- Dillo-dev - lists.mailman3.com

Re: Firefox bookmarks plugin

by Rodrigo Arias

Hi, On Mon, Aug 26, 2024 at 01:31:13AM +0200, Diego wrote: >Good work and very interesting idea. Thanks! >If I can suggest a change I would add this lines to dpidrc at Makefile >the installation > ># Default dillo bookmarks dpi >#bookmarks=bookmarks/bookmarks.dpi ># ffbm: A Dillo plugin to sync bookmarks with Firefox >bookmarks=$(DPI_DIR)/$(BIN) > > >maybe it's time to patch dpidrc to make it more descriptive > >I sugest this >dpi_dir=/usr/local/lib/dillo/dpi > ># standard dillo distributed DPIs >#bookmarks=bookmarks/bookmarks.dpi >#cookies=cookies/cookies.dpi >#downloads=downloads/downloads.dpi >#vsource=vsource/vsource.filter.dpi The file dpidrc only registers the protocols (ftp: -> ftp dpi), the dpis are taken from the dpi_dir or ~/.dillo/dpi automatically. It doesn't require describing the available dpis in the file. Not sure what would be the advantage of doing so. Best, Rodrigo.

Aug. 27, 2024

Re: TLS connect error: "an EOF was observed that violates the protocol"

by Rodrigo Arias

Hi Alex, On Tue, Aug 27, 2024 at 09:56:46PM +0200, a1ex(a)dismail.de wrote: >Hi Rodrigo, > >On Tue, 27 Aug 2024 20:19:32 +0200 >Rodrigo Arias <rodarima(a)gmail.com> wrote: > >> >1) socks5 proxy on 127.0.0.1:8080 not working. >> >> There is currently no support for socks5 proxies, only HTTP proxies >> are supported. > >Thanks for confirming, the docs are not really clear on that. >Could be a nice feature for the future :) Yeah, unfortunately I only have two hands :-) >> >dillo-dev.auriga.wearlab.narkive.com:443 >> > >> >Is this site just broken/misconfigured? >> >> Cannot reproduce with LibreSSL 3.9.2 on Linux. >> ... >> Can you test with the latest LibreSSL 3.9.2? >> >> Is this it happening with the proxy enabled? Also, which user agent >> are you using in curl and Dillo? > >It happens with the stock user agent and no proxy, same with curl. I'm >running the latest snapshot of OpenBSD, which would have the latest >version of LibreSSL. > >I don't care about that site, my only worry is that the error can crash >Dillo. If it happens it is a bug on Dillo side as it is not handling all errors, regardless of the site. I would like to reproduce it to fix it. >This could be an OpenBSD specific issue which wouldn't show up on >Linux. Maybe this is a similar issue: >https://github.com/mitmproxy/mitmproxy/issues/7128 It could be, but I would first reject that is not due to mismatch of versions. The last LibreSSL as per https://www.libressl.org/ is: > The latest stable release is 3.9.2 Which should be printed in the first lines when starting Dillo: > TLS library: LibreSSL 3.9.2 If it says 3.9.0, then Dillo is not using the last release. I can try installing 3.9.0 and see if I can reproduce it with that one. Best, Rodrigo.

Aug. 27, 2024

Re: Issues with HTTP multipart/form-data file upload

by Rodrigo Arias

Hi Xavier, On Mon, Aug 26, 2024 at 01:27:29AM +0200, Xavier Del Campo Romero wrote: >Hello Dillo dev community, > >I am testing support among web browsers for slcl [1], a JS-less >minimalist web storage solution. slcl relies on >"multipart/form-data"-encoded HTTP requests to upload files to a server. >Whereas Dillo helped me to uncover a few wrong assumptions on my code, I >have realised a few issues on Dillo itself related to filue uploads that >should be considered. Glad to read that you also consider Dillo for slcl, and thanks for preparing the patches :-) >Issue #1: > >While Dillo is fine uploading small files (up to a few MiB), things go >wrong with larger files, so much that memory usage increases up to >multiple GiB and can even lock the system up. This is because Dillo is >designed to send requests always from memory, which means file contents >must be dumped into memory first, and this might be unfeasible for large >files. > >Suggestion: > >Instead, Dillo should ideally send file contents on-the-fly, so that >memory usage is kept to a minimum regardless the file size. Sounds good, not sure how complicated it would be to do this. >Issue #2: > >Even if Dillo generates a 70-byte, random boundary string (yet mostly >filled with '-', similarly to Firefox [2]), it ensures it is not found >anywhere inside the file contents. Again, this can be a serious >bottleneck in the case of large files, as it requires to scan the whole >file for a match. > >Suggestion: > >Define *all* of the 70 bytes in the boundary string as random, and >assume they would never be found inside a file. The chance of accidental >collision is so low that it is not worth the effort into checking them. > >Suggested patches: > >- 0001-dialog.cc-Generate-more-random-boundaries.patch) Yes, I think is a good idea. @@ -1246,6 +1246,24 @@ Dstr *DilloHtmlForm::buildQueryData(DilloHtmlInput *active_submit) return DataStr; } +static void generate_boundary(Dstr *boundary) +{ + for (int i = 0; i < 70; i++) { I think this is too long: > Boundary delimiters must not appear within the encapsulated material, > and must be no longer than 70 characters, not counting the two > leading hyphens. Shouldn't it be 68 then? + /* Extracted from RFC 2046, section 5.1.1. */ + static const char set[] = "abcdefghijklmnopqrstuvwxyz" + "ABCDEFGHIJKLMNOPQRSTUVWXYZ" + "0123456789" + "'()+_,-./:=? "; I would leave out all the symbols to avoid quoting and only use A-Z a-z and 0-9. Assuming you are left with 62 symbols (26*2 + 10), the probability of finding a random 70 character string in a random sample of 70 characters is (using ** as ^): p = 1/62**70 = 3.4086e-126 But as you will be dealing with large files, each line of the input could match it. We are interested in the probability that a given boundary will mach any of the N lines. So we compute the probability that it will match none of the lines (1-p)**N and then the opposite of that (matches at least one line): 1 - (1 - p) ** N For example, a 1 TB encoded file, with 2**40/70 lines will have: p = 1 - (1 - 1/62**70) ** (2**40 / 70) Of occurring, which is around 5.354e-116. Which, if I computed it correctly, is still too small to worry about. + char s[sizeof " "] = {0}; Why sizeof " " instead of just 2? + + do { + *s = rand(); + } while (!strspn(s, set)); + + dStr_append(boundary, s); + } +} + /** * Generate a boundary string for use in separating the parts of a * multipart/form-data submission. >Issue #3: > >Dillo only supports uploading 1 file at a time. This is mostly because >it relies (probably temporarily?) on the a_Dialog_save_file function >[3]. However, this is not a limitation on the HTTP protocol, and other >implementation such as Firefox or Chromium-based browsers support this. > >Suggested patches: > >- 0002-dialog-Add-a_Dialog_select_files.patch >- 0003-WIP-multi-file-uploads.patch > >Conclusions: > >Dillo seems designed to always send requests from memory, so it is not >straightforward to break this assumption in order to support large file >uploads. 0003-WIP-multi-file-uploads.patch is an incomplete first step >into fixing this, but it surely needs deeper design changes. > >I did not put more effort into these patches for the time being because, >after seeing the potential complexity behind this task, I thought it was >a better idea to ask the community for feedback and guidelines. I'm not very familiar with this part of Dillo, so I'll have to check a bit more before providing more feedback. However, being able to upload multiple files at the same time sounds reasonable, so feel free to try on your own in the meanwhile. PS: When are you playing? Best, Rodrigo.

Aug. 27, 2024

Re: TLS connect error: "an EOF was observed that violates the protocol"

by a1ex＠dismail.de

Hi Rodrigo, On Tue, 27 Aug 2024 20:19:32 +0200 Rodrigo Arias <rodarima(a)gmail.com> wrote: > >1) socks5 proxy on 127.0.0.1:8080 not working. > > There is currently no support for socks5 proxies, only HTTP proxies > are supported. Thanks for confirming, the docs are not really clear on that. Could be a nice feature for the future :) > >dillo-dev.auriga.wearlab.narkive.com:443 > > > >Is this site just broken/misconfigured? > > Cannot reproduce with LibreSSL 3.9.2 on Linux. > ... > Can you test with the latest LibreSSL 3.9.2? > > Is this it happening with the proxy enabled? Also, which user agent > are you using in curl and Dillo? It happens with the stock user agent and no proxy, same with curl. I'm running the latest snapshot of OpenBSD, which would have the latest version of LibreSSL. I don't care about that site, my only worry is that the error can crash Dillo. This could be an OpenBSD specific issue which wouldn't show up on Linux. Maybe this is a similar issue: https://github.com/mitmproxy/mitmproxy/issues/7128 -Alex

Aug. 27, 2024

Re: TLS connect error: "an EOF was observed that violates the protocol"

by Rodrigo Arias

Hi Alex, On Tue, Aug 27, 2024 at 11:55:04AM +0200, a1ex(a)dismail.de wrote: >Hi, > >Here are 2 separate issues: > >1) socks5 proxy on 127.0.0.1:8080 not working. There is currently no support for socks5 proxies, only HTTP proxies are supported. >This is a proxy over ssh, for example: >ssh -N -D 8080 user(a)example.com > >in dillorc: >http_proxy="http://localhost:8080/" > >console output when trying to connect to a site: >Connecting to 127.0.0.1:8080 >CONNECT through proxy failed. Full reply not received: >(nothing) >** WARNING **: CCC: call on already finished chain. Flags=CCC_Ended > >This setup works fine under Firefox. > >I tried doing a tcpdump while attempting the connection, and there is no >activity. > >So, while trying to research this, I ran into another issue: > >2) Any time I go to a page on this site, Dillo crashes with the >following: > >Nav_open_url: new >url='https://dillo-dev.auriga.wearlab.narkive.com/WT0JYUZq/dillo-won-t-resolve-u…' Dns_server [0]: dillo-dev.auriga.wearlab.narkive.com is 149.248.211.108 >Connecting to 149.248.211.108:443 >TLS connect error: "an EOF was observed that violates the protocol" >Tls_close_by_key: Avoiding SSL shutdown >for: https://dillo-dev.auriga.wearlab.narkive.com/WT0JYUZq/dillo-won-t-resolve-u… fd 6 is done and failed >dillo(13689) in malloc(): write to free mem 0x60383b59140[24..31]@32 >Abort trap > >gdb output: >Program received signal SIGABRT, Aborted. >thrkill () at /tmp/-:2 >2 /tmp/-: No such file or directory. > in /tmp/- > >This is on OpenBSD-current amd64 with LibreSSL 3.9.0, running an >unmodified fresh checkout of Dillo master. Also tested on OpenBSD 7.5 >with the same result. > >I tested the site with: >https://www.ssllabs.com/ssltest/analyze.html?d=dillo-dev.auriga.wearlab.nar… > >There seem to be some handshake failures during the simulation. > >This probably is not be the fault of Dillo, but maybe there is a more >graceful to handle this, rather than crashing. > >Even a test with curl has issues: > >curl: (35) LibreSSL SSL_connect: SSL_ERROR_SYSCALL in connection to >dillo-dev.auriga.wearlab.narkive.com:443 > >Is this site just broken/misconfigured? Cannot reproduce with LibreSSL 3.9.2 on Linux. % LD_LIBRARY_PATH=/usr/lib/libressl src/dillo https://dillo-dev.auriga.wearlab.narkive.com/WT0JYUZq/dillo-won-t-resolve-u… dillo_dns_init: Here we go! (threaded) TLS library: LibreSSL 3.9.2 Enabling cookies as from cookiesrc... Nav_open_url: new url='https://dillo-dev.auriga.wearlab.narkive.com/WT0JYUZq/dillo-won-t-resolve-u…' Dns_server [0]: dillo-dev.auriga.wearlab.narkive.com is 149.248.211.108 Connecting to 149.248.211.108:443 dillo-dev.auriga.wearlab.narkive.com: TLSv1.3, cipher TLS_AES_128_GCM_SHA256 sha256 2048-bit RSA: /CN=narkive.com sha256 2048-bit RSA: /C=US/O=Let's Encrypt/CN=R11 root: /C=US/O=Internet Security Research Group/CN=ISRG Root X1 NumPendingStyleSheets=1 Dns_server [0]: narkive.net is 188.114.97.5 188.114.96.5 narkive.net: TLSv1.3, cipher TLS_AES_256_GCM_SHA384 ecdsa-with-SHA256 256-bit EC: /CN=narkive.net ecdsa-with-SHA384 256-bit EC: /C=US/O=Google Trust Services/CN=WE1 sha256 384-bit EC: /C=US/O=Google Trust Services LLC/CN=GTS Root R4 root: /C=BE/O=GlobalSign nv-sa/OU=Root CA/CN=GlobalSign Root CA Can you test with the latest LibreSSL 3.9.2? Is this it happening with the proxy enabled? Also, which user agent are you using in curl and Dillo? Best, Rodrigo.

Aug. 27, 2024

TLS connect error: "an EOF was observed that violates the protocol"

by a1ex＠dismail.de

Hi, Here are 2 separate issues: 1) socks5 proxy on 127.0.0.1:8080 not working. This is a proxy over ssh, for example: ssh -N -D 8080 user(a)example.com in dillorc: http_proxy="http://localhost:8080/" console output when trying to connect to a site: Connecting to 127.0.0.1:8080 CONNECT through proxy failed. Full reply not received: (nothing) ** WARNING **: CCC: call on already finished chain. Flags=CCC_Ended This setup works fine under Firefox. I tried doing a tcpdump while attempting the connection, and there is no activity. So, while trying to research this, I ran into another issue: 2) Any time I go to a page on this site, Dillo crashes with the following: Nav_open_url: new url='https://dillo-dev.auriga.wearlab.narkive.com/WT0JYUZq/dillo-won-t-resolve-u…' Dns_server [0]: dillo-dev.auriga.wearlab.narkive.com is 149.248.211.108 Connecting to 149.248.211.108:443 TLS connect error: "an EOF was observed that violates the protocol" Tls_close_by_key: Avoiding SSL shutdown for: https://dillo-dev.auriga.wearlab.narkive.com/WT0JYUZq/dillo-won-t-resolve-u… fd 6 is done and failed dillo(13689) in malloc(): write to free mem 0x60383b59140[24..31]@32 Abort trap gdb output: Program received signal SIGABRT, Aborted. thrkill () at /tmp/-:2 2 /tmp/-: No such file or directory. in /tmp/- This is on OpenBSD-current amd64 with LibreSSL 3.9.0, running an unmodified fresh checkout of Dillo master. Also tested on OpenBSD 7.5 with the same result. I tested the site with: https://www.ssllabs.com/ssltest/analyze.html?d=dillo-dev.auriga.wearlab.nar… There seem to be some handshake failures during the simulation. This probably is not be the fault of Dillo, but maybe there is a more graceful to handle this, rather than crashing. Even a test with curl has issues: curl: (35) LibreSSL SSL_connect: SSL_ERROR_SYSCALL in connection to dillo-dev.auriga.wearlab.narkive.com:443 Is this site just broken/misconfigured? Regards, Alex

Aug. 27, 2024

Re: Firefox bookmarks plugin

by Diego

Good work and very interesting idea. If I can suggest a change I would add this lines to dpidrc at Makefile the installation # Default dillo bookmarks dpi #bookmarks=bookmarks/bookmarks.dpi # ffbm: A Dillo plugin to sync bookmarks with Firefox bookmarks=$(DPI_DIR)/$(BIN) maybe it's time to patch dpidrc to make it more descriptive I sugest this dpi_dir=/usr/local/lib/dillo/dpi # standard dillo distributed DPIs #bookmarks=bookmarks/bookmarks.dpi #cookies=cookies/cookies.dpi #downloads=downloads/downloads.dpi #vsource=vsource/vsource.filter.dpi # protocols DPIs proto.file=file/file.dpi proto.ftp=ftp/ftp.filter.dpi proto.data=datauri/datauri.filter.dpi # dpid uses this line when no line for a protocol is found in this file proto.*=misc/ProtocolNotSupported.filter.dpi Thanks for all the work cheers! Diego. El dom, 25 ago 2024 a las 21:55, Rodrigo Arias (<rodarima(a)gmail.com>) escribió: > > Hi, > > Here is another small plugin to sync bookmarks with Firefox Sync: > > https://github.com/dillo-browser/dillo-plugin-ffbm > > It uses the ffsclient program to connect to the Firefox Sync server and > a few lines of shell script for plumbing. It replaces the builtin > bookmarks plugin when installed, so all menu buttons work as expected. > > Best, > Rodrigo > _______________________________________________ > Dillo-dev mailing list -- dillo-dev(a)mailman3.com > To unsubscribe send an email to dillo-dev-leave(a)mailman3.com

Aug. 25, 2024

Issues with HTTP multipart/form-data file upload

by Xavier Del Campo Romero

Hello Dillo dev community, I am testing support among web browsers for slcl [1], a JS-less minimalist web storage solution. slcl relies on "multipart/form-data"-encoded HTTP requests to upload files to a server. Whereas Dillo helped me to uncover a few wrong assumptions on my code, I have realised a few issues on Dillo itself related to filue uploads that should be considered. Issue #1: While Dillo is fine uploading small files (up to a few MiB), things go wrong with larger files, so much that memory usage increases up to multiple GiB and can even lock the system up. This is because Dillo is designed to send requests always from memory, which means file contents must be dumped into memory first, and this might be unfeasible for large files. Suggestion: Instead, Dillo should ideally send file contents on-the-fly, so that memory usage is kept to a minimum regardless the file size. Issue #2: Even if Dillo generates a 70-byte, random boundary string (yet mostly filled with '-', similarly to Firefox [2]), it ensures it is not found anywhere inside the file contents. Again, this can be a serious bottleneck in the case of large files, as it requires to scan the whole file for a match. Suggestion: Define *all* of the 70 bytes in the boundary string as random, and assume they would never be found inside a file. The chance of accidental collision is so low that it is not worth the effort into checking them. Suggested patches: - 0001-dialog.cc-Generate-more-random-boundaries.patch) Issue #3: Dillo only supports uploading 1 file at a time. This is mostly because it relies (probably temporarily?) on the a_Dialog_save_file function [3]. However, this is not a limitation on the HTTP protocol, and other implementation such as Firefox or Chromium-based browsers support this. Suggested patches: - 0002-dialog-Add-a_Dialog_select_files.patch - 0003-WIP-multi-file-uploads.patch Conclusions: Dillo seems designed to always send requests from memory, so it is not straightforward to break this assumption in order to support large file uploads. 0003-WIP-multi-file-uploads.patch is an incomplete first step into fixing this, but it surely needs deeper design changes. I did not put more effort into these patches for the time being because, after seeing the potential complexity behind this task, I thought it was a better idea to ask the community for feedback and guidelines. Thank you very much for reading. Best regards, Xavier Del Campo Romero [1]: https://gitea.privatedns.org/xavi/slcl [2]: https://github.com/dillo-browser/dillo/blob/8a360e32ac3136494a494379a6dbbac… [3]: https://github.com/dillo-browser/dillo/blob/8a360e32ac3136494a494379a6dbbac…

Aug. 25, 2024

Firefox bookmarks plugin

by Rodrigo Arias

Hi, Here is another small plugin to sync bookmarks with Firefox Sync: https://github.com/dillo-browser/dillo-plugin-ffbm It uses the ffsclient program to connect to the Firefox Sync server and a few lines of shell script for plumbing. It replaces the builtin bookmarks plugin when installed, so all menu buttons work as expected. Best, Rodrigo

Aug. 25, 2024

Re: [PATCH] Dark theme for downloads.dpi

by Rodrigo Arias

Hi, On Sat, Aug 24, 2024 at 06:40:33PM +0200, a1ex(a)dismail.de wrote: >Hi, > >On Sat, 24 Aug 2024 15:39:01 +0200 >Rodrigo Arias <rodarima(a)gmail.com> wrote: > >> I'm also thinking we may want to also include the icons, so we avoid >> the problem of having a single icon that must support dark and light >> backgrounds. In this case, maybe a layout like this may be more >> useful: >> >> ~/.dillo/theme/<name>/style.css >> ~/.dillo/theme/<name>/icons.png > >I agree that themed icons would be very nice, so your layout here makes >sense. > >Might be a bit of work to get Dillo to use icons.png instead of the >built-in pixmaps.h though :) > >Anyway, it sounds like this will be a pretty complicated patch. >Not sure where I can help, but can try if desired on smaller parts. Not sure if it would be an easy patch. You may want to review the issues in case you find some that seems more approachable. https://github.com/dillo-browser/dillo/issues I would also appreciate if someone can test this PR by doing some daily google searches over a week or so with the cookies disabled for google: https://github.com/dillo-browser/dillo/pull/251 The question is if it is better to leave the ucbcb=1 parameter in the URL by default or not: search_url="Google https://www.google.com/search?ie=UTF-8&oe=UTF-8&gbv=1&ucbcb=1&q=%s" I tried to skip the consent dialog by denying it by default so it doesn't appear, but I get mixed results even without cookies disabled. Some times it appears again and others it doesn't, not sure how they determine when it should appear. Maybe it is better to leave it out. Testing this PR would be helpful and doesn't require much effort, in case you are interested. I can only do limited tests before Google fingerprints my IP and then it never ask me consent again. (And yes, I'm aware of the implications of using Google, but it is the only search engine that gives me any result in very weird searches like part numbers of unknown electrical components) Best, Rodrigo.

Aug. 25, 2024