You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

1700 lines
57KB

  1. /*
  2. * Apple HTTP Live Streaming demuxer
  3. * Copyright (c) 2010 Martin Storsjo
  4. * Copyright (c) 2013 Anssi Hannula
  5. *
  6. * This file is part of FFmpeg.
  7. *
  8. * FFmpeg is free software; you can redistribute it and/or
  9. * modify it under the terms of the GNU Lesser General Public
  10. * License as published by the Free Software Foundation; either
  11. * version 2.1 of the License, or (at your option) any later version.
  12. *
  13. * FFmpeg is distributed in the hope that it will be useful,
  14. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16. * Lesser General Public License for more details.
  17. *
  18. * You should have received a copy of the GNU Lesser General Public
  19. * License along with FFmpeg; if not, write to the Free Software
  20. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21. */
  22. /**
  23. * @file
  24. * Apple HTTP Live Streaming demuxer
  25. * http://tools.ietf.org/html/draft-pantos-http-live-streaming
  26. */
  27. #include "libavutil/avstring.h"
  28. #include "libavutil/avassert.h"
  29. #include "libavutil/intreadwrite.h"
  30. #include "libavutil/mathematics.h"
  31. #include "libavutil/opt.h"
  32. #include "libavutil/dict.h"
  33. #include "libavutil/time.h"
  34. #include "avformat.h"
  35. #include "internal.h"
  36. #include "avio_internal.h"
  37. #include "url.h"
  38. #include "id3v2.h"
  39. #define INITIAL_BUFFER_SIZE 32768
  40. #define MAX_FIELD_LEN 64
  41. #define MAX_CHARACTERISTICS_LEN 512
  42. #define MPEG_TIME_BASE 90000
  43. #define MPEG_TIME_BASE_Q (AVRational){1, MPEG_TIME_BASE}
  44. /*
  45. * An apple http stream consists of a playlist with media segment files,
  46. * played sequentially. There may be several playlists with the same
  47. * video content, in different bandwidth variants, that are played in
  48. * parallel (preferably only one bandwidth variant at a time). In this case,
  49. * the user supplied the url to a main playlist that only lists the variant
  50. * playlists.
  51. *
  52. * If the main playlist doesn't point at any variants, we still create
  53. * one anonymous toplevel variant for this, to maintain the structure.
  54. */
  55. enum KeyType {
  56. KEY_NONE,
  57. KEY_AES_128,
  58. };
  59. struct segment {
  60. int64_t duration;
  61. int64_t url_offset;
  62. int64_t size;
  63. char url[MAX_URL_SIZE];
  64. char key[MAX_URL_SIZE];
  65. enum KeyType key_type;
  66. uint8_t iv[16];
  67. };
  68. struct rendition;
  69. enum PlaylistType {
  70. PLS_TYPE_UNSPECIFIED,
  71. PLS_TYPE_EVENT,
  72. PLS_TYPE_VOD
  73. };
  74. /*
  75. * Each playlist has its own demuxer. If it currently is active,
  76. * it has an open AVIOContext too, and potentially an AVPacket
  77. * containing the next packet from this stream.
  78. */
  79. struct playlist {
  80. char url[MAX_URL_SIZE];
  81. AVIOContext pb;
  82. uint8_t* read_buffer;
  83. URLContext *input;
  84. AVFormatContext *parent;
  85. int index;
  86. AVFormatContext *ctx;
  87. AVPacket pkt;
  88. int stream_offset;
  89. int finished;
  90. enum PlaylistType type;
  91. int64_t target_duration;
  92. int start_seq_no;
  93. int n_segments;
  94. struct segment **segments;
  95. int needed, cur_needed;
  96. int cur_seq_no;
  97. int64_t cur_seg_offset;
  98. int64_t last_load_time;
  99. char key_url[MAX_URL_SIZE];
  100. uint8_t key[16];
  101. /* ID3 timestamp handling (elementary audio streams have ID3 timestamps
  102. * (and possibly other ID3 tags) in the beginning of each segment) */
  103. int is_id3_timestamped; /* -1: not yet known */
  104. int64_t id3_mpegts_timestamp; /* in mpegts tb */
  105. int64_t id3_offset; /* in stream original tb */
  106. uint8_t* id3_buf; /* temp buffer for id3 parsing */
  107. unsigned int id3_buf_size;
  108. AVDictionary *id3_initial; /* data from first id3 tag */
  109. int id3_found; /* ID3 tag found at some point */
  110. int id3_changed; /* ID3 tag data has changed at some point */
  111. ID3v2ExtraMeta *id3_deferred_extra; /* stored here until subdemuxer is opened */
  112. int64_t seek_timestamp;
  113. int seek_flags;
  114. int seek_stream_index; /* into subdemuxer stream array */
  115. /* Renditions associated with this playlist, if any.
  116. * Alternative rendition playlists have a single rendition associated
  117. * with them, and variant main Media Playlists may have
  118. * multiple (playlist-less) renditions associated with them. */
  119. int n_renditions;
  120. struct rendition **renditions;
  121. };
  122. /*
  123. * Renditions are e.g. alternative subtitle or audio streams.
  124. * The rendition may either be an external playlist or it may be
  125. * contained in the main Media Playlist of the variant (in which case
  126. * playlist is NULL).
  127. */
  128. struct rendition {
  129. enum AVMediaType type;
  130. struct playlist *playlist;
  131. char group_id[MAX_FIELD_LEN];
  132. char language[MAX_FIELD_LEN];
  133. char name[MAX_FIELD_LEN];
  134. int disposition;
  135. };
  136. struct variant {
  137. int bandwidth;
  138. /* every variant contains at least the main Media Playlist in index 0 */
  139. int n_playlists;
  140. struct playlist **playlists;
  141. char audio_group[MAX_FIELD_LEN];
  142. char video_group[MAX_FIELD_LEN];
  143. char subtitles_group[MAX_FIELD_LEN];
  144. };
  145. typedef struct HLSContext {
  146. int n_variants;
  147. struct variant **variants;
  148. int n_playlists;
  149. struct playlist **playlists;
  150. int n_renditions;
  151. struct rendition **renditions;
  152. int cur_seq_no;
  153. int first_packet;
  154. int64_t first_timestamp;
  155. int64_t cur_timestamp;
  156. AVIOInterruptCB *interrupt_callback;
  157. char *user_agent; ///< holds HTTP user agent set as an AVOption to the HTTP protocol context
  158. char *cookies; ///< holds HTTP cookie values set in either the initial response or as an AVOption to the HTTP protocol context
  159. char *headers; ///< holds HTTP headers set as an AVOption to the HTTP protocol context
  160. } HLSContext;
  161. static int read_chomp_line(AVIOContext *s, char *buf, int maxlen)
  162. {
  163. int len = ff_get_line(s, buf, maxlen);
  164. while (len > 0 && av_isspace(buf[len - 1]))
  165. buf[--len] = '\0';
  166. return len;
  167. }
  168. static void free_segment_list(struct playlist *pls)
  169. {
  170. int i;
  171. for (i = 0; i < pls->n_segments; i++)
  172. av_free(pls->segments[i]);
  173. av_freep(&pls->segments);
  174. pls->n_segments = 0;
  175. }
  176. static void free_playlist_list(HLSContext *c)
  177. {
  178. int i;
  179. for (i = 0; i < c->n_playlists; i++) {
  180. struct playlist *pls = c->playlists[i];
  181. free_segment_list(pls);
  182. av_freep(&pls->renditions);
  183. av_freep(&pls->id3_buf);
  184. av_dict_free(&pls->id3_initial);
  185. ff_id3v2_free_extra_meta(&pls->id3_deferred_extra);
  186. av_free_packet(&pls->pkt);
  187. av_free(pls->pb.buffer);
  188. if (pls->input)
  189. ffurl_close(pls->input);
  190. if (pls->ctx) {
  191. pls->ctx->pb = NULL;
  192. avformat_close_input(&pls->ctx);
  193. }
  194. av_free(pls);
  195. }
  196. av_freep(&c->playlists);
  197. av_freep(&c->cookies);
  198. av_freep(&c->user_agent);
  199. c->n_playlists = 0;
  200. }
  201. static void free_variant_list(HLSContext *c)
  202. {
  203. int i;
  204. for (i = 0; i < c->n_variants; i++) {
  205. struct variant *var = c->variants[i];
  206. av_freep(&var->playlists);
  207. av_free(var);
  208. }
  209. av_freep(&c->variants);
  210. c->n_variants = 0;
  211. }
  212. static void free_rendition_list(HLSContext *c)
  213. {
  214. int i;
  215. for (i = 0; i < c->n_renditions; i++)
  216. av_free(c->renditions[i]);
  217. av_freep(&c->renditions);
  218. c->n_renditions = 0;
  219. }
  220. /*
  221. * Used to reset a statically allocated AVPacket to a clean slate,
  222. * containing no data.
  223. */
  224. static void reset_packet(AVPacket *pkt)
  225. {
  226. av_init_packet(pkt);
  227. pkt->data = NULL;
  228. }
  229. static struct playlist *new_playlist(HLSContext *c, const char *url,
  230. const char *base)
  231. {
  232. struct playlist *pls = av_mallocz(sizeof(struct playlist));
  233. if (!pls)
  234. return NULL;
  235. reset_packet(&pls->pkt);
  236. ff_make_absolute_url(pls->url, sizeof(pls->url), base, url);
  237. pls->seek_timestamp = AV_NOPTS_VALUE;
  238. pls->is_id3_timestamped = -1;
  239. pls->id3_mpegts_timestamp = AV_NOPTS_VALUE;
  240. dynarray_add(&c->playlists, &c->n_playlists, pls);
  241. return pls;
  242. }
  243. struct variant_info {
  244. char bandwidth[20];
  245. /* variant group ids: */
  246. char audio[MAX_FIELD_LEN];
  247. char video[MAX_FIELD_LEN];
  248. char subtitles[MAX_FIELD_LEN];
  249. };
  250. static struct variant *new_variant(HLSContext *c, struct variant_info *info,
  251. const char *url, const char *base)
  252. {
  253. struct variant *var;
  254. struct playlist *pls;
  255. pls = new_playlist(c, url, base);
  256. if (!pls)
  257. return NULL;
  258. var = av_mallocz(sizeof(struct variant));
  259. if (!var)
  260. return NULL;
  261. if (info) {
  262. var->bandwidth = atoi(info->bandwidth);
  263. strcpy(var->audio_group, info->audio);
  264. strcpy(var->video_group, info->video);
  265. strcpy(var->subtitles_group, info->subtitles);
  266. }
  267. dynarray_add(&c->variants, &c->n_variants, var);
  268. dynarray_add(&var->playlists, &var->n_playlists, pls);
  269. return var;
  270. }
  271. static void handle_variant_args(struct variant_info *info, const char *key,
  272. int key_len, char **dest, int *dest_len)
  273. {
  274. if (!strncmp(key, "BANDWIDTH=", key_len)) {
  275. *dest = info->bandwidth;
  276. *dest_len = sizeof(info->bandwidth);
  277. } else if (!strncmp(key, "AUDIO=", key_len)) {
  278. *dest = info->audio;
  279. *dest_len = sizeof(info->audio);
  280. } else if (!strncmp(key, "VIDEO=", key_len)) {
  281. *dest = info->video;
  282. *dest_len = sizeof(info->video);
  283. } else if (!strncmp(key, "SUBTITLES=", key_len)) {
  284. *dest = info->subtitles;
  285. *dest_len = sizeof(info->subtitles);
  286. }
  287. }
  288. struct key_info {
  289. char uri[MAX_URL_SIZE];
  290. char method[10];
  291. char iv[35];
  292. };
  293. static void handle_key_args(struct key_info *info, const char *key,
  294. int key_len, char **dest, int *dest_len)
  295. {
  296. if (!strncmp(key, "METHOD=", key_len)) {
  297. *dest = info->method;
  298. *dest_len = sizeof(info->method);
  299. } else if (!strncmp(key, "URI=", key_len)) {
  300. *dest = info->uri;
  301. *dest_len = sizeof(info->uri);
  302. } else if (!strncmp(key, "IV=", key_len)) {
  303. *dest = info->iv;
  304. *dest_len = sizeof(info->iv);
  305. }
  306. }
  307. struct rendition_info {
  308. char type[16];
  309. char uri[MAX_URL_SIZE];
  310. char group_id[MAX_FIELD_LEN];
  311. char language[MAX_FIELD_LEN];
  312. char assoc_language[MAX_FIELD_LEN];
  313. char name[MAX_FIELD_LEN];
  314. char defaultr[4];
  315. char forced[4];
  316. char characteristics[MAX_CHARACTERISTICS_LEN];
  317. };
  318. static struct rendition *new_rendition(HLSContext *c, struct rendition_info *info,
  319. const char *url_base)
  320. {
  321. struct rendition *rend;
  322. enum AVMediaType type = AVMEDIA_TYPE_UNKNOWN;
  323. char *characteristic;
  324. char *chr_ptr;
  325. char *saveptr;
  326. if (!strcmp(info->type, "AUDIO"))
  327. type = AVMEDIA_TYPE_AUDIO;
  328. else if (!strcmp(info->type, "VIDEO"))
  329. type = AVMEDIA_TYPE_VIDEO;
  330. else if (!strcmp(info->type, "SUBTITLES"))
  331. type = AVMEDIA_TYPE_SUBTITLE;
  332. else if (!strcmp(info->type, "CLOSED-CAPTIONS"))
  333. /* CLOSED-CAPTIONS is ignored since we do not support CEA-608 CC in
  334. * AVC SEI RBSP anyway */
  335. return NULL;
  336. if (type == AVMEDIA_TYPE_UNKNOWN)
  337. return NULL;
  338. /* URI is mandatory for subtitles as per spec */
  339. if (type == AVMEDIA_TYPE_SUBTITLE && !info->uri[0])
  340. return NULL;
  341. /* TODO: handle subtitles (each segment has to parsed separately) */
  342. if (type == AVMEDIA_TYPE_SUBTITLE)
  343. return NULL;
  344. rend = av_mallocz(sizeof(struct rendition));
  345. if (!rend)
  346. return NULL;
  347. dynarray_add(&c->renditions, &c->n_renditions, rend);
  348. rend->type = type;
  349. strcpy(rend->group_id, info->group_id);
  350. strcpy(rend->language, info->language);
  351. strcpy(rend->name, info->name);
  352. /* add the playlist if this is an external rendition */
  353. if (info->uri[0]) {
  354. rend->playlist = new_playlist(c, info->uri, url_base);
  355. if (rend->playlist)
  356. dynarray_add(&rend->playlist->renditions,
  357. &rend->playlist->n_renditions, rend);
  358. }
  359. if (info->assoc_language[0]) {
  360. int langlen = strlen(rend->language);
  361. if (langlen < sizeof(rend->language) - 3) {
  362. rend->language[langlen] = ',';
  363. strncpy(rend->language + langlen + 1, info->assoc_language,
  364. sizeof(rend->language) - langlen - 2);
  365. }
  366. }
  367. if (!strcmp(info->defaultr, "YES"))
  368. rend->disposition |= AV_DISPOSITION_DEFAULT;
  369. if (!strcmp(info->forced, "YES"))
  370. rend->disposition |= AV_DISPOSITION_FORCED;
  371. chr_ptr = info->characteristics;
  372. while ((characteristic = av_strtok(chr_ptr, ",", &saveptr))) {
  373. if (!strcmp(characteristic, "public.accessibility.describes-music-and-sound"))
  374. rend->disposition |= AV_DISPOSITION_HEARING_IMPAIRED;
  375. else if (!strcmp(characteristic, "public.accessibility.describes-video"))
  376. rend->disposition |= AV_DISPOSITION_VISUAL_IMPAIRED;
  377. chr_ptr = NULL;
  378. }
  379. return rend;
  380. }
  381. static void handle_rendition_args(struct rendition_info *info, const char *key,
  382. int key_len, char **dest, int *dest_len)
  383. {
  384. if (!strncmp(key, "TYPE=", key_len)) {
  385. *dest = info->type;
  386. *dest_len = sizeof(info->type);
  387. } else if (!strncmp(key, "URI=", key_len)) {
  388. *dest = info->uri;
  389. *dest_len = sizeof(info->uri);
  390. } else if (!strncmp(key, "GROUP-ID=", key_len)) {
  391. *dest = info->group_id;
  392. *dest_len = sizeof(info->group_id);
  393. } else if (!strncmp(key, "LANGUAGE=", key_len)) {
  394. *dest = info->language;
  395. *dest_len = sizeof(info->language);
  396. } else if (!strncmp(key, "ASSOC-LANGUAGE=", key_len)) {
  397. *dest = info->assoc_language;
  398. *dest_len = sizeof(info->assoc_language);
  399. } else if (!strncmp(key, "NAME=", key_len)) {
  400. *dest = info->name;
  401. *dest_len = sizeof(info->name);
  402. } else if (!strncmp(key, "DEFAULT=", key_len)) {
  403. *dest = info->defaultr;
  404. *dest_len = sizeof(info->defaultr);
  405. } else if (!strncmp(key, "FORCED=", key_len)) {
  406. *dest = info->forced;
  407. *dest_len = sizeof(info->forced);
  408. } else if (!strncmp(key, "CHARACTERISTICS=", key_len)) {
  409. *dest = info->characteristics;
  410. *dest_len = sizeof(info->characteristics);
  411. }
  412. /*
  413. * ignored:
  414. * - AUTOSELECT: client may autoselect based on e.g. system language
  415. * - INSTREAM-ID: EIA-608 closed caption number ("CC1".."CC4")
  416. */
  417. }
  418. /* used by parse_playlist to allocate a new variant+playlist when the
  419. * playlist is detected to be a Media Playlist (not Master Playlist)
  420. * and we have no parent Master Playlist (parsing of which would have
  421. * allocated the variant and playlist already) */
  422. static int ensure_playlist(HLSContext *c, struct playlist **pls, const char *url)
  423. {
  424. if (*pls)
  425. return 0;
  426. if (!new_variant(c, NULL, url, NULL))
  427. return AVERROR(ENOMEM);
  428. *pls = c->playlists[c->n_playlists - 1];
  429. return 0;
  430. }
  431. /* pls = NULL => Master Playlist or parentless Media Playlist
  432. * pls = !NULL => parented Media Playlist, playlist+variant allocated */
  433. static int parse_playlist(HLSContext *c, const char *url,
  434. struct playlist *pls, AVIOContext *in)
  435. {
  436. int ret = 0, is_segment = 0, is_variant = 0;
  437. int64_t duration = 0;
  438. enum KeyType key_type = KEY_NONE;
  439. uint8_t iv[16] = "";
  440. int has_iv = 0;
  441. char key[MAX_URL_SIZE] = "";
  442. char line[MAX_URL_SIZE];
  443. const char *ptr;
  444. int close_in = 0;
  445. int64_t seg_offset = 0;
  446. int64_t seg_size = -1;
  447. uint8_t *new_url = NULL;
  448. struct variant_info variant_info;
  449. if (!in) {
  450. AVDictionary *opts = NULL;
  451. close_in = 1;
  452. /* Some HLS servers don't like being sent the range header */
  453. av_dict_set(&opts, "seekable", "0", 0);
  454. // broker prior HTTP options that should be consistent across requests
  455. av_dict_set(&opts, "user-agent", c->user_agent, 0);
  456. av_dict_set(&opts, "cookies", c->cookies, 0);
  457. av_dict_set(&opts, "headers", c->headers, 0);
  458. ret = avio_open2(&in, url, AVIO_FLAG_READ,
  459. c->interrupt_callback, &opts);
  460. av_dict_free(&opts);
  461. if (ret < 0)
  462. return ret;
  463. }
  464. if (av_opt_get(in, "location", AV_OPT_SEARCH_CHILDREN, &new_url) >= 0)
  465. url = new_url;
  466. read_chomp_line(in, line, sizeof(line));
  467. if (strcmp(line, "#EXTM3U")) {
  468. ret = AVERROR_INVALIDDATA;
  469. goto fail;
  470. }
  471. if (pls) {
  472. free_segment_list(pls);
  473. pls->finished = 0;
  474. pls->type = PLS_TYPE_UNSPECIFIED;
  475. }
  476. while (!url_feof(in)) {
  477. read_chomp_line(in, line, sizeof(line));
  478. if (av_strstart(line, "#EXT-X-STREAM-INF:", &ptr)) {
  479. is_variant = 1;
  480. memset(&variant_info, 0, sizeof(variant_info));
  481. ff_parse_key_value(ptr, (ff_parse_key_val_cb) handle_variant_args,
  482. &variant_info);
  483. } else if (av_strstart(line, "#EXT-X-KEY:", &ptr)) {
  484. struct key_info info = {{0}};
  485. ff_parse_key_value(ptr, (ff_parse_key_val_cb) handle_key_args,
  486. &info);
  487. key_type = KEY_NONE;
  488. has_iv = 0;
  489. if (!strcmp(info.method, "AES-128"))
  490. key_type = KEY_AES_128;
  491. if (!strncmp(info.iv, "0x", 2) || !strncmp(info.iv, "0X", 2)) {
  492. ff_hex_to_data(iv, info.iv + 2);
  493. has_iv = 1;
  494. }
  495. av_strlcpy(key, info.uri, sizeof(key));
  496. } else if (av_strstart(line, "#EXT-X-MEDIA:", &ptr)) {
  497. struct rendition_info info = {{0}};
  498. ff_parse_key_value(ptr, (ff_parse_key_val_cb) handle_rendition_args,
  499. &info);
  500. new_rendition(c, &info, url);
  501. } else if (av_strstart(line, "#EXT-X-TARGETDURATION:", &ptr)) {
  502. ret = ensure_playlist(c, &pls, url);
  503. if (ret < 0)
  504. goto fail;
  505. pls->target_duration = atoi(ptr) * AV_TIME_BASE;
  506. } else if (av_strstart(line, "#EXT-X-MEDIA-SEQUENCE:", &ptr)) {
  507. ret = ensure_playlist(c, &pls, url);
  508. if (ret < 0)
  509. goto fail;
  510. pls->start_seq_no = atoi(ptr);
  511. } else if (av_strstart(line, "#EXT-X-PLAYLIST-TYPE:", &ptr)) {
  512. ret = ensure_playlist(c, &pls, url);
  513. if (ret < 0)
  514. goto fail;
  515. if (!strcmp(ptr, "EVENT"))
  516. pls->type = PLS_TYPE_EVENT;
  517. else if (!strcmp(ptr, "VOD"))
  518. pls->type = PLS_TYPE_VOD;
  519. } else if (av_strstart(line, "#EXT-X-ENDLIST", &ptr)) {
  520. if (pls)
  521. pls->finished = 1;
  522. } else if (av_strstart(line, "#EXTINF:", &ptr)) {
  523. is_segment = 1;
  524. duration = atof(ptr) * AV_TIME_BASE;
  525. } else if (av_strstart(line, "#EXT-X-BYTERANGE:", &ptr)) {
  526. seg_size = atoi(ptr);
  527. ptr = strchr(ptr, '@');
  528. if (ptr)
  529. seg_offset = atoi(ptr+1);
  530. } else if (av_strstart(line, "#", NULL)) {
  531. continue;
  532. } else if (line[0]) {
  533. if (is_variant) {
  534. if (!new_variant(c, &variant_info, line, url)) {
  535. ret = AVERROR(ENOMEM);
  536. goto fail;
  537. }
  538. is_variant = 0;
  539. }
  540. if (is_segment) {
  541. struct segment *seg;
  542. if (!pls) {
  543. if (!new_variant(c, 0, url, NULL)) {
  544. ret = AVERROR(ENOMEM);
  545. goto fail;
  546. }
  547. pls = c->playlists[c->n_playlists - 1];
  548. }
  549. seg = av_malloc(sizeof(struct segment));
  550. if (!seg) {
  551. ret = AVERROR(ENOMEM);
  552. goto fail;
  553. }
  554. seg->duration = duration;
  555. seg->key_type = key_type;
  556. if (has_iv) {
  557. memcpy(seg->iv, iv, sizeof(iv));
  558. } else {
  559. int seq = pls->start_seq_no + pls->n_segments;
  560. memset(seg->iv, 0, sizeof(seg->iv));
  561. AV_WB32(seg->iv + 12, seq);
  562. }
  563. ff_make_absolute_url(seg->key, sizeof(seg->key), url, key);
  564. ff_make_absolute_url(seg->url, sizeof(seg->url), url, line);
  565. dynarray_add(&pls->segments, &pls->n_segments, seg);
  566. is_segment = 0;
  567. seg->size = seg_size;
  568. if (seg_size >= 0) {
  569. seg->url_offset = seg_offset;
  570. seg_offset += seg_size;
  571. seg_size = -1;
  572. } else {
  573. seg->url_offset = 0;
  574. seg_offset = 0;
  575. }
  576. }
  577. }
  578. }
  579. if (pls)
  580. pls->last_load_time = av_gettime();
  581. fail:
  582. av_free(new_url);
  583. if (close_in)
  584. avio_close(in);
  585. return ret;
  586. }
  587. enum ReadFromURLMode {
  588. READ_NORMAL,
  589. READ_COMPLETE,
  590. };
  591. /* read from URLContext, limiting read to current segment */
  592. static int read_from_url(struct playlist *pls, uint8_t *buf, int buf_size,
  593. enum ReadFromURLMode mode)
  594. {
  595. int ret;
  596. struct segment *seg = pls->segments[pls->cur_seq_no - pls->start_seq_no];
  597. /* limit read if the segment was only a part of a file */
  598. if (seg->size >= 0)
  599. buf_size = FFMIN(buf_size, seg->size - pls->cur_seg_offset);
  600. if (mode == READ_COMPLETE)
  601. ret = ffurl_read_complete(pls->input, buf, buf_size);
  602. else
  603. ret = ffurl_read(pls->input, buf, buf_size);
  604. if (ret > 0)
  605. pls->cur_seg_offset += ret;
  606. return ret;
  607. }
  608. /* Parse the raw ID3 data and pass contents to caller */
  609. static void parse_id3(AVFormatContext *s, AVIOContext *pb,
  610. AVDictionary **metadata, int64_t *dts,
  611. ID3v2ExtraMetaAPIC **apic, ID3v2ExtraMeta **extra_meta)
  612. {
  613. static const char id3_priv_owner_ts[] = "com.apple.streaming.transportStreamTimestamp";
  614. ID3v2ExtraMeta *meta;
  615. ff_id3v2_read_dict(pb, metadata, ID3v2_DEFAULT_MAGIC, extra_meta);
  616. for (meta = *extra_meta; meta; meta = meta->next) {
  617. if (!strcmp(meta->tag, "PRIV")) {
  618. ID3v2ExtraMetaPRIV *priv = meta->data;
  619. if (priv->datasize == 8 && !strcmp(priv->owner, id3_priv_owner_ts)) {
  620. /* 33-bit MPEG timestamp */
  621. int64_t ts = AV_RB64(priv->data);
  622. av_log(s, AV_LOG_DEBUG, "HLS ID3 audio timestamp %"PRId64"\n", ts);
  623. if ((ts & ~((1ULL << 33) - 1)) == 0)
  624. *dts = ts;
  625. else
  626. av_log(s, AV_LOG_ERROR, "Invalid HLS ID3 audio timestamp %"PRId64"\n", ts);
  627. }
  628. } else if (!strcmp(meta->tag, "APIC") && apic)
  629. *apic = meta->data;
  630. }
  631. }
  632. /* Check if the ID3 metadata contents have changed */
  633. static int id3_has_changed_values(struct playlist *pls, AVDictionary *metadata,
  634. ID3v2ExtraMetaAPIC *apic)
  635. {
  636. AVDictionaryEntry *entry = NULL;
  637. AVDictionaryEntry *oldentry;
  638. /* check that no keys have changed values */
  639. while ((entry = av_dict_get(metadata, "", entry, AV_DICT_IGNORE_SUFFIX))) {
  640. oldentry = av_dict_get(pls->id3_initial, entry->key, NULL, AV_DICT_MATCH_CASE);
  641. if (!oldentry || strcmp(oldentry->value, entry->value) != 0)
  642. return 1;
  643. }
  644. /* check if apic appeared */
  645. if (apic && (pls->ctx->nb_streams != 2 || !pls->ctx->streams[1]->attached_pic.data))
  646. return 1;
  647. if (apic) {
  648. int size = pls->ctx->streams[1]->attached_pic.size;
  649. if (size != apic->buf->size - FF_INPUT_BUFFER_PADDING_SIZE)
  650. return 1;
  651. if (memcmp(apic->buf->data, pls->ctx->streams[1]->attached_pic.data, size) != 0)
  652. return 1;
  653. }
  654. return 0;
  655. }
  656. /* Parse ID3 data and handle the found data */
  657. static void handle_id3(AVIOContext *pb, struct playlist *pls)
  658. {
  659. AVDictionary *metadata = NULL;
  660. ID3v2ExtraMetaAPIC *apic = NULL;
  661. ID3v2ExtraMeta *extra_meta = NULL;
  662. int64_t timestamp = AV_NOPTS_VALUE;
  663. parse_id3(pls->ctx, pb, &metadata, &timestamp, &apic, &extra_meta);
  664. if (timestamp != AV_NOPTS_VALUE) {
  665. pls->id3_mpegts_timestamp = timestamp;
  666. pls->id3_offset = 0;
  667. }
  668. if (!pls->id3_found) {
  669. /* initial ID3 tags */
  670. av_assert0(!pls->id3_deferred_extra);
  671. pls->id3_found = 1;
  672. /* get picture attachment and set text metadata */
  673. if (pls->ctx->nb_streams)
  674. ff_id3v2_parse_apic(pls->ctx, &extra_meta);
  675. else
  676. /* demuxer not yet opened, defer picture attachment */
  677. pls->id3_deferred_extra = extra_meta;
  678. av_dict_copy(&pls->ctx->metadata, metadata, 0);
  679. pls->id3_initial = metadata;
  680. } else {
  681. if (!pls->id3_changed && id3_has_changed_values(pls, metadata, apic)) {
  682. avpriv_report_missing_feature(pls->ctx, "Changing ID3 metadata in HLS audio elementary stream");
  683. pls->id3_changed = 1;
  684. }
  685. av_dict_free(&metadata);
  686. }
  687. if (!pls->id3_deferred_extra)
  688. ff_id3v2_free_extra_meta(&extra_meta);
  689. }
  690. /* Intercept and handle ID3 tags between URLContext and AVIOContext */
  691. static void intercept_id3(struct playlist *pls, uint8_t *buf,
  692. int buf_size, int *len)
  693. {
  694. /* intercept id3 tags, we do not want to pass them to the raw
  695. * demuxer on all segment switches */
  696. int bytes;
  697. int id3_buf_pos = 0;
  698. int fill_buf = 0;
  699. /* gather all the id3 tags */
  700. while (1) {
  701. /* see if we can retrieve enough data for ID3 header */
  702. if (*len < ID3v2_HEADER_SIZE && buf_size >= ID3v2_HEADER_SIZE) {
  703. bytes = read_from_url(pls, buf + *len, ID3v2_HEADER_SIZE - *len, READ_COMPLETE);
  704. if (bytes > 0) {
  705. if (bytes == ID3v2_HEADER_SIZE - *len)
  706. /* no EOF yet, so fill the caller buffer again after
  707. * we have stripped the ID3 tags */
  708. fill_buf = 1;
  709. *len += bytes;
  710. } else if (*len <= 0) {
  711. /* error/EOF */
  712. *len = bytes;
  713. fill_buf = 0;
  714. }
  715. }
  716. if (*len < ID3v2_HEADER_SIZE)
  717. break;
  718. if (ff_id3v2_match(buf, ID3v2_DEFAULT_MAGIC)) {
  719. struct segment *seg = pls->segments[pls->cur_seq_no - pls->start_seq_no];
  720. int64_t segsize = seg->size >= 0 ? seg->size : ffurl_size(pls->input);
  721. int taglen = ff_id3v2_tag_len(buf);
  722. int tag_got_bytes = FFMIN(taglen, *len);
  723. int remaining = taglen - tag_got_bytes;
  724. if (taglen > segsize) {
  725. av_log(pls->ctx, AV_LOG_ERROR, "Too large HLS ID3 tag (%d vs %"PRId64")\n",
  726. taglen, segsize);
  727. break;
  728. }
  729. /*
  730. * Copy the id3 tag to our temporary id3 buffer.
  731. * We could read a small id3 tag directly without memcpy, but
  732. * we would still need to copy the large tags, and handling
  733. * both of those cases together with the possibility for multiple
  734. * tags would make the handling a bit complex.
  735. */
  736. pls->id3_buf = av_fast_realloc(pls->id3_buf, &pls->id3_buf_size, id3_buf_pos + taglen);
  737. if (!pls->id3_buf)
  738. break;
  739. memcpy(pls->id3_buf + id3_buf_pos, buf, tag_got_bytes);
  740. id3_buf_pos += tag_got_bytes;
  741. /* strip the intercepted bytes */
  742. *len -= tag_got_bytes;
  743. memmove(buf, buf + tag_got_bytes, *len);
  744. av_log(pls->ctx, AV_LOG_DEBUG, "Stripped %d HLS ID3 bytes\n", tag_got_bytes);
  745. if (remaining > 0) {
  746. /* read the rest of the tag in */
  747. if (read_from_url(pls, pls->id3_buf + id3_buf_pos, remaining, READ_COMPLETE) != remaining)
  748. break;
  749. id3_buf_pos += remaining;
  750. av_log(pls->ctx, AV_LOG_DEBUG, "Stripped additional %d HLS ID3 bytes\n", remaining);
  751. }
  752. } else {
  753. /* no more ID3 tags */
  754. break;
  755. }
  756. }
  757. /* re-fill buffer for the caller unless EOF */
  758. if (*len >= 0 && (fill_buf || *len == 0)) {
  759. bytes = read_from_url(pls, buf + *len, buf_size - *len, READ_NORMAL);
  760. /* ignore error if we already had some data */
  761. if (bytes >= 0)
  762. *len += bytes;
  763. else if (*len == 0)
  764. *len = bytes;
  765. }
  766. if (pls->id3_buf) {
  767. /* Now parse all the ID3 tags */
  768. AVIOContext id3ioctx;
  769. ffio_init_context(&id3ioctx, pls->id3_buf, id3_buf_pos, 0, NULL, NULL, NULL, NULL);
  770. handle_id3(&id3ioctx, pls);
  771. }
  772. if (pls->is_id3_timestamped == -1)
  773. pls->is_id3_timestamped = (pls->id3_mpegts_timestamp != AV_NOPTS_VALUE);
  774. }
  775. static int open_input(HLSContext *c, struct playlist *pls)
  776. {
  777. AVDictionary *opts = NULL;
  778. AVDictionary *opts2 = NULL;
  779. int ret;
  780. struct segment *seg = pls->segments[pls->cur_seq_no - pls->start_seq_no];
  781. // broker prior HTTP options that should be consistent across requests
  782. av_dict_set(&opts, "user-agent", c->user_agent, 0);
  783. av_dict_set(&opts, "cookies", c->cookies, 0);
  784. av_dict_set(&opts, "headers", c->headers, 0);
  785. av_dict_set(&opts, "seekable", "0", 0);
  786. // Same opts for key request (ffurl_open mutilates the opts so it cannot be used twice)
  787. av_dict_copy(&opts2, opts, 0);
  788. if (seg->size >= 0) {
  789. /* try to restrict the HTTP request to the part we want
  790. * (if this is in fact a HTTP request) */
  791. char offset[24] = { 0 };
  792. char end_offset[24] = { 0 };
  793. snprintf(offset, sizeof(offset) - 1, "%"PRId64,
  794. seg->url_offset);
  795. snprintf(end_offset, sizeof(end_offset) - 1, "%"PRId64,
  796. seg->url_offset + seg->size);
  797. av_dict_set(&opts, "offset", offset, 0);
  798. av_dict_set(&opts, "end_offset", end_offset, 0);
  799. }
  800. av_log(pls->parent, AV_LOG_VERBOSE, "HLS request for url '%s', offset %"PRId64", playlist %d\n",
  801. seg->url, seg->url_offset, pls->index);
  802. if (seg->key_type == KEY_NONE) {
  803. ret = ffurl_open(&pls->input, seg->url, AVIO_FLAG_READ,
  804. &pls->parent->interrupt_callback, &opts);
  805. } else if (seg->key_type == KEY_AES_128) {
  806. char iv[33], key[33], url[MAX_URL_SIZE];
  807. if (strcmp(seg->key, pls->key_url)) {
  808. URLContext *uc;
  809. if (ffurl_open(&uc, seg->key, AVIO_FLAG_READ,
  810. &pls->parent->interrupt_callback, &opts2) == 0) {
  811. if (ffurl_read_complete(uc, pls->key, sizeof(pls->key))
  812. != sizeof(pls->key)) {
  813. av_log(NULL, AV_LOG_ERROR, "Unable to read key file %s\n",
  814. seg->key);
  815. }
  816. ffurl_close(uc);
  817. } else {
  818. av_log(NULL, AV_LOG_ERROR, "Unable to open key file %s\n",
  819. seg->key);
  820. }
  821. av_strlcpy(pls->key_url, seg->key, sizeof(pls->key_url));
  822. }
  823. ff_data_to_hex(iv, seg->iv, sizeof(seg->iv), 0);
  824. ff_data_to_hex(key, pls->key, sizeof(pls->key), 0);
  825. iv[32] = key[32] = '\0';
  826. if (strstr(seg->url, "://"))
  827. snprintf(url, sizeof(url), "crypto+%s", seg->url);
  828. else
  829. snprintf(url, sizeof(url), "crypto:%s", seg->url);
  830. if ((ret = ffurl_alloc(&pls->input, url, AVIO_FLAG_READ,
  831. &pls->parent->interrupt_callback)) < 0)
  832. goto cleanup;
  833. av_opt_set(pls->input->priv_data, "key", key, 0);
  834. av_opt_set(pls->input->priv_data, "iv", iv, 0);
  835. if ((ret = ffurl_connect(pls->input, &opts)) < 0) {
  836. ffurl_close(pls->input);
  837. pls->input = NULL;
  838. goto cleanup;
  839. }
  840. ret = 0;
  841. }
  842. else
  843. ret = AVERROR(ENOSYS);
  844. /* Seek to the requested position. If this was a HTTP request, the offset
  845. * should already be where want it to, but this allows e.g. local testing
  846. * without a HTTP server. */
  847. if (ret == 0) {
  848. int seekret = ffurl_seek(pls->input, seg->url_offset, SEEK_SET);
  849. if (seekret < 0) {
  850. av_log(pls->parent, AV_LOG_ERROR, "Unable to seek to offset %"PRId64" of HLS segment '%s'\n", seg->url_offset, seg->url);
  851. ret = seekret;
  852. ffurl_close(pls->input);
  853. pls->input = NULL;
  854. }
  855. }
  856. cleanup:
  857. av_dict_free(&opts);
  858. av_dict_free(&opts2);
  859. pls->cur_seg_offset = 0;
  860. return ret;
  861. }
  862. static int64_t default_reload_interval(struct playlist *pls)
  863. {
  864. return pls->n_segments > 0 ?
  865. pls->segments[pls->n_segments - 1]->duration :
  866. pls->target_duration;
  867. }
  868. static int read_data(void *opaque, uint8_t *buf, int buf_size)
  869. {
  870. struct playlist *v = opaque;
  871. HLSContext *c = v->parent->priv_data;
  872. int ret, i;
  873. int just_opened = 0;
  874. restart:
  875. if (!v->needed)
  876. return AVERROR_EOF;
  877. if (!v->input) {
  878. int64_t reload_interval;
  879. /* Check that the playlist is still needed before opening a new
  880. * segment. */
  881. if (v->ctx && v->ctx->nb_streams &&
  882. v->parent->nb_streams >= v->stream_offset + v->ctx->nb_streams) {
  883. v->needed = 0;
  884. for (i = v->stream_offset; i < v->stream_offset + v->ctx->nb_streams;
  885. i++) {
  886. if (v->parent->streams[i]->discard < AVDISCARD_ALL)
  887. v->needed = 1;
  888. }
  889. }
  890. if (!v->needed) {
  891. av_log(v->parent, AV_LOG_INFO, "No longer receiving playlist %d\n",
  892. v->index);
  893. return AVERROR_EOF;
  894. }
  895. /* If this is a live stream and the reload interval has elapsed since
  896. * the last playlist reload, reload the playlists now. */
  897. reload_interval = default_reload_interval(v);
  898. reload:
  899. if (!v->finished &&
  900. av_gettime() - v->last_load_time >= reload_interval) {
  901. if ((ret = parse_playlist(c, v->url, v, NULL)) < 0) {
  902. av_log(v->parent, AV_LOG_WARNING, "Failed to reload playlist %d\n",
  903. v->index);
  904. return ret;
  905. }
  906. /* If we need to reload the playlist again below (if
  907. * there's still no more segments), switch to a reload
  908. * interval of half the target duration. */
  909. reload_interval = v->target_duration / 2;
  910. }
  911. if (v->cur_seq_no < v->start_seq_no) {
  912. av_log(NULL, AV_LOG_WARNING,
  913. "skipping %d segments ahead, expired from playlists\n",
  914. v->start_seq_no - v->cur_seq_no);
  915. v->cur_seq_no = v->start_seq_no;
  916. }
  917. if (v->cur_seq_no >= v->start_seq_no + v->n_segments) {
  918. if (v->finished)
  919. return AVERROR_EOF;
  920. while (av_gettime() - v->last_load_time < reload_interval) {
  921. if (ff_check_interrupt(c->interrupt_callback))
  922. return AVERROR_EXIT;
  923. av_usleep(100*1000);
  924. }
  925. /* Enough time has elapsed since the last reload */
  926. goto reload;
  927. }
  928. ret = open_input(c, v);
  929. if (ret < 0) {
  930. av_log(v->parent, AV_LOG_WARNING, "Failed to open segment of playlist %d\n",
  931. v->index);
  932. return ret;
  933. }
  934. just_opened = 1;
  935. }
  936. ret = read_from_url(v, buf, buf_size, READ_NORMAL);
  937. if (ret > 0) {
  938. if (just_opened && v->is_id3_timestamped != 0) {
  939. /* Intercept ID3 tags here, elementary audio streams are required
  940. * to convey timestamps using them in the beginning of each segment. */
  941. intercept_id3(v, buf, buf_size, &ret);
  942. }
  943. return ret;
  944. }
  945. ffurl_close(v->input);
  946. v->input = NULL;
  947. v->cur_seq_no++;
  948. c->cur_seq_no = v->cur_seq_no;
  949. goto restart;
  950. }
  951. static int playlist_in_multiple_variants(HLSContext *c, struct playlist *pls)
  952. {
  953. int variant_count = 0;
  954. int i, j;
  955. for (i = 0; i < c->n_variants && variant_count < 2; i++) {
  956. struct variant *v = c->variants[i];
  957. for (j = 0; j < v->n_playlists; j++) {
  958. if (v->playlists[j] == pls) {
  959. variant_count++;
  960. break;
  961. }
  962. }
  963. }
  964. return variant_count >= 2;
  965. }
  966. static void add_renditions_to_variant(HLSContext *c, struct variant *var,
  967. enum AVMediaType type, const char *group_id)
  968. {
  969. int i;
  970. for (i = 0; i < c->n_renditions; i++) {
  971. struct rendition *rend = c->renditions[i];
  972. if (rend->type == type && !strcmp(rend->group_id, group_id)) {
  973. if (rend->playlist)
  974. /* rendition is an external playlist
  975. * => add the playlist to the variant */
  976. dynarray_add(&var->playlists, &var->n_playlists, rend->playlist);
  977. else
  978. /* rendition is part of the variant main Media Playlist
  979. * => add the rendition to the main Media Playlist */
  980. dynarray_add(&var->playlists[0]->renditions,
  981. &var->playlists[0]->n_renditions,
  982. rend);
  983. }
  984. }
  985. }
  986. static void add_metadata_from_renditions(AVFormatContext *s, struct playlist *pls,
  987. enum AVMediaType type)
  988. {
  989. int rend_idx = 0;
  990. int i;
  991. for (i = 0; i < pls->ctx->nb_streams; i++) {
  992. AVStream *st = s->streams[pls->stream_offset + i];
  993. if (st->codec->codec_type != type)
  994. continue;
  995. for (; rend_idx < pls->n_renditions; rend_idx++) {
  996. struct rendition *rend = pls->renditions[rend_idx];
  997. if (rend->type != type)
  998. continue;
  999. if (rend->language[0])
  1000. av_dict_set(&st->metadata, "language", rend->language, 0);
  1001. if (rend->name[0])
  1002. av_dict_set(&st->metadata, "comment", rend->name, 0);
  1003. st->disposition |= rend->disposition;
  1004. }
  1005. if (rend_idx >=pls->n_renditions)
  1006. break;
  1007. }
  1008. }
  1009. /* if timestamp was in valid range: returns 1 and sets seq_no
  1010. * if not: returns 0 and sets seq_no to closest segment */
  1011. static int find_timestamp_in_playlist(HLSContext *c, struct playlist *pls,
  1012. int64_t timestamp, int *seq_no)
  1013. {
  1014. int i;
  1015. int64_t pos = c->first_timestamp == AV_NOPTS_VALUE ?
  1016. 0 : c->first_timestamp;
  1017. if (timestamp < pos) {
  1018. *seq_no = pls->start_seq_no;
  1019. return 0;
  1020. }
  1021. for (i = 0; i < pls->n_segments; i++) {
  1022. int64_t diff = pos + pls->segments[i]->duration - timestamp;
  1023. if (diff > 0) {
  1024. *seq_no = pls->start_seq_no + i;
  1025. return 1;
  1026. }
  1027. pos += pls->segments[i]->duration;
  1028. }
  1029. *seq_no = pls->start_seq_no + pls->n_segments - 1;
  1030. return 0;
  1031. }
  1032. static int select_cur_seq_no(HLSContext *c, struct playlist *pls)
  1033. {
  1034. int seq_no;
  1035. if (!pls->finished && !c->first_packet &&
  1036. av_gettime() - pls->last_load_time >= default_reload_interval(pls))
  1037. /* reload the playlist since it was suspended */
  1038. parse_playlist(c, pls->url, pls, NULL);
  1039. /* If playback is already in progress (we are just selecting a new
  1040. * playlist) and this is a complete file, find the matching segment
  1041. * by counting durations. */
  1042. if (pls->finished && c->cur_timestamp != AV_NOPTS_VALUE) {
  1043. find_timestamp_in_playlist(c, pls, c->cur_timestamp, &seq_no);
  1044. return seq_no;
  1045. }
  1046. if (!pls->finished) {
  1047. if (!c->first_packet && /* we are doing a segment selection during playback */
  1048. c->cur_seq_no >= pls->start_seq_no &&
  1049. c->cur_seq_no < pls->start_seq_no + pls->n_segments)
  1050. /* While spec 3.4.3 says that we cannot assume anything about the
  1051. * content at the same sequence number on different playlists,
  1052. * in practice this seems to work and doing it otherwise would
  1053. * require us to download a segment to inspect its timestamps. */
  1054. return c->cur_seq_no;
  1055. /* If this is a live stream with more than 3 segments, start at the
  1056. * third last segment. */
  1057. if (pls->n_segments > 3)
  1058. return pls->start_seq_no + pls->n_segments - 3;
  1059. }
  1060. /* Otherwise just start on the first segment. */
  1061. return pls->start_seq_no;
  1062. }
  1063. static int hls_read_header(AVFormatContext *s)
  1064. {
  1065. URLContext *u = (s->flags & AVFMT_FLAG_CUSTOM_IO) ? NULL : s->pb->opaque;
  1066. HLSContext *c = s->priv_data;
  1067. int ret = 0, i, j, stream_offset = 0;
  1068. c->interrupt_callback = &s->interrupt_callback;
  1069. c->first_packet = 1;
  1070. c->first_timestamp = AV_NOPTS_VALUE;
  1071. c->cur_timestamp = AV_NOPTS_VALUE;
  1072. // if the URL context is good, read important options we must broker later
  1073. if (u && u->prot->priv_data_class) {
  1074. // get the previous user agent & set back to null if string size is zero
  1075. av_freep(&c->user_agent);
  1076. av_opt_get(u->priv_data, "user-agent", 0, (uint8_t**)&(c->user_agent));
  1077. if (c->user_agent && !strlen(c->user_agent))
  1078. av_freep(&c->user_agent);
  1079. // get the previous cookies & set back to null if string size is zero
  1080. av_freep(&c->cookies);
  1081. av_opt_get(u->priv_data, "cookies", 0, (uint8_t**)&(c->cookies));
  1082. if (c->cookies && !strlen(c->cookies))
  1083. av_freep(&c->cookies);
  1084. // get the previous headers & set back to null if string size is zero
  1085. av_freep(&c->headers);
  1086. av_opt_get(u->priv_data, "headers", 0, (uint8_t**)&(c->headers));
  1087. if (c->headers && !strlen(c->headers))
  1088. av_freep(&c->headers);
  1089. }
  1090. if ((ret = parse_playlist(c, s->filename, NULL, s->pb)) < 0)
  1091. goto fail;
  1092. if (c->n_variants == 0) {
  1093. av_log(NULL, AV_LOG_WARNING, "Empty playlist\n");
  1094. ret = AVERROR_EOF;
  1095. goto fail;
  1096. }
  1097. /* If the playlist only contained playlists (Master Playlist),
  1098. * parse each individual playlist. */
  1099. if (c->n_playlists > 1 || c->playlists[0]->n_segments == 0) {
  1100. for (i = 0; i < c->n_playlists; i++) {
  1101. struct playlist *pls = c->playlists[i];
  1102. if ((ret = parse_playlist(c, pls->url, pls, NULL)) < 0)
  1103. goto fail;
  1104. }
  1105. }
  1106. if (c->variants[0]->playlists[0]->n_segments == 0) {
  1107. av_log(NULL, AV_LOG_WARNING, "Empty playlist\n");
  1108. ret = AVERROR_EOF;
  1109. goto fail;
  1110. }
  1111. /* If this isn't a live stream, calculate the total duration of the
  1112. * stream. */
  1113. if (c->variants[0]->playlists[0]->finished) {
  1114. int64_t duration = 0;
  1115. for (i = 0; i < c->variants[0]->playlists[0]->n_segments; i++)
  1116. duration += c->variants[0]->playlists[0]->segments[i]->duration;
  1117. s->duration = duration;
  1118. }
  1119. /* Associate renditions with variants */
  1120. for (i = 0; i < c->n_variants; i++) {
  1121. struct variant *var = c->variants[i];
  1122. if (var->audio_group[0])
  1123. add_renditions_to_variant(c, var, AVMEDIA_TYPE_AUDIO, var->audio_group);
  1124. if (var->video_group[0])
  1125. add_renditions_to_variant(c, var, AVMEDIA_TYPE_VIDEO, var->video_group);
  1126. if (var->subtitles_group[0])
  1127. add_renditions_to_variant(c, var, AVMEDIA_TYPE_SUBTITLE, var->subtitles_group);
  1128. }
  1129. /* Open the demuxer for each playlist */
  1130. for (i = 0; i < c->n_playlists; i++) {
  1131. struct playlist *pls = c->playlists[i];
  1132. AVInputFormat *in_fmt = NULL;
  1133. if (pls->n_segments == 0)
  1134. continue;
  1135. if (!(pls->ctx = avformat_alloc_context())) {
  1136. ret = AVERROR(ENOMEM);
  1137. goto fail;
  1138. }
  1139. pls->index = i;
  1140. pls->needed = 1;
  1141. pls->parent = s;
  1142. pls->cur_seq_no = select_cur_seq_no(c, pls);
  1143. pls->read_buffer = av_malloc(INITIAL_BUFFER_SIZE);
  1144. ffio_init_context(&pls->pb, pls->read_buffer, INITIAL_BUFFER_SIZE, 0, pls,
  1145. read_data, NULL, NULL);
  1146. pls->pb.seekable = 0;
  1147. ret = av_probe_input_buffer(&pls->pb, &in_fmt, pls->segments[0]->url,
  1148. NULL, 0, 0);
  1149. if (ret < 0) {
  1150. /* Free the ctx - it isn't initialized properly at this point,
  1151. * so avformat_close_input shouldn't be called. If
  1152. * avformat_open_input fails below, it frees and zeros the
  1153. * context, so it doesn't need any special treatment like this. */
  1154. av_log(s, AV_LOG_ERROR, "Error when loading first segment '%s'\n", pls->segments[0]->url);
  1155. avformat_free_context(pls->ctx);
  1156. pls->ctx = NULL;
  1157. goto fail;
  1158. }
  1159. pls->ctx->pb = &pls->pb;
  1160. pls->stream_offset = stream_offset;
  1161. ret = avformat_open_input(&pls->ctx, pls->segments[0]->url, in_fmt, NULL);
  1162. if (ret < 0)
  1163. goto fail;
  1164. if (pls->id3_deferred_extra && pls->ctx->nb_streams == 1) {
  1165. ff_id3v2_parse_apic(pls->ctx, &pls->id3_deferred_extra);
  1166. avformat_queue_attached_pictures(pls->ctx);
  1167. ff_id3v2_free_extra_meta(&pls->id3_deferred_extra);
  1168. pls->id3_deferred_extra = NULL;
  1169. }
  1170. pls->ctx->ctx_flags &= ~AVFMTCTX_NOHEADER;
  1171. ret = avformat_find_stream_info(pls->ctx, NULL);
  1172. if (ret < 0)
  1173. goto fail;
  1174. if (pls->is_id3_timestamped == -1)
  1175. av_log(s, AV_LOG_WARNING, "No expected HTTP requests have been made\n");
  1176. /* Create new AVStreams for each stream in this playlist */
  1177. for (j = 0; j < pls->ctx->nb_streams; j++) {
  1178. AVStream *st = avformat_new_stream(s, NULL);
  1179. AVStream *ist = pls->ctx->streams[j];
  1180. if (!st) {
  1181. ret = AVERROR(ENOMEM);
  1182. goto fail;
  1183. }
  1184. st->id = i;
  1185. avcodec_copy_context(st->codec, pls->ctx->streams[j]->codec);
  1186. if (pls->is_id3_timestamped) /* custom timestamps via id3 */
  1187. avpriv_set_pts_info(st, 33, 1, MPEG_TIME_BASE);
  1188. else
  1189. avpriv_set_pts_info(st, ist->pts_wrap_bits, ist->time_base.num, ist->time_base.den);
  1190. }
  1191. add_metadata_from_renditions(s, pls, AVMEDIA_TYPE_AUDIO);
  1192. add_metadata_from_renditions(s, pls, AVMEDIA_TYPE_VIDEO);
  1193. add_metadata_from_renditions(s, pls, AVMEDIA_TYPE_SUBTITLE);
  1194. stream_offset += pls->ctx->nb_streams;
  1195. }
  1196. /* Create a program for each variant */
  1197. for (i = 0; i < c->n_variants; i++) {
  1198. struct variant *v = c->variants[i];
  1199. char bitrate_str[20];
  1200. AVProgram *program;
  1201. snprintf(bitrate_str, sizeof(bitrate_str), "%d", v->bandwidth);
  1202. program = av_new_program(s, i);
  1203. if (!program)
  1204. goto fail;
  1205. av_dict_set(&program->metadata, "variant_bitrate", bitrate_str, 0);
  1206. for (j = 0; j < v->n_playlists; j++) {
  1207. struct playlist *pls = v->playlists[j];
  1208. int is_shared = playlist_in_multiple_variants(c, pls);
  1209. int k;
  1210. for (k = 0; k < pls->ctx->nb_streams; k++) {
  1211. struct AVStream *st = s->streams[pls->stream_offset + k];
  1212. ff_program_add_stream_index(s, i, pls->stream_offset + k);
  1213. /* Set variant_bitrate for streams unique to this variant */
  1214. if (!is_shared && v->bandwidth)
  1215. av_dict_set(&st->metadata, "variant_bitrate", bitrate_str, 0);
  1216. }
  1217. }
  1218. }
  1219. return 0;
  1220. fail:
  1221. free_playlist_list(c);
  1222. free_variant_list(c);
  1223. free_rendition_list(c);
  1224. return ret;
  1225. }
  1226. static int recheck_discard_flags(AVFormatContext *s, int first)
  1227. {
  1228. HLSContext *c = s->priv_data;
  1229. int i, changed = 0;
  1230. /* Check if any new streams are needed */
  1231. for (i = 0; i < c->n_playlists; i++)
  1232. c->playlists[i]->cur_needed = 0;
  1233. for (i = 0; i < s->nb_streams; i++) {
  1234. AVStream *st = s->streams[i];
  1235. struct playlist *pls = c->playlists[s->streams[i]->id];
  1236. if (st->discard < AVDISCARD_ALL)
  1237. pls->cur_needed = 1;
  1238. }
  1239. for (i = 0; i < c->n_playlists; i++) {
  1240. struct playlist *pls = c->playlists[i];
  1241. if (pls->cur_needed && !pls->needed) {
  1242. pls->needed = 1;
  1243. changed = 1;
  1244. pls->cur_seq_no = select_cur_seq_no(c, pls);
  1245. pls->pb.eof_reached = 0;
  1246. if (c->cur_timestamp != AV_NOPTS_VALUE) {
  1247. /* catch up */
  1248. pls->seek_timestamp = c->cur_timestamp;
  1249. pls->seek_flags = AVSEEK_FLAG_ANY;
  1250. pls->seek_stream_index = -1;
  1251. }
  1252. av_log(s, AV_LOG_INFO, "Now receiving playlist %d, segment %d\n", i, pls->cur_seq_no);
  1253. } else if (first && !pls->cur_needed && pls->needed) {
  1254. if (pls->input)
  1255. ffurl_close(pls->input);
  1256. pls->input = NULL;
  1257. pls->needed = 0;
  1258. changed = 1;
  1259. av_log(s, AV_LOG_INFO, "No longer receiving playlist %d\n", i);
  1260. }
  1261. }
  1262. return changed;
  1263. }
  1264. static void fill_timing_for_id3_timestamped_stream(struct playlist *pls)
  1265. {
  1266. if (pls->id3_offset >= 0) {
  1267. pls->pkt.dts = pls->id3_mpegts_timestamp +
  1268. av_rescale_q(pls->id3_offset,
  1269. pls->ctx->streams[pls->pkt.stream_index]->time_base,
  1270. MPEG_TIME_BASE_Q);
  1271. if (pls->pkt.duration)
  1272. pls->id3_offset += pls->pkt.duration;
  1273. else
  1274. pls->id3_offset = -1;
  1275. } else {
  1276. /* there have been packets with unknown duration
  1277. * since the last id3 tag, should not normally happen */
  1278. pls->pkt.dts = AV_NOPTS_VALUE;
  1279. }
  1280. if (pls->pkt.duration)
  1281. pls->pkt.duration = av_rescale_q(pls->pkt.duration,
  1282. pls->ctx->streams[pls->pkt.stream_index]->time_base,
  1283. MPEG_TIME_BASE_Q);
  1284. pls->pkt.pts = AV_NOPTS_VALUE;
  1285. }
  1286. static AVRational get_timebase(struct playlist *pls)
  1287. {
  1288. if (pls->is_id3_timestamped)
  1289. return MPEG_TIME_BASE_Q;
  1290. return pls->ctx->streams[pls->pkt.stream_index]->time_base;
  1291. }
  1292. static int compare_ts_with_wrapdetect(int64_t ts_a, struct playlist *pls_a,
  1293. int64_t ts_b, struct playlist *pls_b)
  1294. {
  1295. int64_t scaled_ts_a = av_rescale_q(ts_a, get_timebase(pls_a), MPEG_TIME_BASE_Q);
  1296. int64_t scaled_ts_b = av_rescale_q(ts_b, get_timebase(pls_b), MPEG_TIME_BASE_Q);
  1297. return av_compare_mod(scaled_ts_a, scaled_ts_b, 1LL << 33);
  1298. }
  1299. static int hls_read_packet(AVFormatContext *s, AVPacket *pkt)
  1300. {
  1301. HLSContext *c = s->priv_data;
  1302. int ret, i, minplaylist = -1;
  1303. recheck_discard_flags(s, c->first_packet);
  1304. for (i = 0; i < c->n_playlists; i++) {
  1305. struct playlist *pls = c->playlists[i];
  1306. /* Make sure we've got one buffered packet from each open playlist
  1307. * stream */
  1308. if (pls->needed && !pls->pkt.data) {
  1309. while (1) {
  1310. int64_t ts_diff;
  1311. AVRational tb;
  1312. ret = av_read_frame(pls->ctx, &pls->pkt);
  1313. if (ret < 0) {
  1314. if (!url_feof(&pls->pb) && ret != AVERROR_EOF)
  1315. return ret;
  1316. reset_packet(&pls->pkt);
  1317. break;
  1318. } else {
  1319. /* stream_index check prevents matching picture attachments etc. */
  1320. if (pls->is_id3_timestamped && pls->pkt.stream_index == 0) {
  1321. /* audio elementary streams are id3 timestamped */
  1322. fill_timing_for_id3_timestamped_stream(pls);
  1323. }
  1324. if (c->first_timestamp == AV_NOPTS_VALUE &&
  1325. pls->pkt.dts != AV_NOPTS_VALUE)
  1326. c->first_timestamp = av_rescale_q(pls->pkt.dts,
  1327. get_timebase(pls), AV_TIME_BASE_Q);
  1328. }
  1329. if (pls->seek_timestamp == AV_NOPTS_VALUE)
  1330. break;
  1331. if (pls->seek_stream_index < 0 ||
  1332. pls->seek_stream_index == pls->pkt.stream_index) {
  1333. if (pls->pkt.dts == AV_NOPTS_VALUE) {
  1334. pls->seek_timestamp = AV_NOPTS_VALUE;
  1335. break;
  1336. }
  1337. tb = get_timebase(pls);
  1338. ts_diff = av_rescale_rnd(pls->pkt.dts, AV_TIME_BASE,
  1339. tb.den, AV_ROUND_DOWN) -
  1340. pls->seek_timestamp;
  1341. if (ts_diff >= 0 && (pls->seek_flags & AVSEEK_FLAG_ANY ||
  1342. pls->pkt.flags & AV_PKT_FLAG_KEY)) {
  1343. pls->seek_timestamp = AV_NOPTS_VALUE;
  1344. break;
  1345. }
  1346. }
  1347. av_free_packet(&pls->pkt);
  1348. reset_packet(&pls->pkt);
  1349. }
  1350. }
  1351. /* Check if this stream has the packet with the lowest dts */
  1352. if (pls->pkt.data) {
  1353. struct playlist *minpls = minplaylist < 0 ?
  1354. NULL : c->playlists[minplaylist];
  1355. if (minplaylist < 0) {
  1356. minplaylist = i;
  1357. } else {
  1358. int64_t dts = pls->pkt.dts;
  1359. int64_t mindts = minpls->pkt.dts;
  1360. if (dts == AV_NOPTS_VALUE ||
  1361. (mindts != AV_NOPTS_VALUE && compare_ts_with_wrapdetect(dts, pls, mindts, minpls) < 0))
  1362. minplaylist = i;
  1363. }
  1364. }
  1365. }
  1366. /* If we got a packet, return it */
  1367. if (minplaylist >= 0) {
  1368. struct playlist *pls = c->playlists[minplaylist];
  1369. *pkt = pls->pkt;
  1370. pkt->stream_index += pls->stream_offset;
  1371. reset_packet(&c->playlists[minplaylist]->pkt);
  1372. if (pkt->dts != AV_NOPTS_VALUE)
  1373. c->cur_timestamp = av_rescale_q(pkt->dts,
  1374. pls->ctx->streams[pls->pkt.stream_index]->time_base,
  1375. AV_TIME_BASE_Q);
  1376. return 0;
  1377. }
  1378. return AVERROR_EOF;
  1379. }
  1380. static int hls_close(AVFormatContext *s)
  1381. {
  1382. HLSContext *c = s->priv_data;
  1383. free_playlist_list(c);
  1384. free_variant_list(c);
  1385. free_rendition_list(c);
  1386. return 0;
  1387. }
  1388. static int hls_read_seek(AVFormatContext *s, int stream_index,
  1389. int64_t timestamp, int flags)
  1390. {
  1391. HLSContext *c = s->priv_data;
  1392. struct playlist *seek_pls = NULL;
  1393. int i, seq_no;
  1394. int64_t first_timestamp, seek_timestamp, duration;
  1395. if ((flags & AVSEEK_FLAG_BYTE) ||
  1396. !(c->variants[0]->playlists[0]->finished || c->variants[0]->playlists[0]->type == PLS_TYPE_EVENT))
  1397. return AVERROR(ENOSYS);
  1398. first_timestamp = c->first_timestamp == AV_NOPTS_VALUE ?
  1399. 0 : c->first_timestamp;
  1400. seek_timestamp = av_rescale_rnd(timestamp, AV_TIME_BASE,
  1401. s->streams[stream_index]->time_base.den,
  1402. flags & AVSEEK_FLAG_BACKWARD ?
  1403. AV_ROUND_DOWN : AV_ROUND_UP);
  1404. duration = s->duration == AV_NOPTS_VALUE ?
  1405. 0 : s->duration;
  1406. if (0 < duration && duration < seek_timestamp - first_timestamp)
  1407. return AVERROR(EIO);
  1408. /* find the playlist with the specified stream */
  1409. for (i = 0; i < c->n_playlists; i++) {
  1410. struct playlist *pls = c->playlists[i];
  1411. if (stream_index >= pls->stream_offset &&
  1412. stream_index - pls->stream_offset < pls->ctx->nb_streams) {
  1413. seek_pls = pls;
  1414. break;
  1415. }
  1416. }
  1417. /* check if the timestamp is valid for the playlist with the
  1418. * specified stream index */
  1419. if (!seek_pls || !find_timestamp_in_playlist(c, seek_pls, seek_timestamp, &seq_no))
  1420. return AVERROR(EIO);
  1421. /* set segment now so we do not need to search again below */
  1422. seek_pls->cur_seq_no = seq_no;
  1423. seek_pls->seek_stream_index = stream_index - seek_pls->stream_offset;
  1424. for (i = 0; i < c->n_playlists; i++) {
  1425. /* Reset reading */
  1426. struct playlist *pls = c->playlists[i];
  1427. if (pls->input) {
  1428. ffurl_close(pls->input);
  1429. pls->input = NULL;
  1430. }
  1431. av_free_packet(&pls->pkt);
  1432. reset_packet(&pls->pkt);
  1433. pls->pb.eof_reached = 0;
  1434. /* Clear any buffered data */
  1435. pls->pb.buf_end = pls->pb.buf_ptr = pls->pb.buffer;
  1436. /* Reset the pos, to let the mpegts demuxer know we've seeked. */
  1437. pls->pb.pos = 0;
  1438. /* Flush the packet queue of the subdemuxer. */
  1439. ff_read_frame_flush(pls->ctx);
  1440. pls->seek_timestamp = seek_timestamp;
  1441. pls->seek_flags = flags;
  1442. if (pls != seek_pls) {
  1443. /* set closest segment seq_no for playlists not handled above */
  1444. find_timestamp_in_playlist(c, pls, seek_timestamp, &pls->cur_seq_no);
  1445. /* seek the playlist to the given position without taking
  1446. * keyframes into account since this playlist does not have the
  1447. * specified stream where we should look for the keyframes */
  1448. pls->seek_stream_index = -1;
  1449. pls->seek_flags |= AVSEEK_FLAG_ANY;
  1450. }
  1451. }
  1452. c->cur_timestamp = seek_timestamp;
  1453. return 0;
  1454. }
  1455. static int hls_probe(AVProbeData *p)
  1456. {
  1457. /* Require #EXTM3U at the start, and either one of the ones below
  1458. * somewhere for a proper match. */
  1459. if (strncmp(p->buf, "#EXTM3U", 7))
  1460. return 0;
  1461. if (strstr(p->buf, "#EXT-X-STREAM-INF:") ||
  1462. strstr(p->buf, "#EXT-X-TARGETDURATION:") ||
  1463. strstr(p->buf, "#EXT-X-MEDIA-SEQUENCE:"))
  1464. return AVPROBE_SCORE_MAX;
  1465. return 0;
  1466. }
  1467. AVInputFormat ff_hls_demuxer = {
  1468. .name = "hls,applehttp",
  1469. .long_name = NULL_IF_CONFIG_SMALL("Apple HTTP Live Streaming"),
  1470. .priv_data_size = sizeof(HLSContext),
  1471. .read_probe = hls_probe,
  1472. .read_header = hls_read_header,
  1473. .read_packet = hls_read_packet,
  1474. .read_close = hls_close,
  1475. .read_seek = hls_read_seek,
  1476. };