Message ID | f8e7e935-4b11-c596-d55b-2f1836b28144@gentoo.org |
---|---|
State | New |
Headers | show |
Series | [FFmpeg-devel] avformat/file: Fix handing of file URIs | expand |
Context | Check | Description |
---|---|---|
andriy/configure_x86 | warning | Failed to apply patch |
yinshiyou/configure_loongarch64 | warning | Failed to apply patch |
Nick Sarnie (12022-11-24): > The current URI handling only removes the file: prefix, but we also need to > consider the case of percent encoding. > Percent encoding can happen with non-ASCII characters in the path. NAK, this is a huge compatibility break. But this fix is necessary, you are right. What needs to happen to handle this correctly: - Introduce a “fs:” protocol as a synonym to file:. - Change “file:” (but not “fs:”!) to detect if the rest of the path looks like a real file URL or a raw file path. - If it looks like a real file URL, de-percent-escape it. - If it looks like a raw file path, treat it the legacy way after printing a warning. - In a few years, remove the heuristics and always handle “file:” in the standards-compliant way. I believe we can ignore incomplete / relative file:// URLs in the transition period. The triple / would be a very reliable way of identification. Regards,
Okay, thanks. I tried asking if this patch would be accepted in IRC before posting this to prevent this kind of situation but nobody responded. I don't know enough about the codebase to implement the suggested solution, so I'll let you all decide if you want to do it.
diff --git a/libavformat/file.c b/libavformat/file.c index 6103c37b34..416e7a0b6b 100644 --- a/libavformat/file.c +++ b/libavformat/file.c @@ -21,6 +21,7 @@ #include "config_components.h" +#include "libavformat/urldecode.h" #include "libavutil/avstring.h" #include "libavutil/file_open.h" #include "libavutil/internal.h" @@ -142,7 +143,9 @@ static int file_check(URLContext *h, int mask) int ret = 0;
The current URI handling only removes the file: prefix, but we also need to consider the case of percent encoding. Percent encoding can happen with non-ASCII characters in the path. I hit this through mpv with vobsub subtitles when dragging subtitles onto the player using the mouse. Vobsub uses two files: an .idx file and a .sub file. mpv decodes the file path for direct inputs to mpv (the .idx file here), so that's why mpv works for most things today. When passed the .idx file from a mouse-drag, mpv calls the FFmpeg mpeg demuxer which then tries to find the corresponding .sub file in vobsub_read_header. However, mpv does not decode the path for the vobsub track name, which FFmpeg uses to find the sub file, so open() was receiving the path with "file:" removed but still had the percent encoding, so it obviously failed. I have a similar patch for mpv to make mpv decode the vobsub sub file name before passing to FFmpeg as well, but I have received feedback from multiple mpv developers that this is also a FFmpeg issue, so hopefully we can fix it in both places. Signed-off-by: Nick Sarnie <sarnex@gentoo.org> --- libavformat/file.c | 18 ++++++++++++++++-- 1 file changed, 16 insertions(+), 2 deletions(-) const char *filename = h->filename; av_strstart(filename, "file:", &filename); - + filename = ff_urldecode(filename, 0); + if(!filename) + return AVERROR(ENOMEM); { #if HAVE_ACCESS && defined(R_OK) if (access(filename, F_OK) < 0) @@ -174,7 +177,9 @@ static int file_delete(URLContext *h) int ret; const char *filename = h->filename; av_strstart(filename, "file:", &filename); - + filename = ff_urldecode(filename, 0); + if(!filename) + return AVERROR(ENOMEM); ret = rmdir(filename); if (ret < 0 && (errno == ENOTDIR # ifdef _WIN32 @@ -196,7 +201,13 @@ static int file_move(URLContext *h_src, URLContext *h_dst) const char *filename_src = h_src->filename; const char *filename_dst = h_dst->filename; av_strstart(filename_src, "file:", &filename_src); + filename_src = ff_urldecode(filename_src, 0); + if(!filename_src) + return AVERROR(ENOMEM); av_strstart(filename_dst, "file:", &filename_dst); + filename_dst = ff_urldecode(filename_dst, 0); + if(!filename_dst) + return AVERROR(ENOMEM); if (rename(filename_src, filename_dst) < 0) return AVERROR(errno); @@ -212,6 +223,9 @@ static int file_open(URLContext *h, const char *filename, int flags) struct stat st; av_strstart(filename, "file:", &filename); + filename = ff_urldecode(filename, 0); + if(!filename) + return AVERROR(ENOMEM); if (flags & AVIO_FLAG_WRITE && flags & AVIO_FLAG_READ) { access = O_CREAT | O_RDWR;