diff --git a/src/readpst.c b/src/readpst.c index 57b1b81..0135d81 100644 --- a/src/readpst.c +++ b/src/readpst.c @@ -1,2491 +1,2492 @@ /*** * readpst.c * Part of the LibPST project * Written by David Smith * dave.s@earthcorp.com */ #include "define.h" #include "lzfu.h" #include "msg.h" #define OUTPUT_TEMPLATE "%s.%s" #define OUTPUT_KMAIL_DIR_TEMPLATE ".%s.directory" #define KMAIL_INDEX "../.%s.index" // max size of the c_time char*. It will store the date of the email #define C_TIME_SIZE 500 struct file_ll { char *name[PST_TYPE_MAX]; char *dname; FILE * output[PST_TYPE_MAX]; int32_t stored_count; int32_t item_count; int32_t skip_count; }; int grim_reaper(); pid_t try_fork(char* folder); void process(pst_item *outeritem, pst_desc_tree *d_ptr); void write_email_body(FILE *f, char *body); void removeCR(char *c); void usage(); void version(); void mk_kmail_dir(char* fname); int close_kmail_dir(); void mk_recurse_dir(char* dir); int close_recurse_dir(); void mk_separate_dir(char *dir); int close_separate_dir(); void mk_separate_file(struct file_ll *f, int32_t t, char *extension, int openit); void close_separate_file(struct file_ll *f); char* my_stristr(char *haystack, char *needle); void check_filename(char *fname); int acceptable_ext(pst_item_attach* attach); void write_separate_attachment(char f_name[], pst_item_attach* attach, int attach_num, pst_file* pst); void write_embedded_message(FILE* f_output, pst_item_attach* attach, char *boundary, pst_file* pf, int save_rtf, char** extra_mime_headers); void write_inline_attachment(FILE* f_output, pst_item_attach* attach, char *boundary, pst_file* pst); int valid_headers(char *header); void header_has_field(char *header, char *field, int *flag); void header_get_subfield(char *field, const char *subfield, char *body_subfield, size_t size_subfield); char* header_get_field(char *header, char *field); char* header_end_field(char *field); void header_strip_field(char *header, char *field); int test_base64(char *body, size_t len); void find_html_charset(char *html, char *charset, size_t charsetlen); void find_rfc822_headers(char** extra_mime_headers); void write_body_part(FILE* f_output, pst_string *body, char *mime, char *charset, char *boundary, pst_file* pst); void write_schedule_part_data(FILE* f_output, pst_item* item, const char* sender, const char* method); void write_schedule_part(FILE* f_output, pst_item* item, const char* sender, const char* boundary); void write_normal_email(FILE* f_output, char f_name[], pst_item* item, int current_mode, int mode_MH, pst_file* pst, int save_rtf, int embedding, char** extra_mime_headers); void write_vcard(FILE* f_output, pst_item *item, pst_item_contact* contact, char comment[]); int write_extra_categories(FILE* f_output, pst_item* item); void write_journal(FILE* f_output, pst_item* item); void write_appointment(FILE* f_output, pst_item *item); void create_enter_dir(struct file_ll* f, pst_item *item); void close_enter_dir(struct file_ll *f); char* quote_string(char *inp); const char* prog_name; char* output_dir = "."; // Normal mode just creates mbox format files in the current directory. Each file is named // the same as the folder's name that it represents #define MODE_NORMAL 0 // KMail mode creates a directory structure suitable for being used directly // by the KMail application #define MODE_KMAIL 1 // recurse mode creates a directory structure like the PST file. Each directory // contains only one file which stores the emails in mboxrd format. #define MODE_RECURSE 2 // separate mode creates the same directory structure as recurse. The emails are stored in // separate files, numbering from 1 upward. Attachments belonging to the emails are // saved as email_no-filename (e.g. 1-samplefile.doc or 1-Attachment2.zip) #define MODE_SEPARATE 3 // Output Normal just prints the standard information about what is going on #define OUTPUT_NORMAL 0 // Output Quiet is provided so that only errors are printed #define OUTPUT_QUIET 1 // default mime-type for attachments that have a null mime-type #define MIME_TYPE_DEFAULT "application/octet-stream" #define RFC822 "message/rfc822" // output mode for contacts #define CMODE_VCARD 0 #define CMODE_LIST 1 // output mode for deleted items #define DMODE_EXCLUDE 0 #define DMODE_INCLUDE 1 // Output type mode flags #define OTMODE_EMAIL 1 #define OTMODE_APPOINTMENT 2 #define OTMODE_JOURNAL 4 #define OTMODE_CONTACT 8 // output settings for RTF bodies // filename for the attachment #define RTF_ATTACH_NAME "rtf-body.rtf" // mime type for the attachment #define RTF_ATTACH_TYPE "application/rtf" // global settings int mode = MODE_NORMAL; int mode_MH = 0; // a submode of MODE_SEPARATE int mode_EX = 0; // a submode of MODE_SEPARATE int mode_MSG = 0; // a submode of MODE_SEPARATE int mode_thunder = 0; // a submode of MODE_RECURSE int output_mode = OUTPUT_NORMAL; int contact_mode = CMODE_VCARD; int deleted_mode = DMODE_EXCLUDE; int output_type_mode = 0xff; // Default to all. int contact_mode_specified = 0; int overwrite = 0; int prefer_utf8 = 0; int save_rtf_body = 1; int file_name_len = 10; // enough room for MODE_SPEARATE file name pst_file pstfile; regex_t meta_charset_pattern; char* default_charset = NULL; char* acceptable_extensions = NULL; int number_processors = 1; // number of cpus we have int max_children = 0; // based on number of cpus and command line args int max_child_specified = 0;// have command line arg -j int active_children; // number of children of this process, cannot be larger than max_children pid_t* child_processes; // setup by main(), and at the start of new child process #ifdef HAVE_SEMAPHORE_H int shared_memory_id; sem_t* global_children = NULL; sem_t* output_mutex = NULL; #endif int grim_reaper(int waitall) { int available = 0; #ifdef HAVE_FORK #ifdef HAVE_SEMAPHORE_H if (global_children) { //sem_getvalue(global_children, &available); //printf("grim reaper %s for pid %d (parent %d) with %d children, %d available\n", (waitall) ? "all" : "", getpid(), getppid(), active_children, available); //fflush(stdout); int i,j; for (i=0; i 0 but active_children == max_children if (available && active_children < max_children) { sem_wait(global_children); pid_t child = fork(); if (child < 0) { // fork failed, pretend it worked and we are the child return 0; } else if (child == 0) { // fork worked, and we are the child, reinitialize *our* list of children active_children = 0; memset(child_processes, 0, sizeof(pid_t) * max_children); pst_reopen(&pstfile); // close and reopen the pst file to get an independent file position pointer } else { // fork worked, and we are the parent, record this child that we need to wait for //pid_t me = getpid(); //printf("parent %d forked child pid %d to process folder %s\n", me, child, folder); //fflush(stdout); child_processes[active_children++] = child; } return child; } else { return 0; // pretend to have forked and we are the child } #endif #endif return 0; } void process(pst_item *outeritem, pst_desc_tree *d_ptr) { struct file_ll ff; pst_item *item = NULL; DEBUG_ENT("process"); create_enter_dir(&ff, outeritem); for (; d_ptr; d_ptr = d_ptr->next) { DEBUG_INFO(("New item record\n")); if (!d_ptr->desc) { ff.skip_count++; DEBUG_WARN(("ERROR item's desc record is NULL\n")); continue; } DEBUG_INFO(("Desc Email ID %#" PRIx64 " [d_ptr->d_id = %#" PRIx64 "]\n", d_ptr->desc->i_id, d_ptr->d_id)); item = pst_parse_item(&pstfile, d_ptr, NULL); DEBUG_INFO(("About to process item\n")); if (!item) { ff.skip_count++; DEBUG_INFO(("A NULL item was seen\n")); continue; } if (item->subject.str) { DEBUG_INFO(("item->subject = %s\n", item->subject.str)); } if (item->folder && item->file_as.str) { DEBUG_INFO(("Processing Folder \"%s\"\n", item->file_as.str)); if (output_mode != OUTPUT_QUIET) { pst_debug_lock(); printf("Processing Folder \"%s\"\n", item->file_as.str); fflush(stdout); pst_debug_unlock(); } ff.item_count++; if (d_ptr->child && (deleted_mode == DMODE_INCLUDE || strcasecmp(item->file_as.str, "Deleted Items"))) { //if this is a non-empty folder other than deleted items, we want to recurse into it pid_t parent = getpid(); pid_t child = try_fork(item->file_as.str); if (child == 0) { // we are the child process, or the original parent if no children were available pid_t me = getpid(); process(item, d_ptr->child); #ifdef HAVE_FORK #ifdef HAVE_SEMAPHORE_H if (me != parent) { // we really were a child, forked for the sole purpose of processing this folder // free my child count slot before really exiting, since // all I am doing here is waiting for my children to exit sem_post(global_children); grim_reaper(1); // wait for all my child processes to exit exit(0); // really exit } #endif #endif } } } else if (item->contact && (item->type == PST_TYPE_CONTACT)) { DEBUG_INFO(("Processing Contact\n")); if (!(output_type_mode & OTMODE_CONTACT)) { ff.skip_count++; DEBUG_INFO(("skipping contact: not in output type list\n")); } else { ff.item_count++; if (mode == MODE_SEPARATE) mk_separate_file(&ff, PST_TYPE_CONTACT, (mode_EX) ? ".vcf" : "", 1); if (contact_mode == CMODE_VCARD) { pst_convert_utf8_null(item, &item->comment); write_vcard(ff.output[PST_TYPE_CONTACT], item, item->contact, item->comment.str); } else { pst_convert_utf8(item, &item->contact->fullname); pst_convert_utf8(item, &item->contact->address1); fprintf(ff.output[PST_TYPE_CONTACT], "%s <%s>\n", item->contact->fullname.str, item->contact->address1.str); } if (mode == MODE_SEPARATE) close_separate_file(&ff); } } else if (item->email && ((item->type == PST_TYPE_NOTE) || (item->type == PST_TYPE_SCHEDULE) || (item->type == PST_TYPE_REPORT))) { DEBUG_INFO(("Processing Email\n")); if (!(output_type_mode & OTMODE_EMAIL)) { ff.skip_count++; DEBUG_INFO(("skipping email: not in output type list\n")); } else { char *extra_mime_headers = NULL; ff.item_count++; if (mode == MODE_SEPARATE) { // process this single email message, possibly forking pid_t parent = getpid(); pid_t child = try_fork(item->file_as.str); if (child == 0) { // we are the child process, or the original parent if no children were available pid_t me = getpid(); mk_separate_file(&ff, PST_TYPE_NOTE, (mode_EX) ? ".eml" : "", 1); write_normal_email(ff.output[PST_TYPE_NOTE], ff.name[PST_TYPE_NOTE], item, mode, mode_MH, &pstfile, save_rtf_body, PST_TYPE_NOTE, &extra_mime_headers); close_separate_file(&ff); if (mode_MSG) { mk_separate_file(&ff, PST_TYPE_NOTE, ".msg", 0); write_msg_email(ff.name[PST_TYPE_NOTE], item, &pstfile); } #ifdef HAVE_FORK #ifdef HAVE_SEMAPHORE_H if (me != parent) { // we really were a child, forked for the sole purpose of processing this message // free my child count slot before really exiting, since // all I am doing here is waiting for my children to exit sem_post(global_children); grim_reaper(1); // wait for all my child processes to exit - there should not be any exit(0); // really exit } #endif #endif } } else { // process this single email message, cannot fork since not separate mode write_normal_email(ff.output[PST_TYPE_NOTE], ff.name[PST_TYPE_NOTE], item, mode, mode_MH, &pstfile, save_rtf_body, 0, &extra_mime_headers); } } } else if (item->journal && (item->type == PST_TYPE_JOURNAL)) { DEBUG_INFO(("Processing Journal Entry\n")); if (!(output_type_mode & OTMODE_JOURNAL)) { ff.skip_count++; DEBUG_INFO(("skipping journal entry: not in output type list\n")); } else { ff.item_count++; if (mode == MODE_SEPARATE) mk_separate_file(&ff, PST_TYPE_JOURNAL, (mode_EX) ? ".ics" : "", 1); write_journal(ff.output[PST_TYPE_JOURNAL], item); fprintf(ff.output[PST_TYPE_JOURNAL], "\n"); if (mode == MODE_SEPARATE) close_separate_file(&ff); } } else if (item->appointment && (item->type == PST_TYPE_APPOINTMENT)) { DEBUG_INFO(("Processing Appointment Entry\n")); if (!(output_type_mode & OTMODE_APPOINTMENT)) { ff.skip_count++; DEBUG_INFO(("skipping appointment: not in output type list\n")); } else { ff.item_count++; if (mode == MODE_SEPARATE) mk_separate_file(&ff, PST_TYPE_APPOINTMENT, (mode_EX) ? ".ics" : "", 1); write_schedule_part_data(ff.output[PST_TYPE_APPOINTMENT], item, NULL, NULL); fprintf(ff.output[PST_TYPE_APPOINTMENT], "\n"); if (mode == MODE_SEPARATE) close_separate_file(&ff); } } else if (item->message_store) { // there should only be one message_store, and we have already done it ff.skip_count++; DEBUG_WARN(("item with message store content, type %i %s, skipping it\n", item->type, item->ascii_type)); } else { ff.skip_count++; DEBUG_WARN(("Unknown item type %i (%s) name (%s)\n", item->type, item->ascii_type, item->file_as.str)); } pst_freeItem(item); } close_enter_dir(&ff); DEBUG_RET(); } int main(int argc, char* const* argv) { pst_item *item = NULL; pst_desc_tree *d_ptr; char * fname = NULL; char *d_log = NULL; int c,x; char *temp = NULL; //temporary char pointer prog_name = argv[0]; if (regcomp(&meta_charset_pattern, "]*content=\"[^>]*charset=([^>\";]*)[\";]", REG_ICASE | REG_EXTENDED)) { printf("cannot compile regex pattern to find content charset in html bodies\n"); exit(3); } // command-line option handling while ((c = getopt(argc, argv, "a:bC:c:Dd:emhj:kMo:qrSt:uVwL:8"))!= -1) { switch (c) { case 'a': if (optarg) { int n = strlen(optarg); acceptable_extensions = (char*)pst_malloc(n+2); strcpy(acceptable_extensions, optarg); acceptable_extensions[n+1] = '\0'; // double null terminates array of non-empty null terminated strings. char *p = acceptable_extensions; while (*p) { if (*p == ',') *p = '\0'; p++; } } break; case 'b': save_rtf_body = 0; break; case 'C': if (optarg) { default_charset = optarg; } else { usage(); exit(0); } break; case 'c': if (optarg && optarg[0]=='v') { contact_mode=CMODE_VCARD; contact_mode_specified = 1; } else if (optarg && optarg[0]=='l') { contact_mode=CMODE_LIST; contact_mode_specified = 1; } else { usage(); exit(0); } break; case 'D': deleted_mode = DMODE_INCLUDE; break; case 'd': d_log = optarg; break; case 'h': usage(); exit(0); break; case 'j': max_children = atoi(optarg); max_child_specified = 1; break; case 'k': mode = MODE_KMAIL; break; case 'M': mode = MODE_SEPARATE; mode_MH = 1; mode_EX = 0; mode_MSG = 0; break; case 'e': mode = MODE_SEPARATE; mode_MH = 1; mode_EX = 1; mode_MSG = 0; file_name_len = 14; break; case 'L': pst_debug_setlevel(atoi(optarg)); break; case 'm': mode = MODE_SEPARATE; mode_MH = 1; mode_EX = 1; mode_MSG = 1; file_name_len = 14; break; case 'o': output_dir = optarg; break; case 'q': output_mode = OUTPUT_QUIET; break; case 'r': mode = MODE_RECURSE; mode_thunder = 0; break; case 'S': mode = MODE_SEPARATE; mode_MH = 0; mode_EX = 0; mode_MSG = 0; break; case 't': // email, appointment, contact, other if (!optarg) { usage(); exit(0); } temp = optarg; output_type_mode = 0; while (*temp > 0) { switch (temp[0]) { case 'e': output_type_mode |= OTMODE_EMAIL; break; case 'a': output_type_mode |= OTMODE_APPOINTMENT; break; case 'j': output_type_mode |= OTMODE_JOURNAL; break; case 'c': output_type_mode |= OTMODE_CONTACT; break; default: usage(); exit(0); break; } temp++; } break; case 'u': mode = MODE_RECURSE; mode_thunder = 1; break; case 'V': version(); exit(0); break; case 'w': overwrite = 1; break; case '8': prefer_utf8 = 1; break; default: usage(); exit(1); break; } } if (argc > optind) { fname = argv[optind]; } else { usage(); exit(2); } #ifdef _SC_NPROCESSORS_ONLN number_processors = sysconf(_SC_NPROCESSORS_ONLN); #endif max_children = (max_child_specified) ? max_children : number_processors * 4; active_children = 0; child_processes = (pid_t *)pst_malloc(sizeof(pid_t) * max_children); memset(child_processes, 0, sizeof(pid_t) * max_children); #if defined(HAVE_SEMAPHORE_H) && defined(HAVE_SYS_IPC_H) && defined(HAVE_SYS_SHM_H) if (max_children) { shared_memory_id = shmget(IPC_PRIVATE, sizeof(sem_t)*2, 0777); if (shared_memory_id >= 0) { global_children = (sem_t *)shmat(shared_memory_id, NULL, 0); if (global_children == (sem_t *)-1) global_children = NULL; if (global_children) { output_mutex = &(global_children[1]); sem_init(global_children, 1, max_children); sem_init(output_mutex, 1, 1); } shmctl(shared_memory_id, IPC_RMID, NULL); } } #endif #ifdef DEBUG_ALL // force a log file if (!d_log) d_log = "readpst.log"; #endif // defined DEBUG_ALL #ifdef HAVE_SEMAPHORE_H DEBUG_INIT(d_log, output_mutex); #else DEBUG_INIT(d_log, NULL); #endif DEBUG_ENT("main"); if (output_mode != OUTPUT_QUIET) printf("Opening PST file and indexes...\n"); RET_DERROR(pst_open(&pstfile, fname, default_charset), 1, ("Error opening File\n")); RET_DERROR(pst_load_index(&pstfile), 2, ("Index Error\n")); pst_load_extended_attributes(&pstfile); if (chdir(output_dir)) { x = errno; pst_close(&pstfile); DEBUG_RET(); DIE(("Cannot change to output dir %s: %s\n", output_dir, strerror(x))); } d_ptr = pstfile.d_head; // first record is main record item = pst_parse_item(&pstfile, d_ptr, NULL); if (!item || !item->message_store) { DEBUG_RET(); DIE(("Could not get root record\n")); } // default the file_as to the same as the main filename if it doesn't exist if (!item->file_as.str) { if (!(temp = strrchr(fname, '/'))) if (!(temp = strrchr(fname, '\\'))) temp = fname; else temp++; // get past the "\\" else temp++; // get past the "/" item->file_as.str = (char*)pst_malloc(strlen(temp)+1); strcpy(item->file_as.str, temp); item->file_as.is_utf8 = 1; DEBUG_INFO(("file_as was blank, so am using %s\n", item->file_as.str)); } DEBUG_INFO(("Root Folder Name: %s\n", item->file_as.str)); d_ptr = pst_getTopOfFolders(&pstfile, item); if (!d_ptr) { DEBUG_RET(); DIE(("Top of folders record not found. Cannot continue\n")); } process(item, d_ptr->child); // do the children of TOPF grim_reaper(1); // wait for all child processes pst_freeItem(item); pst_close(&pstfile); DEBUG_RET(); #ifdef HAVE_SEMAPHORE_H if (global_children) { sem_destroy(global_children); sem_destroy(output_mutex); shmdt(global_children); } #endif + free(child_processes); regfree(&meta_charset_pattern); return 0; } void write_email_body(FILE *f, char *body) { char *n = body; DEBUG_ENT("write_email_body"); if (mode != MODE_SEPARATE) { while (n) { char *p = body; while (*p == '>') p++; if (strncmp(p, "From ", 5) == 0) fprintf(f, ">"); if ((n = strchr(body, '\n'))) { n++; pst_fwrite(body, n-body, 1, f); //write just a line body = n; } } } pst_fwrite(body, strlen(body), 1, f); DEBUG_RET(); } void removeCR (char *c) { // converts \r\n to \n char *a, *b; DEBUG_ENT("removeCR"); a = b = c; while (*a != '\0') { *b = *a; if (*a != '\r') b++; a++; } *b = '\0'; DEBUG_RET(); } void usage() { DEBUG_ENT("usage"); version(); printf("Usage: %s [OPTIONS] {PST FILENAME}\n", prog_name); printf("OPTIONS:\n"); printf("\t-V\t- Version. Display program version\n"); printf("\t-C charset\t- character set for items with an unspecified character set\n"); printf("\t-D\t- Include deleted items in output\n"); printf("\t-L \t- Set debug level; 1=debug,2=info,3=warn.\n"); printf("\t-M\t- Write emails in the MH (rfc822) format\n"); printf("\t-S\t- Separate. Write emails in the separate format\n"); printf("\t-a \t- Discard any attachment without an extension on the list\n"); printf("\t-b\t- Don't save RTF-Body attachments\n"); printf("\t-c[v|l]\t- Set the Contact output mode. -cv = VCard, -cl = EMail list\n"); printf("\t-d \t- Debug to file.\n"); printf("\t-e\t- As with -M, but include extensions on output files\n"); printf("\t-h\t- Help. This screen\n"); printf("\t-j \t- Number of parallel jobs to run\n"); printf("\t-k\t- KMail. Output in kmail format\n"); printf("\t-m\t- As with -e, but write .msg files also\n"); printf("\t-o \t- Output directory to write files to. CWD is changed *after* opening pst file\n"); printf("\t-q\t- Quiet. Only print error messages\n"); printf("\t-r\t- Recursive. Output in a recursive format\n"); printf("\t-t[eajc]\t- Set the output type list. e = email, a = attachment, j = journal, c = contact\n"); printf("\t-u\t- Thunderbird mode. Write two extra .size and .type files\n"); printf("\t-w\t- Overwrite any output mbox files\n"); printf("\t-8\t- Output bodies in UTF-8, rather than original encoding, if UTF-8 version is available\n"); printf("\n"); printf("Only one of -M -S -e -k -m -r should be specified\n"); DEBUG_RET(); } void version() { DEBUG_ENT("version"); printf("ReadPST / LibPST v%s\n", VERSION); #if BYTE_ORDER == BIG_ENDIAN printf("Big Endian implementation being used.\n"); #elif BYTE_ORDER == LITTLE_ENDIAN printf("Little Endian implementation being used.\n"); #else # error "Byte order not supported by this library" #endif DEBUG_RET(); } void mk_kmail_dir(char *fname) { //make a directory based on OUTPUT_KMAIL_DIR_TEMPLATE //change to that directory char *dir, *index; int x; DEBUG_ENT("mk_kmail_dir"); dir = pst_malloc(strlen(fname)+strlen(OUTPUT_KMAIL_DIR_TEMPLATE)+1); sprintf(dir, OUTPUT_KMAIL_DIR_TEMPLATE, fname); check_filename(dir); if (D_MKDIR(dir)) { if (errno != EEXIST) { // not an error because it exists x = errno; DIE(("mk_kmail_dir: Cannot create directory %s: %s\n", dir, strerror(x))); } } if (chdir(dir)) { x = errno; DIE(("mk_kmail_dir: Cannot change to directory %s: %s\n", dir, strerror(x))); } free (dir); //we should remove any existing indexes created by KMail, cause they might be different now index = pst_malloc(strlen(fname)+strlen(KMAIL_INDEX)+1); sprintf(index, KMAIL_INDEX, fname); unlink(index); free(index); DEBUG_RET(); } int close_kmail_dir() { int x; DEBUG_ENT("close_kmail_dir"); if (chdir("..")) { x = errno; DIE(("close_kmail_dir: Cannot move up dir (..): %s\n", strerror(x))); } DEBUG_RET(); return 0; } char *item_type_to_name(int32_t item_type) { char *name; switch (item_type) { case PST_TYPE_APPOINTMENT: name = "calendar"; break; case PST_TYPE_CONTACT: name = "contacts"; break; case PST_TYPE_JOURNAL: name = "journal"; break; case PST_TYPE_STICKYNOTE: case PST_TYPE_TASK: case PST_TYPE_NOTE: case PST_TYPE_OTHER: case PST_TYPE_REPORT: default: name = "mbox"; break; } return name; } int32_t reduced_item_type(int32_t item_type) { int32_t reduced; switch (item_type) { case PST_TYPE_APPOINTMENT: case PST_TYPE_CONTACT: case PST_TYPE_JOURNAL: reduced = item_type; break; case PST_TYPE_STICKYNOTE: case PST_TYPE_TASK: case PST_TYPE_NOTE: case PST_TYPE_OTHER: case PST_TYPE_REPORT: default: reduced = PST_TYPE_NOTE; break; } return reduced; } // this will create a directory by that name void mk_recurse_dir(char *dir) { int x; DEBUG_ENT("mk_recurse_dir"); check_filename(dir); if (D_MKDIR (dir)) { if (errno != EEXIST) { // not an error because it exists x = errno; DIE(("mk_recurse_dir: Cannot create directory %s: %s\n", dir, strerror(x))); } } if (chdir(dir)) { x = errno; DIE(("mk_recurse_dir: Cannot change to directory %s: %s\n", dir, strerror(x))); } DEBUG_RET(); } int close_recurse_dir() { int x; DEBUG_ENT("close_recurse_dir"); if (chdir("..")) { x = errno; DIE(("close_recurse_dir: Cannot go up dir (..): %s\n", strerror(x))); } DEBUG_RET(); return 0; } void mk_separate_dir(char *dir) { size_t dirsize = strlen(dir) + 10; char dir_name[dirsize]; int x = 0, y = 0; DEBUG_ENT("mk_separate_dir"); do { if (y == 0) snprintf(dir_name, dirsize, "%s", dir); else snprintf(dir_name, dirsize, "%s%i%s", dir, y, ""); // enough for 9 digits allocated above check_filename(dir_name); DEBUG_INFO(("about to try creating %s\n", dir_name)); if (D_MKDIR(dir_name)) { if (errno != EEXIST) { // if there is an error, and it doesn't already exist x = errno; DIE(("mk_separate_dir: Cannot create directory %s: %s\n", dir, strerror(x))); } } else { break; } y++; } while (overwrite == 0); if (chdir(dir_name)) { x = errno; DIE(("mk_separate_dir: Cannot change to directory %s: %s\n", dir, strerror(x))); } if (overwrite) { // we should probably delete all files from this directory #if !defined(WIN32) && !defined(__CYGWIN__) DIR * sdir = NULL; struct dirent *dirent = NULL; struct stat filestat; if (!(sdir = opendir("./"))) { DEBUG_WARN(("mk_separate_dir: Cannot open dir \"%s\" for deletion of old contents\n", "./")); } else { while ((dirent = readdir(sdir))) { if (lstat(dirent->d_name, &filestat) != -1) if (S_ISREG(filestat.st_mode)) { if (unlink(dirent->d_name)) { y = errno; DIE(("mk_separate_dir: unlink returned error on file %s: %s\n", dirent->d_name, strerror(y))); } } } closedir(sdir); // cppcheck detected leak } #endif } DEBUG_RET(); } int close_separate_dir() { int x; DEBUG_ENT("close_separate_dir"); if (chdir("..")) { x = errno; DIE(("close_separate_dir: Cannot go up dir (..): %s\n", strerror(x))); } DEBUG_RET(); return 0; } void mk_separate_file(struct file_ll *f, int32_t t, char *extension, int openit) { DEBUG_ENT("mk_separate_file"); DEBUG_INFO(("opening next file to save email type %s\n", item_type_to_name(t))); if (f->item_count > 999999999) { // bigger than nine 9's DIE(("mk_separate_file: The number of emails in this folder has become too high to handle\n")); } sprintf(f->name[t], "%" PRIi32 "%s", f->item_count, extension); check_filename(f->name[t]); if (openit) { if (!(f->output[t] = fopen(f->name[t], "w"))) { DIE(("mk_separate_file: Cannot open file to save email \"%s\"\n", f->name[t])); } } DEBUG_RET(); } void close_separate_file(struct file_ll *f) { int32_t t; DEBUG_ENT("close_separate_file"); for (t=0; toutput[t]) { struct stat st; fclose(f->output[t]); stat(f->name[t], &st); if (!st.st_size) { DEBUG_WARN(("removing empty output file %s\n", f->name[t])); remove(f->name[t]); } f->output[t] = NULL; } } DEBUG_RET(); } char *my_stristr(char *haystack, char *needle) { // my_stristr varies from strstr in that its searches are case-insensitive char *x=haystack, *y=needle, *z = NULL; if (!haystack || !needle) { return NULL; } while (*y != '\0' && *x != '\0') { if (tolower(*y) == tolower(*x)) { // move y on one y++; if (!z) { z = x; // store first position in haystack where a match is made } } else { y = needle; // reset y to the beginning of the needle z = NULL; // reset the haystack storage point } x++; // advance the search in the haystack } // If the haystack ended before our search finished, it's not a match. if (*y != '\0') return NULL; return z; } void check_filename(char *fname) { char *t = fname; DEBUG_ENT("check_filename"); if (!t) { DEBUG_RET(); return; } while ((t = strpbrk(t, "/\\:"))) { // while there are characters in the second string that we don't want *t = '_'; //replace them with an underscore } DEBUG_RET(); } /** * check if the file name extension is acceptable. If not, the attachment * will be discarded * @param attach pst attachment object * @return true if the attachment filename contains an extension that we want. */ int acceptable_ext(pst_item_attach* attach) { if (!acceptable_extensions || *acceptable_extensions == '\0') return 1; // acceptable list missing or empty char *attach_filename = (attach->filename2.str) ? attach->filename2.str : attach->filename1.str; if (!attach_filename) return 1; // attachment with no name is always acceptable char *e = strrchr(attach_filename, '.'); if (!e) return 1; // attachment with no extension is always acceptable. DEBUG_ENT("acceptable_ext"); DEBUG_INFO(("attachment extension %s\n", e)); int rc = 0; char *a = acceptable_extensions; while (*a) { if (pst_stricmp(a, e) == 0) { rc = 1; break; } a += strlen(a) + 1; } DEBUG_INFO(("attachment acceptable returns %d\n", rc)); DEBUG_RET(); return rc; } void write_separate_attachment(char f_name[], pst_item_attach* attach, int attach_num, pst_file* pst) { FILE *fp = NULL; int x = 0; char *temp = NULL; // If there is a long filename (filename2) use that, otherwise // use the 8.3 filename (filename1) char *attach_filename = (attach->filename2.str) ? attach->filename2.str : attach->filename1.str; DEBUG_ENT("write_separate_attachment"); DEBUG_INFO(("Attachment %s Size is %#zx, data = %p, id %#" PRIx64 "\n", attach_filename, attach->data.size, (void*)attach->data.data, attach->i_id)); if (!attach->data.data) { // make sure we can fetch data from the id pst_index_ll *ptr = pst_getID(pst, attach->i_id); if (!ptr) { DEBUG_WARN(("Couldn't find i_id %#" PRIx64 ". Cannot save attachment to file\n", attach->i_id)); DEBUG_RET(); return; } } check_filename(f_name); if (!attach_filename) { // generate our own (dummy) filename for the attachment temp = pst_malloc(strlen(f_name)+15); sprintf(temp, "%s-attach%i", f_name, attach_num); } else { // have an attachment name, make sure it's unique temp = pst_malloc(strlen(f_name)+strlen(attach_filename)+15); do { if (fp) fclose(fp); if (x == 0) sprintf(temp, "%s-%s", f_name, attach_filename); else sprintf(temp, "%s-%s-%i", f_name, attach_filename, x); } while ((fp = fopen(temp, "r")) && ++x < 99999999); if (x > 99999999) { DIE(("error finding attachment name. exhausted possibilities to %s\n", temp)); } } DEBUG_INFO(("Saving attachment to %s\n", temp)); if (!(fp = fopen(temp, "w"))) { DEBUG_WARN(("write_separate_attachment: Cannot open attachment save file \"%s\"\n", temp)); } else { (void)pst_attach_to_file(pst, attach, fp); fclose(fp); } if (temp) free(temp); DEBUG_RET(); } void write_embedded_message(FILE* f_output, pst_item_attach* attach, char *boundary, pst_file* pf, int save_rtf, char** extra_mime_headers) { pst_index_ll *ptr; DEBUG_ENT("write_embedded_message"); ptr = pst_getID(pf, attach->i_id); pst_desc_tree d_ptr; d_ptr.d_id = 0; d_ptr.parent_d_id = 0; d_ptr.assoc_tree = NULL; d_ptr.desc = ptr; d_ptr.no_child = 0; d_ptr.prev = NULL; d_ptr.next = NULL; d_ptr.parent = NULL; d_ptr.child = NULL; d_ptr.child_tail = NULL; pst_item *item = pst_parse_item(pf, &d_ptr, attach->id2_head); // It appears that if the embedded message contains an appointment/ // calendar item, pst_parse_item returns NULL due to the presence of // an unexpected reference type of 0x1048, which seems to represent // an array of GUIDs representing a CLSID. It's likely that this is // a reference to an internal Outlook COM class. // Log the skipped item and continue on. if (!item) { DEBUG_WARN(("write_embedded_message: pst_parse_item was unable to parse the embedded message in attachment ID %" PRIu64, attach->i_id)); } else { if (!item->email) { DEBUG_WARN(("write_embedded_message: pst_parse_item returned type %d, not an email message", item->type)); } else { fprintf(f_output, "\n--%s\n", boundary); fprintf(f_output, "Content-Type: %s\n\n", attach->mimetype.str); write_normal_email(f_output, "", item, MODE_NORMAL, 0, pf, save_rtf, 1, extra_mime_headers); } pst_freeItem(item); } DEBUG_RET(); } /** * Backslash-escape quotes and backslashes in the given string. */ char *quote_string(char *inp) { int i = 0; int count = 0; char *curr = inp; while (*curr) { if (*curr == '\"' || *curr == '\\') { count++; } curr++; i++; } char *res = malloc(i + count + 1); char *curr_in = inp; char *curr_out = res; while (*curr_in) { if (*curr_in == '\"' || *curr_in == '\\') { *curr_out++ = '\\'; } *curr_out++ = *curr_in++; } *curr_out = '\0'; return res; } /* * RFC 2231 says: * attribute-char := * Equivalent, but clearer from RFC 5987: * attr-char = ALPHA / DIGIT * / "!" / "#" / "$" / "&" / "+" / "-" / "." * / "^" / "_" / "`" / "|" / "~" */ static int is_attribute_char(int8_t x) { if ((x >= 'A' && x <= 'Z') || (x >= 'a' && x <= 'z') || (x >= '0' && x <= '9')) { return 1; } switch(x) { case '!': case '#': case '$': case '&': case '+': case '-': case '.': case '^': case '_': case '`': case '|': case '~': return 1; } return 0; } /** Convert inp to rfc2231 encoding of string * * @param inp pointer to the string of interest * @return pointer to converted string -- caller must free */ char *rfc2231_string(char *inp) { int needs = 0; const int8_t *x = (int8_t *)inp; while (*x) { if (!is_attribute_char(*x)) needs++; x++; } int n = strlen(inp) + 2*needs + 15; char *buffer = pst_malloc(n); strcpy(buffer, "utf-8''"); x = (int8_t *)inp; const uint8_t *y = (uint8_t *)inp; char *z = buffer; z += strlen(buffer); // skip the utf8 prefix while (*y) { if (!is_attribute_char(*x)) { *(z++) = '%'; snprintf(z, 3, "%2" PRIx8, *y); z += 2; } else { *(z++) = *y; } x++; y++; } *z = '\0'; return buffer; } void write_inline_attachment(FILE* f_output, pst_item_attach* attach, char *boundary, pst_file* pst) { DEBUG_ENT("write_inline_attachment"); DEBUG_INFO(("Attachment Size is %#zx, data = %p, id %#" PRIx64 "\n", attach->data.size, (void*)attach->data.data, attach->i_id)); if (!attach->data.data) { // make sure we can fetch data from the id pst_index_ll *ptr = pst_getID(pst, attach->i_id); if (!ptr) { DEBUG_WARN(("Couldn't find ID pointer. Cannot save attachment to file\n")); DEBUG_RET(); return; } } fprintf(f_output, "\n--%s\n", boundary); if (!attach->mimetype.str) { fprintf(f_output, "Content-Type: %s\n", MIME_TYPE_DEFAULT); } else { fprintf(f_output, "Content-Type: %s\n", attach->mimetype.str); } fprintf(f_output, "Content-Transfer-Encoding: base64\n"); if (attach->content_id.str) { fprintf(f_output, "Content-ID: <%s>\n", attach->content_id.str); } if (attach->filename2.str) { // use the long filename, converted to proper encoding if needed. // it is already utf8 char *escaped = quote_string(attach->filename2.str); // encode long filename as rfc2231 without modifying original -- we may still need the original long filename char *rfc2231 = rfc2231_string(attach->filename2.str); fprintf(f_output, "Content-Disposition: attachment; \n filename*=%s;\n", rfc2231); free (rfc2231); // Also include the (escaped) utf8 filename in the 'filename' header directly - this is not strictly valid // (since this header should be ASCII) but is almost always handled correctly (and in fact this is the only // way to get MS Outlook to correctly read a UTF8 filename, AFAICT, which is why we're doing it). fprintf(f_output, " filename=\"%s\"\n\n", escaped); free(escaped); } else if (attach->filename1.str) { // short filename never needs encoding fprintf(f_output, "Content-Disposition: attachment; filename=\"%s\"\n\n", attach->filename1.str); } else { // no filename is inline fprintf(f_output, "Content-Disposition: inline\n\n"); } (void)pst_attach_to_file_base64(pst, attach, f_output); fprintf(f_output, "\n\n"); DEBUG_RET(); } int header_match(char *header, char*field) { int n = strlen(field); if (strncasecmp(header, field, n) == 0) return 1; // tag:{space} if ((field[n-1] == ' ') && (strncasecmp(header, field, n-1) == 0)) { char *crlftab = "\r\n\t"; char *crlfspc = "\r\n "; DEBUG_INFO(("Possible wrapped header = %s\n", header)); if (strncasecmp(header+n-1, crlftab, 3) == 0) return 1; // tag:{cr}{lf}{tab} if (strncasecmp(header+n-1, crlfspc, 3) == 0) return 1; // tag:{cr}{lf}{space} } return 0; } // https://en.wikipedia.org/wiki/Email#Message_header // https://www.rfc-editor.org/rfc/rfc5322.html // https://www.iana.org/assignments/message-headers/message-headers.xhtml int header_is_reasonable(char *header) { char *c; #define C *c // The header must not be NULL if (header) c = header; else return 0; // usually the header field name starts with upper-case: A-Z if (C >= 'A' && C <= 'Z') c++; else return 0; while(1) { // most header field names use a limited set of characters: - 0-9 A-Z a-z if ( (C >= 'A' && C <= 'Z') || (C >= 'a' && C <= 'z') || (C >= '0' && C <= '9') || (C == '-') ) { c++; // the header field name is then terminated with a colon } else if (C == ':') { c++; goto parse_header_field_value; // other characters are an indicator of an invalid header } else { return 0; } } parse_header_field_value: while(1) { // header field values are printable US-ASCII plus space/tab if ( (C >= 33 && C <= 126) || (C == ' ' || C == '\t') ) { c++; // the header field value is then terminated with CRLF } else if (C == '\r' && *(c+1) == '\n') { c += 2; // the value could continue to the next line though if (C == ' ' || C == '\t') c++; else return 1; // other characters are an indicator of an invalid header } else { return 0; } } #undef C } int valid_headers(char *header) { // headers are sometimes really bogus - they seem to be fragments of the // message body, so we only use them if they seem to be real rfc822 headers. // this list is composed of ones that we have seen in real pst files. // there are surely others. the problem is - given an arbitrary character // string, is it a valid (or even reasonable) set of rfc822 headers? if (header) { if (header_match(header, "Content-Language: " )) return 1; if (header_match(header, "Content-Type: " )) return 1; if (header_match(header, "DKIM-Signature: " )) return 1; if (header_match(header, "Date: " )) return 1; if (header_match(header, "Delivered-To: " )) return 1; if (header_match(header, "From: " )) return 1; if (header_match(header, "MIME-Version: " )) return 1; if (header_match(header, "Microsoft Mail Internet Headers")) return 1; if (header_match(header, "Received: " )) return 1; if (header_match(header, "Return-Path: " )) return 1; if (header_match(header, "Subject: " )) return 1; if (header_match(header, "To: " )) return 1; if (header_match(header, "User-Agent: " )) return 1; if (header_match(header, "X-ASG-Debug-ID: " )) return 1; if (header_match(header, "X-Barracuda-URL: " )) return 1; if (header_match(header, "X-x: " )) return 1; if (header_is_reasonable(header)) return 1; if (strlen(header) > 2) { DEBUG_INFO(("Ignore bogus headers = %s\n", header)); } return 0; } else return 0; } void header_has_field(char *header, char *field, int *flag) { DEBUG_ENT("header_has_field"); if (my_stristr(header, field) || (strncasecmp(header, field+1, strlen(field)-1) == 0)) { DEBUG_INFO(("header block has %s header\n", field+1)); *flag = 1; } DEBUG_RET(); } void header_get_subfield(char *field, const char *subfield, char *body_subfield, size_t size_subfield) { if (!field) return; DEBUG_ENT("header_get_subfield"); char search[60]; snprintf(search, sizeof(search), " %s=", subfield); field++; char *n = header_end_field(field); char *s = my_stristr(field, search); if (n && s && (s < n)) { char *e, *f, save; s += strlen(search); // skip over subfield= if (*s == '"') { s++; e = strchr(s, '"'); } else { e = strchr(s, ';'); f = strchr(s, '\n'); if (e && f && (f < e)) e = f; } if (!e || (e > n)) e = n; // use the trailing lf as terminator if nothing better save = *e; *e = '\0'; snprintf(body_subfield, size_subfield, "%s", s); // copy the subfield to our buffer *e = save; DEBUG_INFO(("body %s %s from headers\n", subfield, body_subfield)); } DEBUG_RET(); } char* header_get_field(char *header, char *field) { char *t = my_stristr(header, field); if (!t && (strncasecmp(header, field+1, strlen(field)-1) == 0)) t = header; return t; } // return pointer to \n at the end of this header field, // or NULL if this field goes to the end of the string. char *header_end_field(char *field) { char *e = strchr(field+1, '\n'); while (e && ((e[1] == ' ') || (e[1] == '\t'))) { e = strchr(e+1, '\n'); } return e; } void header_strip_field(char *header, char *field) { char *t; while ((t = header_get_field(header, field))) { char *e = header_end_field(t); if (e) { if (t == header) e++; // if *t is not \n, we don't want to keep the \n at *e either. while (*e != '\0') { *t = *e; t++; e++; } *t = '\0'; } else { // this was the last header field, truncate the headers *t = '\0'; } } } int test_base64(char *body, size_t len) { int b64 = 0; uint8_t *b = (uint8_t *)body; DEBUG_ENT("test_base64"); while (len--) { if ((*b < 32) && (*b != 9) && (*b != 10)) { DEBUG_INFO(("found base64 byte %" PRId8 "\n", *b)); DEBUG_HEXDUMPC(body, strlen(body), 0x10); b64 = 1; break; } b++; } DEBUG_RET(); return b64; } void find_html_charset(char *html, char *charset, size_t charsetlen) { const int index = 1; const int nmatch = index+1; regmatch_t match[nmatch]; DEBUG_ENT("find_html_charset"); int rc = regexec(&meta_charset_pattern, html, nmatch, match, 0); if (rc == 0) { int s = match[index].rm_so; int e = match[index].rm_eo; if (s != -1) { char save = html[e]; html[e] = '\0'; snprintf(charset, charsetlen, "%s", html+s); // copy the html charset html[e] = save; DEBUG_INFO(("charset %s from html text\n", charset)); } else { DEBUG_INFO(("matching %d %d %d %d\n", match[0].rm_so, match[0].rm_eo, match[1].rm_so, match[1].rm_eo)); DEBUG_HEXDUMPC(html, strlen(html), 0x10); } } else { DEBUG_INFO(("regexec returns %d\n", rc)); } DEBUG_RET(); } void find_rfc822_headers(char** extra_mime_headers) { DEBUG_ENT("find_rfc822_headers"); char *headers = *extra_mime_headers; if (headers) { char *temp, *t; while ((temp = strstr(headers, "\n\n"))) { temp[1] = '\0'; t = header_get_field(headers, "\nContent-Type:"); if (t) { t++; DEBUG_INFO(("found content type header\n")); char *n = strchr(t, '\n'); char *s = strstr(t, ": "); char *e = strchr(t, ';'); if (!e || (e > n)) e = n; if (s && (s < e)) { s += 2; if (!strncasecmp(s, RFC822, e-s)) { headers = temp+2; // found rfc822 header DEBUG_INFO(("found 822 headers\n%s\n", headers)); break; } } } //DEBUG_INFO(("skipping to next block after\n%s\n", headers)); headers = temp+2; // skip to next chunk of headers } *extra_mime_headers = headers; } DEBUG_RET(); } void write_body_part(FILE* f_output, pst_string *body, char *mime, char *charset, char *boundary, pst_file* pst) { DEBUG_ENT("write_body_part"); removeCR(body->str); size_t body_len = strlen(body->str); if (body->is_utf8 && (strcasecmp("utf-8", charset))) { if (prefer_utf8) { charset = "utf-8"; } else { // try to convert to the specified charset since the target // is not utf-8, and the data came from a unicode (utf16) field // and is now in utf-8. size_t rc; DEBUG_INFO(("Convert %s utf-8 to %s\n", mime, charset)); pst_vbuf *newer = pst_vballoc(2); rc = pst_vb_utf8to8bit(newer, body->str, body_len, charset); if (rc == (size_t)-1) { // unable to convert, change the charset to utf8 free(newer->b); DEBUG_INFO(("Failed to convert %s utf-8 to %s\n", mime, charset)); charset = "utf-8"; } else { // null terminate the output string pst_vbgrow(newer, 1); newer->b[newer->dlen] = '\0'; free(body->str); body->str = newer->b; body_len = newer->dlen; } free(newer); } } int base64 = test_base64(body->str, body_len); fprintf(f_output, "\n--%s\n", boundary); fprintf(f_output, "Content-Type: %s; charset=\"%s\"\n", mime, charset); if (base64) fprintf(f_output, "Content-Transfer-Encoding: base64\n"); fprintf(f_output, "\n"); // Any body that uses an encoding with NULLs, e.g. UTF16, will be base64-encoded here. if (base64) { char *enc = pst_base64_encode(body->str, body_len); if (enc) { write_email_body(f_output, enc); fprintf(f_output, "\n"); free(enc); } } else { write_email_body(f_output, body->str); } DEBUG_RET(); } void write_schedule_part_data(FILE* f_output, pst_item* item, const char* sender, const char* method) { fprintf(f_output, "BEGIN:VCALENDAR\n"); fprintf(f_output, "VERSION:2.0\n"); fprintf(f_output, "PRODID:LibPST v%s\n", VERSION); if (method) fprintf(f_output, "METHOD:%s\n", method); fprintf(f_output, "BEGIN:VEVENT\n"); if (sender) { if (item->email->outlook_sender_name.str) { fprintf(f_output, "ORGANIZER;CN=\"%s\":MAILTO:%s\n", item->email->outlook_sender_name.str, sender); } else { fprintf(f_output, "ORGANIZER;CN=\"\":MAILTO:%s\n", sender); } } write_appointment(f_output, item); fprintf(f_output, "END:VCALENDAR\n"); } void write_schedule_part(FILE* f_output, pst_item* item, const char* sender, const char* boundary) { const char* method = "REQUEST"; const char* charset = "utf-8"; char fname[30]; if (!item->appointment) return; // inline appointment request fprintf(f_output, "\n--%s\n", boundary); fprintf(f_output, "Content-Type: %s; method=\"%s\"; charset=\"%s\"\n\n", "text/calendar", method, charset); write_schedule_part_data(f_output, item, sender, method); fprintf(f_output, "\n"); // attachment appointment request snprintf(fname, sizeof(fname), "i%" PRIu64 ".ics", item->block_id); fprintf(f_output, "\n--%s\n", boundary); fprintf(f_output, "Content-Type: %s; charset=\"%s\"; name=\"%s\"\n", "text/calendar", "utf-8", fname); fprintf(f_output, "Content-Disposition: attachment; filename=\"%s\"\n\n", fname); write_schedule_part_data(f_output, item, sender, method); fprintf(f_output, "\n"); } void write_normal_email(FILE* f_output, char f_name[], pst_item* item, int current_mode, int mode_MH, pst_file* pst, int save_rtf, int embedding, char** extra_mime_headers) { char boundary[60]; char altboundary[66]; char *altboundaryp = NULL; char body_charset[30]; char buffer_charset[30]; char body_report[60]; char sender[60]; int sender_known = 0; char *temp = NULL; time_t em_time; char *c_time = NULL; char *headers = NULL; int has_from, has_subject, has_to, has_cc, has_date, has_msgid; has_from = has_subject = has_to = has_cc = has_date = has_msgid = 0; DEBUG_ENT("write_normal_email"); pst_convert_utf8_null(item, &item->email->header); DEBUG_INFO(("PST headers\n%s\n", item->email->header.str)); DEBUG_INFO(("Extra MIME headers\n%s\n", *extra_mime_headers)); headers = valid_headers(item->email->header.str) ? item->email->header.str : valid_headers(*extra_mime_headers) ? *extra_mime_headers : NULL; // setup default body character set and report type strncpy(body_charset, pst_default_charset(item, sizeof(buffer_charset), buffer_charset), sizeof(body_charset)); body_charset[sizeof(body_charset)-1] = '\0'; strncpy(body_report, "delivery-status", sizeof(body_report)); body_report[sizeof(body_report)-1] = '\0'; // setup default sender pst_convert_utf8(item, &item->email->sender_address); if (item->email->sender_address.str && strchr(item->email->sender_address.str, '@')) { temp = item->email->sender_address.str; sender_known = 1; } else { temp = "MAILER-DAEMON"; } strncpy(sender, temp, sizeof(sender)); sender[sizeof(sender)-1] = '\0'; // convert the sent date if it exists, or set it to a fixed date if (item->email->sent_date) { em_time = pst_fileTimeToUnixTime(item->email->sent_date); c_time = ctime(&em_time); if (c_time) c_time[strlen(c_time)-1] = '\0'; //remove end \n } if (!c_time) c_time = "Thu Jan 1 00:00:00 1970"; // create our MIME boundaries here. snprintf(boundary, sizeof(boundary), "--boundary-LibPST-iamunique-%" PRIu64 "_-_-", item->block_id); snprintf(altboundary, sizeof(altboundary), "alt-%s", boundary); // we will always look at the headers to discover some stuff if (headers ) { char *t; removeCR(headers); temp = strstr(headers, "\n\n"); if (temp) { // cut off our real rfc822 headers here temp[1] = '\0'; // pointer to all the embedded MIME headers. // we use these to find the actual rfc822 headers for embedded message/rfc822 mime parts // but only for the outermost message if (!*extra_mime_headers) *extra_mime_headers = temp+2; DEBUG_INFO(("Found extra mime headers\n%s\n", temp+2)); } // Check if the headers have all the necessary fields header_has_field(headers, "\nFrom:", &has_from); header_has_field(headers, "\nTo:", &has_to); header_has_field(headers, "\nSubject:", &has_subject); header_has_field(headers, "\nDate:", &has_date); header_has_field(headers, "\nCC:", &has_cc); header_has_field(headers, "\nMessage-Id:", &has_msgid); // look for charset and report-type in Content-Type header t = header_get_field(headers, "\nContent-Type:"); header_get_subfield(t, "charset", body_charset, sizeof(body_charset)); header_get_subfield(t, "report-type", body_report, sizeof(body_report)); // derive a proper sender email address if (!sender_known) { t = header_get_field(headers, "\nFrom:"); if (t) { // assume address is on the first line, rather than on a continuation line t++; char *n = strchr(t, '\n'); char *s = strchr(t, '<'); char *e = strchr(t, '>'); if (s && e && n && (s < e) && (e < n)) { char save = *e; *e = '\0'; snprintf(sender, sizeof(sender), "%s", s+1); *e = save; } } } // Strip out the mime headers and some others that we don't want to emit header_strip_field(headers, "\nMicrosoft Mail Internet Headers"); header_strip_field(headers, "\nMIME-Version:"); header_strip_field(headers, "\nContent-Type:"); header_strip_field(headers, "\nContent-Transfer-Encoding:"); header_strip_field(headers, "\nContent-class:"); header_strip_field(headers, "\nX-MimeOLE:"); header_strip_field(headers, "\nX-From_:"); } DEBUG_INFO(("About to print Header\n")); if (item && item->subject.str) { pst_convert_utf8(item, &item->subject); DEBUG_INFO(("item->subject = %s\n", item->subject.str)); } if (mode != MODE_SEPARATE) { // most modes need this separator line. // procmail produces this separator without the quotes around the // sender email address, but apparently some Mac email client needs // those quotes, and they don't seem to cause problems for anyone else. char *quo = (embedding) ? ">" : ""; fprintf(f_output, "%sFrom \"%s\" %s\n", quo, sender, c_time); } // print the supplied email headers if (headers) { int len = strlen(headers); if (len > 0) { fprintf(f_output, "%s", headers); // make sure the headers end with a \n if (headers[len-1] != '\n') fprintf(f_output, "\n"); //char *h = headers; //while (*h) { // char *e = strchr(h, '\n'); // int d = 1; // normally e points to trailing \n // if (!e) { // e = h + strlen(h); // e points to trailing null // d = 0; // } // // we could do rfc2047 encoding here if needed // fprintf(f_output, "%.*s\n", (int)(e-h), h); // h = e + d; //} } } // record read status if ((item->flags & PST_FLAG_READ) == PST_FLAG_READ) { fprintf(f_output, "Status: RO\n"); } // create required header fields that are not already written if (!has_from) { if (item->email->outlook_sender_name.str){ pst_rfc2047(item, &item->email->outlook_sender_name, 1); fprintf(f_output, "From: %s <%s>\n", item->email->outlook_sender_name.str, sender); } else { fprintf(f_output, "From: <%s>\n", sender); } } if (!has_subject) { if (item->subject.str) { pst_rfc2047(item, &item->subject, 0); fprintf(f_output, "Subject: %s\n", item->subject.str); } else { fprintf(f_output, "Subject: \n"); } } if (!has_to && item->email->sentto_address.str) { pst_rfc2047(item, &item->email->sentto_address, 0); fprintf(f_output, "To: %s\n", item->email->sentto_address.str); } if (!has_cc && item->email->cc_address.str) { pst_rfc2047(item, &item->email->cc_address, 0); fprintf(f_output, "Cc: %s\n", item->email->cc_address.str); } if (!has_date && item->email->sent_date) { char c_time[C_TIME_SIZE]; struct tm stm; gmtime_r(&em_time, &stm); strftime(c_time, C_TIME_SIZE, "%a, %d %b %Y %H:%M:%S %z", &stm); fprintf(f_output, "Date: %s\n", c_time); } if (!has_msgid && item->email->messageid.str) { pst_convert_utf8(item, &item->email->messageid); fprintf(f_output, "Message-Id: %s\n", item->email->messageid.str); } // add forensic headers to capture some .pst stuff that is not really // needed or used by mail clients pst_convert_utf8_null(item, &item->email->sender_address); if (item->email->sender_address.str && !strchr(item->email->sender_address.str, '@') && strcmp(item->email->sender_address.str, ".") && (strlen(item->email->sender_address.str) > 0)) { fprintf(f_output, "X-libpst-forensic-sender: %s\n", item->email->sender_address.str); } if (item->email->bcc_address.str) { pst_convert_utf8(item, &item->email->bcc_address); fprintf(f_output, "X-libpst-forensic-bcc: %s\n", item->email->bcc_address.str); } // add our own mime headers fprintf(f_output, "MIME-Version: 1.0\n"); if (item->type == PST_TYPE_REPORT) { // multipart/report for DSN/MDN reports fprintf(f_output, "Content-Type: multipart/report; report-type=%s;\n\tboundary=\"%s\"\n", body_report, boundary); } else { fprintf(f_output, "Content-Type: multipart/mixed;\n\tboundary=\"%s\"\n", boundary); } fprintf(f_output, "\n"); // end of headers, start of body // now dump the body parts if ((item->type == PST_TYPE_REPORT) && (item->email->report_text.str)) { write_body_part(f_output, &item->email->report_text, "text/plain", body_charset, boundary, pst); fprintf(f_output, "\n"); } if (item->body.str && item->email->htmlbody.str) { // start the nested alternative part fprintf(f_output, "\n--%s\n", boundary); fprintf(f_output, "Content-Type: multipart/alternative;\n\tboundary=\"%s\"\n", altboundary); altboundaryp = altboundary; } else { altboundaryp = boundary; } if (item->body.str) { write_body_part(f_output, &item->body, "text/plain", body_charset, altboundaryp, pst); } if (item->email->htmlbody.str) { find_html_charset(item->email->htmlbody.str, body_charset, sizeof(body_charset)); write_body_part(f_output, &item->email->htmlbody, "text/html", body_charset, altboundaryp, pst); } if (item->body.str && item->email->htmlbody.str) { // end the nested alternative part fprintf(f_output, "\n--%s--\n", altboundary); } if (item->email->rtf_compressed.data && save_rtf) { pst_item_attach* attach = (pst_item_attach*)pst_malloc(sizeof(pst_item_attach)); DEBUG_INFO(("Adding RTF body as attachment\n")); memset(attach, 0, sizeof(pst_item_attach)); attach->next = item->attach; item->attach = attach; attach->data.data = pst_lzfu_decompress(item->email->rtf_compressed.data, item->email->rtf_compressed.size, &attach->data.size); attach->filename2.str = strdup(RTF_ATTACH_NAME); attach->filename2.is_utf8 = 1; attach->mimetype.str = strdup(RTF_ATTACH_TYPE); attach->mimetype.is_utf8 = 1; } if (item->email->encrypted_body.data) { pst_item_attach* attach = (pst_item_attach*)pst_malloc(sizeof(pst_item_attach)); DEBUG_INFO(("Adding encrypted text body as attachment\n")); memset(attach, 0, sizeof(pst_item_attach)); attach->next = item->attach; item->attach = attach; attach->data.data = item->email->encrypted_body.data; attach->data.size = item->email->encrypted_body.size; item->email->encrypted_body.data = NULL; } if (item->email->encrypted_htmlbody.data) { pst_item_attach* attach = (pst_item_attach*)pst_malloc(sizeof(pst_item_attach)); DEBUG_INFO(("Adding encrypted HTML body as attachment\n")); memset(attach, 0, sizeof(pst_item_attach)); attach->next = item->attach; item->attach = attach; attach->data.data = item->email->encrypted_htmlbody.data; attach->data.size = item->email->encrypted_htmlbody.size; item->email->encrypted_htmlbody.data = NULL; } if (item->type == PST_TYPE_SCHEDULE) { write_schedule_part(f_output, item, sender, boundary); } // other attachments { pst_item_attach* attach; int attach_num = 0; for (attach = item->attach; attach; attach = attach->next) { pst_convert_utf8_null(item, &attach->filename1); pst_convert_utf8_null(item, &attach->filename2); pst_convert_utf8_null(item, &attach->mimetype); DEBUG_INFO(("Attempting Attachment encoding\n")); if (attach->method == PST_ATTACH_EMBEDDED) { DEBUG_INFO(("have an embedded rfc822 message attachment\n")); if (attach->mimetype.str) { DEBUG_INFO(("which already has a mime-type of %s\n", attach->mimetype.str)); free(attach->mimetype.str); } attach->mimetype.str = strdup(RFC822); attach->mimetype.is_utf8 = 1; find_rfc822_headers(extra_mime_headers); write_embedded_message(f_output, attach, boundary, pst, save_rtf, extra_mime_headers); } else if (attach->data.data || attach->i_id) { if (acceptable_ext(attach)) { if (current_mode == MODE_SEPARATE && !mode_MH) write_separate_attachment(f_name, attach, ++attach_num, pst); else write_inline_attachment(f_output, attach, boundary, pst); } } } } fprintf(f_output, "\n--%s--\n\n", boundary); DEBUG_RET(); } void write_vcard(FILE* f_output, pst_item* item, pst_item_contact* contact, char comment[]) { char* result = NULL; size_t resultlen = 0; char time_buffer[30]; // We can only call rfc escape once per printf, since the second call // may free the buffer returned by the first call. // I had tried to place those into a single printf - Carl. DEBUG_ENT("write_vcard"); // make everything utf8 pst_convert_utf8_null(item, &contact->fullname); pst_convert_utf8_null(item, &contact->surname); pst_convert_utf8_null(item, &contact->first_name); pst_convert_utf8_null(item, &contact->middle_name); pst_convert_utf8_null(item, &contact->display_name_prefix); pst_convert_utf8_null(item, &contact->suffix); pst_convert_utf8_null(item, &contact->nickname); pst_convert_utf8_null(item, &contact->address1); pst_convert_utf8_null(item, &contact->address2); pst_convert_utf8_null(item, &contact->address3); pst_convert_utf8_null(item, &contact->home_po_box); pst_convert_utf8_null(item, &contact->home_street); pst_convert_utf8_null(item, &contact->home_city); pst_convert_utf8_null(item, &contact->home_state); pst_convert_utf8_null(item, &contact->home_postal_code); pst_convert_utf8_null(item, &contact->home_country); pst_convert_utf8_null(item, &contact->home_address); pst_convert_utf8_null(item, &contact->business_po_box); pst_convert_utf8_null(item, &contact->business_street); pst_convert_utf8_null(item, &contact->business_city); pst_convert_utf8_null(item, &contact->business_state); pst_convert_utf8_null(item, &contact->business_postal_code); pst_convert_utf8_null(item, &contact->business_country); pst_convert_utf8_null(item, &contact->business_address); pst_convert_utf8_null(item, &contact->other_po_box); pst_convert_utf8_null(item, &contact->other_street); pst_convert_utf8_null(item, &contact->other_city); pst_convert_utf8_null(item, &contact->other_state); pst_convert_utf8_null(item, &contact->other_postal_code); pst_convert_utf8_null(item, &contact->other_country); pst_convert_utf8_null(item, &contact->other_address); pst_convert_utf8_null(item, &contact->business_fax); pst_convert_utf8_null(item, &contact->business_phone); pst_convert_utf8_null(item, &contact->business_phone2); pst_convert_utf8_null(item, &contact->car_phone); pst_convert_utf8_null(item, &contact->home_fax); pst_convert_utf8_null(item, &contact->home_phone); pst_convert_utf8_null(item, &contact->home_phone2); pst_convert_utf8_null(item, &contact->isdn_phone); pst_convert_utf8_null(item, &contact->mobile_phone); pst_convert_utf8_null(item, &contact->other_phone); pst_convert_utf8_null(item, &contact->pager_phone); pst_convert_utf8_null(item, &contact->primary_fax); pst_convert_utf8_null(item, &contact->primary_phone); pst_convert_utf8_null(item, &contact->radio_phone); pst_convert_utf8_null(item, &contact->telex); pst_convert_utf8_null(item, &contact->job_title); pst_convert_utf8_null(item, &contact->profession); pst_convert_utf8_null(item, &contact->assistant_name); pst_convert_utf8_null(item, &contact->assistant_phone); pst_convert_utf8_null(item, &contact->company_name); pst_convert_utf8_null(item, &item->body); // the specification I am following is (hopefully) RFC2426 vCard Mime Directory Profile fprintf(f_output, "BEGIN:VCARD\n"); fprintf(f_output, "FN:%s\n", pst_rfc2426_escape(contact->fullname.str, &result, &resultlen)); //fprintf(f_output, "N:%s;%s;%s;%s;%s\n", fprintf(f_output, "N:%s;", (!contact->surname.str) ? "" : pst_rfc2426_escape(contact->surname.str, &result, &resultlen)); fprintf(f_output, "%s;", (!contact->first_name.str) ? "" : pst_rfc2426_escape(contact->first_name.str, &result, &resultlen)); fprintf(f_output, "%s;", (!contact->middle_name.str) ? "" : pst_rfc2426_escape(contact->middle_name.str, &result, &resultlen)); fprintf(f_output, "%s;", (!contact->display_name_prefix.str) ? "" : pst_rfc2426_escape(contact->display_name_prefix.str, &result, &resultlen)); fprintf(f_output, "%s\n", (!contact->suffix.str) ? "" : pst_rfc2426_escape(contact->suffix.str, &result, &resultlen)); if (contact->nickname.str) fprintf(f_output, "NICKNAME:%s\n", pst_rfc2426_escape(contact->nickname.str, &result, &resultlen)); if (contact->address1.str) fprintf(f_output, "EMAIL:%s\n", pst_rfc2426_escape(contact->address1.str, &result, &resultlen)); if (contact->address2.str) fprintf(f_output, "EMAIL:%s\n", pst_rfc2426_escape(contact->address2.str, &result, &resultlen)); if (contact->address3.str) fprintf(f_output, "EMAIL:%s\n", pst_rfc2426_escape(contact->address3.str, &result, &resultlen)); if (contact->birthday) fprintf(f_output, "BDAY:%s\n", pst_rfc2425_datetime_format(contact->birthday, sizeof(time_buffer), time_buffer)); if (contact->home_address.str) { //fprintf(f_output, "ADR;TYPE=home:%s;%s;%s;%s;%s;%s;%s\n", fprintf(f_output, "ADR;TYPE=home:%s;", (!contact->home_po_box.str) ? "" : pst_rfc2426_escape(contact->home_po_box.str, &result, &resultlen)); fprintf(f_output, "%s;", ""); // extended Address fprintf(f_output, "%s;", (!contact->home_street.str) ? "" : pst_rfc2426_escape(contact->home_street.str, &result, &resultlen)); fprintf(f_output, "%s;", (!contact->home_city.str) ? "" : pst_rfc2426_escape(contact->home_city.str, &result, &resultlen)); fprintf(f_output, "%s;", (!contact->home_state.str) ? "" : pst_rfc2426_escape(contact->home_state.str, &result, &resultlen)); fprintf(f_output, "%s;", (!contact->home_postal_code.str) ? "" : pst_rfc2426_escape(contact->home_postal_code.str, &result, &resultlen)); fprintf(f_output, "%s\n", (!contact->home_country.str) ? "" : pst_rfc2426_escape(contact->home_country.str, &result, &resultlen)); fprintf(f_output, "LABEL;TYPE=home:%s\n", pst_rfc2426_escape(contact->home_address.str, &result, &resultlen)); } if (contact->business_address.str) { //fprintf(f_output, "ADR;TYPE=work:%s;%s;%s;%s;%s;%s;%s\n", fprintf(f_output, "ADR;TYPE=work:%s;", (!contact->business_po_box.str) ? "" : pst_rfc2426_escape(contact->business_po_box.str, &result, &resultlen)); fprintf(f_output, "%s;", ""); // extended Address fprintf(f_output, "%s;", (!contact->business_street.str) ? "" : pst_rfc2426_escape(contact->business_street.str, &result, &resultlen)); fprintf(f_output, "%s;", (!contact->business_city.str) ? "" : pst_rfc2426_escape(contact->business_city.str, &result, &resultlen)); fprintf(f_output, "%s;", (!contact->business_state.str) ? "" : pst_rfc2426_escape(contact->business_state.str, &result, &resultlen)); fprintf(f_output, "%s;", (!contact->business_postal_code.str) ? "" : pst_rfc2426_escape(contact->business_postal_code.str, &result, &resultlen)); fprintf(f_output, "%s\n", (!contact->business_country.str) ? "" : pst_rfc2426_escape(contact->business_country.str, &result, &resultlen)); fprintf(f_output, "LABEL;TYPE=work:%s\n", pst_rfc2426_escape(contact->business_address.str, &result, &resultlen)); } if (contact->other_address.str) { //fprintf(f_output, "ADR;TYPE=postal:%s;%s;%s;%s;%s;%s;%s\n", fprintf(f_output, "ADR;TYPE=postal:%s;",(!contact->other_po_box.str) ? "" : pst_rfc2426_escape(contact->other_po_box.str, &result, &resultlen)); fprintf(f_output, "%s;", ""); // extended Address fprintf(f_output, "%s;", (!contact->other_street.str) ? "" : pst_rfc2426_escape(contact->other_street.str, &result, &resultlen)); fprintf(f_output, "%s;", (!contact->other_city.str) ? "" : pst_rfc2426_escape(contact->other_city.str, &result, &resultlen)); fprintf(f_output, "%s;", (!contact->other_state.str) ? "" : pst_rfc2426_escape(contact->other_state.str, &result, &resultlen)); fprintf(f_output, "%s;", (!contact->other_postal_code.str) ? "" : pst_rfc2426_escape(contact->other_postal_code.str, &result, &resultlen)); fprintf(f_output, "%s\n", (!contact->other_country.str) ? "" : pst_rfc2426_escape(contact->other_country.str, &result, &resultlen)); fprintf(f_output, "LABEL;TYPE=postal:%s\n", pst_rfc2426_escape(contact->other_address.str, &result, &resultlen)); } if (contact->business_fax.str) fprintf(f_output, "TEL;TYPE=work,fax:%s\n", pst_rfc2426_escape(contact->business_fax.str, &result, &resultlen)); if (contact->business_phone.str) fprintf(f_output, "TEL;TYPE=work,voice:%s\n", pst_rfc2426_escape(contact->business_phone.str, &result, &resultlen)); if (contact->business_phone2.str) fprintf(f_output, "TEL;TYPE=work,voice:%s\n", pst_rfc2426_escape(contact->business_phone2.str, &result, &resultlen)); if (contact->car_phone.str) fprintf(f_output, "TEL;TYPE=car,voice:%s\n", pst_rfc2426_escape(contact->car_phone.str, &result, &resultlen)); if (contact->home_fax.str) fprintf(f_output, "TEL;TYPE=home,fax:%s\n", pst_rfc2426_escape(contact->home_fax.str, &result, &resultlen)); if (contact->home_phone.str) fprintf(f_output, "TEL;TYPE=home,voice:%s\n", pst_rfc2426_escape(contact->home_phone.str, &result, &resultlen)); if (contact->home_phone2.str) fprintf(f_output, "TEL;TYPE=home,voice:%s\n", pst_rfc2426_escape(contact->home_phone2.str, &result, &resultlen)); if (contact->isdn_phone.str) fprintf(f_output, "TEL;TYPE=isdn:%s\n", pst_rfc2426_escape(contact->isdn_phone.str, &result, &resultlen)); if (contact->mobile_phone.str) fprintf(f_output, "TEL;TYPE=cell,voice:%s\n", pst_rfc2426_escape(contact->mobile_phone.str, &result, &resultlen)); if (contact->other_phone.str) fprintf(f_output, "TEL;TYPE=msg:%s\n", pst_rfc2426_escape(contact->other_phone.str, &result, &resultlen)); if (contact->pager_phone.str) fprintf(f_output, "TEL;TYPE=pager:%s\n", pst_rfc2426_escape(contact->pager_phone.str, &result, &resultlen)); if (contact->primary_fax.str) fprintf(f_output, "TEL;TYPE=fax,pref:%s\n", pst_rfc2426_escape(contact->primary_fax.str, &result, &resultlen)); if (contact->primary_phone.str) fprintf(f_output, "TEL;TYPE=phone,pref:%s\n", pst_rfc2426_escape(contact->primary_phone.str, &result, &resultlen)); if (contact->radio_phone.str) fprintf(f_output, "TEL;TYPE=pcs:%s\n", pst_rfc2426_escape(contact->radio_phone.str, &result, &resultlen)); if (contact->telex.str) fprintf(f_output, "TEL;TYPE=bbs:%s\n", pst_rfc2426_escape(contact->telex.str, &result, &resultlen)); if (contact->job_title.str) fprintf(f_output, "TITLE:%s\n", pst_rfc2426_escape(contact->job_title.str, &result, &resultlen)); if (contact->profession.str) fprintf(f_output, "ROLE:%s\n", pst_rfc2426_escape(contact->profession.str, &result, &resultlen)); if (contact->assistant_name.str || contact->assistant_phone.str) { fprintf(f_output, "AGENT:BEGIN:VCARD\n"); if (contact->assistant_name.str) fprintf(f_output, "FN:%s\n", pst_rfc2426_escape(contact->assistant_name.str, &result, &resultlen)); if (contact->assistant_phone.str) fprintf(f_output, "TEL:%s\n", pst_rfc2426_escape(contact->assistant_phone.str, &result, &resultlen)); } if (contact->company_name.str) fprintf(f_output, "ORG:%s\n", pst_rfc2426_escape(contact->company_name.str, &result, &resultlen)); if (comment) fprintf(f_output, "NOTE:%s\n", pst_rfc2426_escape(comment, &result, &resultlen)); if (item->body.str) fprintf(f_output, "NOTE:%s\n", pst_rfc2426_escape(item->body.str, &result, &resultlen)); write_extra_categories(f_output, item); fprintf(f_output, "VERSION:3.0\n"); fprintf(f_output, "END:VCARD\n\n"); if (result) free(result); DEBUG_RET(); } /** * write extra vcard or vcalendar categories from the extra keywords fields * * @param f_output open file pointer * @param item pst item containing the keywords * @return true if we write a categories line */ int write_extra_categories(FILE* f_output, pst_item* item) { char* result = NULL; size_t resultlen = 0; pst_item_extra_field *ef = item->extra_fields; const char *fmt = "CATEGORIES:%s"; int category_started = 0; while (ef) { if (strcmp(ef->field_name, "Keywords") == 0) { fprintf(f_output, fmt, pst_rfc2426_escape(ef->value, &result, &resultlen)); fmt = ", %s"; category_started = 1; } ef = ef->next; } if (category_started) fprintf(f_output, "\n"); if (result) free(result); return category_started; } void write_journal(FILE* f_output, pst_item* item) { char* result = NULL; size_t resultlen = 0; char time_buffer[30]; pst_item_journal* journal = item->journal; // make everything utf8 pst_convert_utf8_null(item, &item->subject); pst_convert_utf8_null(item, &item->body); fprintf(f_output, "BEGIN:VJOURNAL\n"); // FIXME: use the attendee/owner critical change property for DTSTAMP // https://learn.microsoft.com/en-us/openspecs/exchange_server_protocols/ms-oxcical/4b93e7b6-142e-4f0c-ac08-1505a6fa0199 // https://learn.microsoft.com/en-us/office/client-developer/outlook/mapi/pidlidattendeecriticalchange-canonical-property // https://learn.microsoft.com/en-us/office/client-developer/outlook/mapi/pidlidownercriticalchange-canonical-property if (item->create_date) fprintf(f_output, "DTSTAMP:%s\n", pst_rfc2445_datetime_format(item->create_date, sizeof(time_buffer), time_buffer)); else fprintf(f_output, "DTSTAMP:%s\n", pst_rfc2445_datetime_format_now(sizeof(time_buffer), time_buffer)); if (item->create_date) fprintf(f_output, "CREATED:%s\n", pst_rfc2445_datetime_format(item->create_date, sizeof(time_buffer), time_buffer)); if (item->modify_date) fprintf(f_output, "LAST-MOD:%s\n", pst_rfc2445_datetime_format(item->modify_date, sizeof(time_buffer), time_buffer)); if (item->subject.str) fprintf(f_output, "SUMMARY:%s\n", pst_rfc2426_escape(item->subject.str, &result, &resultlen)); if (item->body.str) fprintf(f_output, "DESCRIPTION:%s\n", pst_rfc2426_escape(item->body.str, &result, &resultlen)); if (journal && journal->start) fprintf(f_output, "DTSTART;VALUE=DATE-TIME:%s\n", pst_rfc2445_datetime_format(journal->start, sizeof(time_buffer), time_buffer)); fprintf(f_output, "END:VJOURNAL\n"); if (result) free(result); } void write_appointment(FILE* f_output, pst_item* item) { char* result = NULL; size_t resultlen = 0; char time_buffer[30]; pst_item_appointment* appointment = item->appointment; // make everything utf8 pst_convert_utf8_null(item, &item->subject); pst_convert_utf8_null(item, &item->body); pst_convert_utf8_null(item, &appointment->location); fprintf(f_output, "UID:%#" PRIx64 "\n", item->block_id); // FIXME: use the attendee/owner critical change property for DTSTAMP // https://learn.microsoft.com/en-us/openspecs/exchange_server_protocols/ms-oxcical/4b93e7b6-142e-4f0c-ac08-1505a6fa0199 // https://learn.microsoft.com/en-us/office/client-developer/outlook/mapi/pidlidattendeecriticalchange-canonical-property // https://learn.microsoft.com/en-us/office/client-developer/outlook/mapi/pidlidownercriticalchange-canonical-property if (item->create_date) fprintf(f_output, "DTSTAMP:%s\n", pst_rfc2445_datetime_format(item->create_date, sizeof(time_buffer), time_buffer)); else fprintf(f_output, "DTSTAMP:%s\n", pst_rfc2445_datetime_format_now(sizeof(time_buffer), time_buffer)); if (item->create_date) fprintf(f_output, "CREATED:%s\n", pst_rfc2445_datetime_format(item->create_date, sizeof(time_buffer), time_buffer)); if (item->modify_date) fprintf(f_output, "LAST-MOD:%s\n", pst_rfc2445_datetime_format(item->modify_date, sizeof(time_buffer), time_buffer)); if (item->subject.str) fprintf(f_output, "SUMMARY:%s\n", pst_rfc2426_escape(item->subject.str, &result, &resultlen)); if (item->body.str) fprintf(f_output, "DESCRIPTION:%s\n", pst_rfc2426_escape(item->body.str, &result, &resultlen)); if (appointment && appointment->start) fprintf(f_output, "DTSTART;VALUE=DATE-TIME:%s\n", pst_rfc2445_datetime_format(appointment->start, sizeof(time_buffer), time_buffer)); if (appointment && appointment->end) fprintf(f_output, "DTEND;VALUE=DATE-TIME:%s\n", pst_rfc2445_datetime_format(appointment->end, sizeof(time_buffer), time_buffer)); if (appointment && appointment->location.str) fprintf(f_output, "LOCATION:%s\n", pst_rfc2426_escape(appointment->location.str, &result, &resultlen)); if (appointment) { switch (appointment->showas) { case PST_FREEBUSY_TENTATIVE: fprintf(f_output, "STATUS:TENTATIVE\n"); break; case PST_FREEBUSY_FREE: // mark as transparent and as confirmed fprintf(f_output, "TRANSP:TRANSPARENT\n"); fprintf(f_output, "STATUS:CONFIRMED\n"); break; case PST_FREEBUSY_BUSY: case PST_FREEBUSY_OUT_OF_OFFICE: fprintf(f_output, "STATUS:CONFIRMED\n"); break; } if (appointment->is_recurring) { const char* rules[] = {"DAILY", "WEEKLY", "MONTHLY", "YEARLY"}; const char* days[] = {"SU", "MO", "TU", "WE", "TH", "FR", "SA"}; pst_recurrence *rdata = pst_convert_recurrence(appointment); fprintf(f_output, "RRULE:FREQ=%s", rules[rdata->type]); if (rdata->count) fprintf(f_output, ";COUNT=%" PRIu32, rdata->count); if ((rdata->interval != 1) && (rdata->interval)) fprintf(f_output, ";INTERVAL=%" PRIu32, rdata->interval); if (rdata->dayofmonth) fprintf(f_output, ";BYMONTHDAY=%" PRIu32, rdata->dayofmonth); if (rdata->monthofyear) fprintf(f_output, ";BYMONTH=%" PRIu32, rdata->monthofyear); if (rdata->position) fprintf(f_output, ";BYSETPOS=%" PRIu32, rdata->position); if (rdata->bydaymask) { char byday[40]; int empty = 1; int i=0; memset(byday, 0, sizeof(byday)); for (i=0; i<7; i++) { int bit = 1 << i; if (bit & rdata->bydaymask) { char temp[49]; snprintf(temp, sizeof(temp), "%s%s%s", byday, (empty) ? ";BYDAY=" : ";", days[i]); strcpy(byday, temp); empty = 0; } } fprintf(f_output, "%s", byday); } fprintf(f_output, "\n"); pst_free_recurrence(rdata); } switch (appointment->label) { case PST_APP_LABEL_NONE: if (!write_extra_categories(f_output, item)) fprintf(f_output, "CATEGORIES:NONE\n"); break; case PST_APP_LABEL_IMPORTANT: fprintf(f_output, "CATEGORIES:IMPORTANT\n"); break; case PST_APP_LABEL_BUSINESS: fprintf(f_output, "CATEGORIES:BUSINESS\n"); break; case PST_APP_LABEL_PERSONAL: fprintf(f_output, "CATEGORIES:PERSONAL\n"); break; case PST_APP_LABEL_VACATION: fprintf(f_output, "CATEGORIES:VACATION\n"); break; case PST_APP_LABEL_MUST_ATTEND: fprintf(f_output, "CATEGORIES:MUST-ATTEND\n"); break; case PST_APP_LABEL_TRAVEL_REQ: fprintf(f_output, "CATEGORIES:TRAVEL-REQUIRED\n"); break; case PST_APP_LABEL_NEEDS_PREP: fprintf(f_output, "CATEGORIES:NEEDS-PREPARATION\n"); break; case PST_APP_LABEL_BIRTHDAY: fprintf(f_output, "CATEGORIES:BIRTHDAY\n"); break; case PST_APP_LABEL_ANNIVERSARY: fprintf(f_output, "CATEGORIES:ANNIVERSARY\n"); break; case PST_APP_LABEL_PHONE_CALL: fprintf(f_output, "CATEGORIES:PHONE-CALL\n"); break; } // ignore bogus alarms if (appointment->alarm && (appointment->alarm_minutes >= 0) && (appointment->alarm_minutes < 1440)) { fprintf(f_output, "BEGIN:VALARM\n"); fprintf(f_output, "TRIGGER:-PT%" PRId32 "M\n", appointment->alarm_minutes); fprintf(f_output, "ACTION:DISPLAY\n"); fprintf(f_output, "DESCRIPTION:Reminder\n"); fprintf(f_output, "END:VALARM\n"); } } fprintf(f_output, "END:VEVENT\n"); if (result) free(result); } void create_enter_dir(struct file_ll* f, pst_item *item) { memset(f, 0, sizeof(*f)); f->stored_count = (item->folder) ? item->folder->item_count : 0; pst_convert_utf8(item, &item->file_as); f->dname = (char*) pst_malloc(strlen(item->file_as.str)+1); strcpy(f->dname, item->file_as.str); DEBUG_ENT("create_enter_dir"); if (mode == MODE_KMAIL) { int32_t t; mk_kmail_dir(item->file_as.str); for (t=0; tname[t] = (char*) pst_malloc(strlen(item->file_as.str)+strlen(OUTPUT_TEMPLATE)+30); sprintf(f->name[t], OUTPUT_TEMPLATE, item->file_as.str, item_type_to_name(t)); } } } else if (mode == MODE_RECURSE) { int32_t t; mk_recurse_dir(item->file_as.str); for (t=0; tname[t] = strdup(item_type_to_name(t)); } } if (mode_thunder) { FILE *type_file = fopen(".type", "w"); if (type_file) { fprintf(type_file, "%d\n", item->type); fclose(type_file); } else { DEBUG_WARN(("could not write .type file: %d\n", item->type)); } } } else if (mode == MODE_SEPARATE) { // do similar stuff to recurse here. int32_t t; mk_separate_dir(item->file_as.str); for (t=0; tname[t] = (char*) pst_malloc(file_name_len); memset(f->name[t], 0, file_name_len); } } } else { // MODE_NORMAL int32_t t; for (t=0; tname[t] = (char*) pst_malloc(strlen(item->file_as.str)+strlen(OUTPUT_TEMPLATE)+30); sprintf(f->name[t], OUTPUT_TEMPLATE, item->file_as.str, item_type_to_name(t)); } } } if (mode != MODE_SEPARATE) { int32_t t; for (t=0; tname[t]) { if (!overwrite) { int x = 0; char *temp = (char*) pst_malloc (strlen(f->name[t])+10); //enough room for 10 digits sprintf(temp, "%s", f->name[t]); check_filename(temp); while ((f->output[t] = fopen(temp, "r"))) { DEBUG_INFO(("need to increase filename because one already exists with that name\n")); x++; sprintf(temp, "%s%08d", f->name[t], x); DEBUG_INFO(("- bump file name and try \"%s\"\n", temp)); if (x == 99999999) { DIE(("create_enter_dir: Why can I not create a folder %s? I have tried %i extensions...\n", f->name[t], x)); } fclose(f->output[t]); } if (x > 0) { //then the f->name should change free (f->name[t]); f->name[t] = temp; } else { free(temp); } } check_filename(f->name[t]); if (!(f->output[t] = fopen(f->name[t], "w"))) { DIE(("create_enter_dir: Could not open file \"%s\" for write\n", f->name[t])); } DEBUG_INFO(("f->name = %s\nitem->folder_name = %s\n", f->name[t], item->file_as.str)); } } } DEBUG_RET(); } void close_enter_dir(struct file_ll *f) { int32_t t; DEBUG_INFO(("processed item count for folder %s is %i, skipped %i, total %i \n", f->dname, f->item_count, f->skip_count, f->stored_count)); if (output_mode != OUTPUT_QUIET) { pst_debug_lock(); printf("\t\"%s\" - %" PRIi32 " items done, %" PRIi32 " items skipped.\n", f->dname, f->item_count, f->skip_count); fflush(stdout); pst_debug_unlock(); } for (t=0; toutput[t]) { if (mode == MODE_SEPARATE) DEBUG_WARN(("close_enter_dir finds open separate file\n")); fclose(f->output[t]); f->output[t] = NULL; } if (f->name[t]) { struct stat st; stat(f->name[t], &st); if (!st.st_size) { DEBUG_WARN(("removing empty output file %s\n", f->name[t])); remove(f->name[t]); } free(f->name[t]); f->name[t] = NULL; } } free(f->dname); if (mode == MODE_KMAIL) close_kmail_dir(); else if (mode == MODE_RECURSE) { if (mode_thunder) { FILE *type_file = fopen(".size", "w"); if (type_file) { fprintf(type_file, "%" PRIi32 " %" PRIi32 "\n", f->item_count, f->stored_count); fclose(type_file); } else { DEBUG_WARN(("could not write .size file: %" PRIi32 " %" PRIi32 "\n", f->item_count, f->stored_count)); } } close_recurse_dir(); } else if (mode == MODE_SEPARATE) close_separate_dir(); }