, sb, strlen(sb)) != strlen(sb)) {
*error_text = ERR_FS_SHORT_WRITE;
return FALSE;
}
if (! next_file(number_of_files)) {
*error_text = ERR_FS_FULL;
more_files = FALSE;
}
trigger_response(reqtxt_file, "n", filenr, file_length);
usleep(UEMIS_TIMEOUT);
mbuf = NULL;
mbuf_size = 0;
while (searching || assembling_mbuf) {
progress_bar_fraction = filenr / 4000.0;
snprintf(fl, 13, "ANS%d.TXT", filenr - 1);
ans_path = g_build_filename(path, "ANS", fl, NULL);
ans_file = g_open(ans_path, O_RDONLY, 0666);
read(ans_file, tmp, 100);
close(ans_file);
#if UEMIS_DEBUG > 3
tmp[100]='\0';
fprintf(debugfile, "::t %s \"%s\"\n", ans_path, tmp);
#endif
g_free(ans_path);
if (tmp[0] == '1') {
searching = FALSE;
if (tmp[1] == 'm') {
assembling_mbuf = TRUE;
ismulti = TRUE;
}
if (tmp[2] == 'e')
assembling_mbuf = FALSE;
if (assembling_mbuf) {
if (! next_file(number_of_files)) {
*error_text = ERR_FS_FULL;
more_files = FALSE;
assembling_mbuf = FALSE;
}
reqtxt_file = g_open(reqtxt_path, O_RDWR | O_CREAT, 0666);
trigger_response(reqtxt_file, "n", filenr, file_length);
}
} else {
if (! next_file(number_of_files - 1)) {
*error_text = ERR_FS_FULL;
more_files = FALSE;
assembling_mbuf = FALSE;
searching = FALSE;
}
reqtxt_file = g_open(reqtxt_path, O_RDWR | O_CREAT, 0666);
trigger_response(reqtxt_file, "r", filenr, file_length);
usleep(UEMIS_TIMEOUT);
}
if (ismulti && more_files && tmp[0] == '1') {
int size;
snprintf(fl, 13, "ANS%d.TXT", assembling_mbuf ? filenr - 2 : filenr - 1);
ans_path = g_build_filename(path, "ANS", fl, NULL);
ans_file = g_open(ans_path, O_RDONLY, 0666);
size = bytes_available(ans_file);
if (size > 3) {
char *buf = malloc(size - 2);
lseek(ans_file, 3, SEEK_CUR);
read(ans_file, buf, size - 3);
buf[size -3 ] = '\0';
buffer_add(&mbuf, &mbuf_size, buf);
show_progress(buf);
free(buf);
param_buff[3]++;
}
close(ans_file);
usleep(UEMIS_TIMEOUT);
}
}
if (more_files) {
int size = 0, j = 0;
char *buf = NULL;
if (!ismulti) {
snprintf(fl, 13, "ANS%d.TXT", filenr - 1);
ans_path = g_build_filename(path, "ANS", fl, NULL);
ans_file = g_open(ans_path, O_RDONLY, 0666);
size = bytes_available(ans_file);
if (size > 3) {
buf = malloc(size - 2);
lseek(ans_file, 3, SEEK_CUR);
read(ans_file, buf, size - 3);
buf[size - 3] = '\0';
#if UEMIS_DEBUG > 2
fprintf(debugfile, "::r %s \"%s\"\n", ans_path, buf);
#endif
}
size -= 3;
close(ans_file);
free(ans_path);
} else {
ismulti = FALSE;
}
#if UEMIS_DEBUG > 1
fprintf(debugfile,":r: %s\n", buf);
#endif
if (!answer_in_mbuf)
for (i = 0; i < n_param_out && j < size; i++)
param_buff[i] = next_segment(buf, &j, size);
found_answer = TRUE;
}
#if UEMIS_DEBUG
for (i = 0; i < n_param_out; i++)
fprintf(debugfile,"::: %d: %s\n", i, param_buff[i]);
#endif
return found_answer;
}
/* Turn what we get from the dive computer into something
* that we can pass to the parse_xml function.
* The last 'object_id' that we see is returned as our current
* approximation of a last dive number */
static char *process_raw_buffer(char *inbuf, char **max_divenr)
{
char *buf = strdup(inbuf);
char *tp, *bp, *tag, *type, *val;
gboolean done = FALSE;
int inbuflen = strlen(inbuf);
char *endptr = buf + inbuflen;
char *conv_buffer = NULL;
int conv_buffer_size = 0;
bp = buf + 1;
tp = next_token(&bp);
if (strcmp(tp,"divelog") != 0)
return NULL;
tp = next_token(&bp);
if (strcmp(tp,"1.0") != 0)
return NULL;
buffer_add(&conv_buffer, &conv_buffer_size,
"<dive type=\"uemis\" ref=\"divelog\" version=\"1.0\">\n");
while (!done) {
char *tmp;
int tmp_size;
tag = next_token(&bp);
type = next_token(&bp);
val = next_token(&bp);
if (! strcmp(tag, "object_id")) {
free(*max_divenr);
*max_divenr = strdup(val);
}
if (! strcmp(tag, "file_content")) {
tmp_size = 44 + strlen(tag) + strlen(val);
done = TRUE;
} else {
tmp_size = 27 + strlen(tag) + 2 * strlen(type) + strlen(val);
}
tmp = malloc(tmp_size);
snprintf(tmp, tmp_size,"<val key=\"%s\">\n<%s>%s</%s>\n</val>\n",
tag, type, val, type);
buffer_add(&conv_buffer, &conv_buffer_size, tmp);
free(tmp);
/* done with one dive (got the file_content tag), but there could be more:
* a '{' indicates the end of the record - but we need to see another "{{"
* later in the buffer to know that the next record is complete (it could
* be a short read because of some error */
if (done && ++bp < endptr && *bp != '{' && strstr(bp, "{{")) {
done = FALSE;
buffer_add(&conv_buffer, &conv_buffer_size,
"</dive>\n<dive type=\"uemis\" ref=\"divelog\" version=\"1.0\">\n");
}
}
buffer_add(&conv_buffer, &conv_buffer_size, "</dive>\n");
free(buf);
return strdup(conv_buffer);
}
/* to keep track of multiple computers we simply encode the last dive read
in tuples "{deviceid,nr},{deviceid,nr}..." no spaces to make parsing easier */
static char *find_deviceid(char *max_dive_data, char *deviceid)
{
char *pattern;
char *result;
if (! deviceid || *deviceid == '\0')
return NULL;
pattern = malloc(3 + strlen(deviceid));
sprintf(pattern, "{%s,", deviceid);
result = strstr(max_dive_data, pattern);
free(pattern);
return result;
}
static char *get_divenr(char *max_dive_data, char *deviceid)
{
char *q, *p = max_dive_data;
char *result = NULL;
if (!p || !deviceid)
return strdup("0");
p = find_deviceid(max_dive_data, deviceid);
if (p) {
p += strlen(deviceid) + 2;
q = strchr(p, '}');
if (!q)
return result;
result = malloc(q - p + 1);
strncpy(result, p, q - p);
result[q - p] = '\0';
}
if (!result)
result = strdup("0");
return result;
}
static char *update_max_dive_data(char *max_dive_data, char *deviceid, char *newmax)
{
char *p;
char *result;
int len;
if (! newmax || *newmax == '\0')
return max_dive_data;
p = find_deviceid(max_dive_data, deviceid);
if (p) {
/* if there are more entries after this one, copy them,
otherwise just remove the existing entry for this device */
char *q = strstr(p, "},{");
if (q) {
memcpy(p + 1, q + 3, strlen(q + 3) + 1);
} else {
if (p > max_dive_data)
*(p-1) = '\0';
else
*p = '\0';
}
}
/* now add the new one at the end */
len = strlen(max_dive_data) + strlen(deviceid) + strlen(newmax) + 4 + (strlen(max_dive_data) ? 1 : 0);
result = malloc(len);
snprintf(result, len, "%s%s{%s,%s}", max_dive_data, strlen(max_dive_data) ? "," : "", deviceid, newmax);
free(max_dive_data);
return result;
}
static char *do_uemis_download(struct argument_block *args)
{
const char *mountpath = args->mountpath;
char **max_dive_data = args->max_dive_data;
char **xml_buffer = args->xml_buffer;
int xml_buffer_size;
char *error_text = "";
char *newmax = NULL;
char *deviceid;
char *result = NULL;
char *endptr;
gboolean success;
buffer_add(xml_buffer, &xml_buffer_size, "<dives type='uemis'><string></string>\n<list>\n");
uemis_info("Init Communication");
if (! uemis_init(mountpath))
return "Uemis init failed";
if (! uemis_get_answer(mountpath, "getDeviceId", 0, 1, &error_text))
return error_text;
deviceid = strdup(param_buff[0]);
/* the answer from the DeviceId call becomes the input parameter for getDeviceData */
if (! uemis_get_answer(mountpath, "getDeviceData", 1, 0, &error_text))
return error_text;
/* param_buff[0] is still valid */
if (! uemis_get_answer(mountpath, "initSession", 1, 6, &error_text))
return error_text;
uemis_info("Start download");
if (! uemis_get_answer(mountpath, "processSync", 0, 2, &error_text))
return error_text;
param_buff[1] = "notempty";
newmax = get_divenr(*max_dive_data, deviceid);
for (;;) {
param_buff[2] = newmax;
param_buff[3] = 0;
success = uemis_get_answer(mountpath, "getDivelogs", 3, 0, &error_text);
/* process the buffer we have assembled */
if (mbuf) {
char *next_seg = process_raw_buffer(mbuf, &newmax);
buffer_add(xml_buffer, &xml_buffer_size, next_seg);
}
/* if we got an error, deal with it */
if (!success) {
result = error_text;
break;
}
/* also, if we got nothing back, we should stop trying */
if (!param_buff[3])
break;
/* finally, if the memory is getting too full, maybe we better stop, too */
if (progress_bar_fraction > 0.85) {
result = ERR_FS_ALMOST_FULL;
break;
}
/* clean up mbuf */
endptr = strstr(mbuf, "{{{");
if (endptr)
*(endptr + 2) = '\0';
}
*args->max_dive_data = update_max_dive_data(*max_dive_data, deviceid, newmax);
free(newmax);
if (! uemis_get_answer(mountpath, "terminateSync", 0, 3, &error_text))
return error_text;
if (! strcmp(param_buff[0], "error")) {
if (! strcmp(param_buff[2],"Out of Memory"))
result = ERR_FS_FULL;
else
result = param_buff[2];
}
buffer_add(xml_buffer, &xml_buffer_size, "</list></dives>");
#if UEMIS_DEBUG > 5
fprintf(debugfile, "XML buffer \"%s\"", *xml_buffer);
#endif
return result;
}
static void *pthread_wrapper(void *_data)
{
struct argument_block *args = _data;
const char *err_string = do_uemis_download(args);
import_thread_done = 1;
return (void *)err_string;
}
GError *uemis_download(const char *mountpath, char **max_dive_data, char **xml_buffer, progressbar_t *progress)
{
pthread_t pthread;
void *retval;
struct argument_block args = {mountpath, max_dive_data, xml_buffer, progress};
/* I'm sure there is some better interface for waiting on a thread in a UI main loop */
import_thread_done = 0;
progress_bar_text = "";
progress_bar_fraction = 0.0;
pthread_create(&pthread, NULL, pthread_wrapper, &args);
while (!import_thread_done) {
import_thread_cancelled = process_ui_events();
update_progressbar(args.progress, progress_bar_fraction);
update_progressbar_text(args.progress, progress_bar_text);
usleep(100000);
}
if (pthread_join(pthread, &retval) < 0)
return error("Pthread return with error");
if (retval)
return error(retval);
return NULL;
}