- Dillo-dev - lists.mailman3.com

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

by Xavier Del Campo Romero

Hi Rodrigo, > Not sure what you mean with "unnecessary use of the heap". I meant > something like this: You are right: I was too fixated with strcspn(3), when in fact strchr(3) would already do. Also, this use of dStr_append_c() looks good to me. > The variable c is stored in a register, as you can see in the > disassembly (built with -Og): Thanks for the detailed explanation. Anyway, I do not expect this function to become a performance bottleneck. > This other method has a slightly not > uniform distribution, but only runs rand() 70 times: > > Which is probably okay for this case. I also think it is good enough. > Then, I think using our own set may be the most readable solution here, > which also allows the quick module method. I agree. Best regards, Xavi On 1/9/24 16:12, Rodrigo Arias wrote: > Hi Xavier, > >> Thank you. I am still unfamiliar with that part of Dillo, so please let >> me know about any progress. > > For now I'm still writing the RFC and doing some proof of concepts. I > can see that it will take a while. > >> Limiting ourselves to a-z, A-Z and 0-9 would still account for 62 out of >> the 75 possible characters, so roughly 82% of the set. I think that >> removing the quoting in favour of the limited set reduce the risk for >> broken implementations, yet still provide a good amount of randomness. > > Yes, I think so too. > >> I am not sure whether this was an intentional modification from your >> side. My patch is adding a <space> as defined by POSIX.1-2017 [1], so >> that sizeof " " would always return 2. Was it your intention to flag >> this potential confusion? > > Yes, my point was that it is that is not easy to determine the length of > a UTF-8 string by just looking at it. > >> Also, there was not strict reason to use sizeof " ". Any other >> character would do e.g.: sizeof "x", sizeof "A", etc. > > Same problem: > > ᕁ 𝓍 𝙭 х 𝐱 𝗑 ⤫ 𝑥 𝘅 ⤬ ᙮ ⨯ 𝕩 𝖝 × 𝔁 𝚡 x ｘ ⅹ 𝔵 ᕽ 𝒙 𝘹 > > 𝙰 A 𝐴 ᗅ 𝑨 𝚨 𝕬 𝖠 А Α 𝛢 𝝖 𝒜 𝜜 𖽀 𝓐 𝞐 Ａ 𝗔 𝘈 Ꭺ ꓮ 𝐀 𝔄 𝔸 𐊠 𝘼 > > Check: https://util.unicode.org/UnicodeJsps/confusables.jsp > > It is generally safer to use the explicit length. > >>> You can also use dStr_append_c() to only append one character, so you >>> only need a single character. >> >> That would be an unnecessary use of the heap, because the size is static. > > Not sure what you mean with "unnecessary use of the heap". I meant > something like this: > > static void generate_boundary(Dstr *boundary) > { > for (int i = 0; i < 70; i++) { > /* Extracted from RFC 2046, section 5.1.1. */ > static const char set[] = "abcdefghijklmnopqrstuvwxyz" > "ABCDEFGHIJKLMNOPQRSTUVWXYZ" > "0123456789"; > int c; > > do { > c = rand() & 0xff; > } while (!strchr(set, c)); > > dStr_append_c(boundary, c); > } > } > > The variable c is stored in a register, as you can see in the > disassembly (built with -Og): > > hop% r2 build/src/dillo > WARN: Relocs has not been applied. Please use `-e bin.relocs.apply=true` > or `-e bin.cache=true` next time > [0x000e3860]> s sym.generate_boundary_Dstr_ > [0x0012cf99]> af > [0x0012cf99]> pdf > ┌ 85: sym.generate_boundary_Dstr_ (int64_t arg1); > │ ; arg int64_t arg1 @ rdi > │ 0x0012cf99 55 push rbp ; > Fl_Pixmap.H:1250 ; generate_boundary(Dstr*) > │ 0x0012cf9a 4889e5 mov rbp, rsp > │ 0x0012cf9d 4155 push r13 > │ 0x0012cf9f 4154 push r12 > │ 0x0012cfa1 53 push rbx > │ 0x0012cfa2 4883ec08 sub rsp, 8 > │ 0x0012cfa6 4989fd mov r13, rdi ; > arg1 > │ 0x0012cfa9 41bc00000000 mov r12d, 0 ; > Fl_Pixmap.H:1251 > │ ┌─< 0x0012cfaf eb2c jmp 0x12cfdd > │ ┌┌──> 0x0012cfb1 ff1509cd1e00 call qword [reloc.rand] ; > Fl_Pixmap.H:1253 ; [0x319cc0:8]=0 > │ ╎╎│ 0x0012cfb7 0fb6d8 movzx ebx, al ; > <--- Perform the "& 0xff" by zero-extending the lowest byte in eax > │ ╎╎│ 0x0012cfba 89de mov esi, ebx ; > Fl_Pixmap.H:1260 > │ ╎╎│ 0x0012cfbc 488d3dbd5b.. lea rdi, > obj.generate_boundary_Dstr_::set ; 0x2a2b80 ; > "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789" > │ ╎╎│ 0x0012cfc3 ff15afc51e00 call qword [reloc.strchr] ; > [0x319578:8]=0 > │ ╎╎│ 0x0012cfc9 4885c0 test rax, rax > │ └───< 0x0012cfcc 74e3 je 0x12cfb1 > │ ╎│ 0x0012cfce 89de mov esi, ebx ; > Fl_Pixmap.H:1262 > │ ╎│ 0x0012cfd0 4c89ef mov rdi, r13 > │ ╎│ 0x0012cfd3 67e830570800 call sym.dStr_append_c > │ ╎│ 0x0012cfd9 4183c401 add r12d, 1 ; > Fl_Pixmap.H:1251 > │ ╎│ ; CODE XREF from generate_boundary(Dstr*) @ 0x12cfaf(x) > │ ╎└─> 0x0012cfdd 4183fc45 cmp r12d, 0x45 ; > 'E' > │ └──< 0x0012cfe1 7ece jle 0x12cfb1 > │ 0x0012cfe3 4883c408 add rsp, 8 ; > Fl_Pixmap.H:1264 > │ 0x0012cfe7 5b pop rbx > │ 0x0012cfe8 415c pop r12 > │ 0x0012cfea 415d pop r13 > │ 0x0012cfec 5d pop rbp > └ 0x0012cfed c3 ret > > This method gives us a perfect uniform distribution when RAND_MAX is a > power of 2, at the cost of executing rand() more times than required > (70/(62/256) ≈ 289 on average). This other method has a slightly not > uniform distribution, but only runs rand() 70 times: > > static void generate_boundary(Dstr *boundary) > { > /* Extracted from RFC 2046, section 5.1.1. */ > static const char set[] = "abcdefghijklmnopqrstuvwxyz" > "ABCDEFGHIJKLMNOPQRSTUVWXYZ" > "0123456789"; > static const int n = strlen(set); > > for (int i = 0; i < 70; i++) { > int c = (unsigned char) set[rand() % n]; > dStr_append_c(boundary, c); > } > } > > Which is probably okay for this case. > > Notice I haven't tested any of these methods yet, I'll need to add a > form upload test case to be able to see them in action (or a unit test). > >>> If we only use alphanumeric characters, we can just use isalnum() right? >> >> According to POSIX.1-2017 [2], isalnum(3) depends on the current locale >> configured by the system. For example, characters such as Ä or ú could >> return non-zero. > > Oh right, for some reason I was thinking this was the other way around, > and isalnum only worked with ASCII. I think we should review the other > uses of isalnum and friends as I think there may be used under similar > assumptions. > >> To avoid this, there are two possible solutions: >> >> 1. Use isalnum_l(3) to specify a locale_t object corresponding to the >> "POSIX" locale (equivalent to "C" [3]), which must be previously >> allocated by the newlocale(3) function [3] and released by the >> freelocal(3) function [4]. A minimalist example is shown below: >> >> locale_t l = newlocale(LC_CTYPE, "POSIX", NULL); >> >> for (unsigned char i = 0; i < 255; i++) >> printf("hhu=%hhu, c=%c, isalnum=%d\n", i, i, >> isalnum_l(i, l)); >> >> freelocale(l); >> >> 2. Define a known subset from the portable character set defined by >> POSIX.1-2017 [5] and use strspn(3), as already suggested by the patch. >> IMHO this approach is better because: >> - It does not deal with locales, so developers not familiar with them >> would understand the code better. >> - It is also portable outside a POSIX environment (not sure if this a >> requirement, though). >> - It does not require dynamic allication via newlocale(3). >> - It is the only possible option if non-alnum characters, such as ':' >> or '/', are appended to the boundary string. > > Then, I think using our own set may be the most readable solution here, > which also allows the quick module method. > > Best, > Rodrigo > _______________________________________________ > Dillo-dev mailing list -- dillo-dev(a)mailman3.com > To unsubscribe send an email to dillo-dev-leave(a)mailman3.com

Sept. 1, 2024

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

by Rodrigo Arias

Hi Xavier, >Thank you. I am still unfamiliar with that part of Dillo, so please let >me know about any progress. For now I'm still writing the RFC and doing some proof of concepts. I can see that it will take a while. >Limiting ourselves to a-z, A-Z and 0-9 would still account for 62 out of >the 75 possible characters, so roughly 82% of the set. I think that >removing the quoting in favour of the limited set reduce the risk for >broken implementations, yet still provide a good amount of randomness. Yes, I think so too. >I am not sure whether this was an intentional modification from your >side. My patch is adding a <space> as defined by POSIX.1-2017 [1], so >that sizeof " " would always return 2. Was it your intention to flag >this potential confusion? Yes, my point was that it is that is not easy to determine the length of a UTF-8 string by just looking at it. >Also, there was not strict reason to use sizeof " ". Any other >character would do e.g.: sizeof "x", sizeof "A", etc. Same problem: ᕁ 𝓍 𝙭 х 𝐱 𝗑 ⤫ 𝑥 𝘅 ⤬ ᙮ ⨯ 𝕩 𝖝 × 𝔁 𝚡 x ｘ ⅹ 𝔵 ᕽ 𝒙 𝘹 𝙰 A 𝐴 ᗅ 𝑨 𝚨 𝕬 𝖠 А Α 𝛢 𝝖 𝒜 𝜜 𖽀 𝓐 𝞐 Ａ 𝗔 𝘈 Ꭺ ꓮ 𝐀 𝔄 𝔸 𐊠 𝘼 Check: https://util.unicode.org/UnicodeJsps/confusables.jsp It is generally safer to use the explicit length. >> You can also use dStr_append_c() to only append one character, so you >> only need a single character. > >That would be an unnecessary use of the heap, because the size is static. Not sure what you mean with "unnecessary use of the heap". I meant something like this: static void generate_boundary(Dstr *boundary) { for (int i = 0; i < 70; i++) { /* Extracted from RFC 2046, section 5.1.1. */ static const char set[] = "abcdefghijklmnopqrstuvwxyz" "ABCDEFGHIJKLMNOPQRSTUVWXYZ" "0123456789"; int c; do { c = rand() & 0xff; } while (!strchr(set, c)); dStr_append_c(boundary, c); } } The variable c is stored in a register, as you can see in the disassembly (built with -Og): hop% r2 build/src/dillo WARN: Relocs has not been applied. Please use `-e bin.relocs.apply=true` or `-e bin.cache=true` next time [0x000e3860]> s sym.generate_boundary_Dstr_ [0x0012cf99]> af [0x0012cf99]> pdf ┌ 85: sym.generate_boundary_Dstr_ (int64_t arg1); │ ; arg int64_t arg1 @ rdi │ 0x0012cf99 55 push rbp ; Fl_Pixmap.H:1250 ; generate_boundary(Dstr*) │ 0x0012cf9a 4889e5 mov rbp, rsp │ 0x0012cf9d 4155 push r13 │ 0x0012cf9f 4154 push r12 │ 0x0012cfa1 53 push rbx │ 0x0012cfa2 4883ec08 sub rsp, 8 │ 0x0012cfa6 4989fd mov r13, rdi ; arg1 │ 0x0012cfa9 41bc00000000 mov r12d, 0 ; Fl_Pixmap.H:1251 │ ┌─< 0x0012cfaf eb2c jmp 0x12cfdd │ ┌┌──> 0x0012cfb1 ff1509cd1e00 call qword [reloc.rand] ; Fl_Pixmap.H:1253 ; [0x319cc0:8]=0 │ ╎╎│ 0x0012cfb7 0fb6d8 movzx ebx, al ; <--- Perform the "& 0xff" by zero-extending the lowest byte in eax │ ╎╎│ 0x0012cfba 89de mov esi, ebx ; Fl_Pixmap.H:1260 │ ╎╎│ 0x0012cfbc 488d3dbd5b.. lea rdi, obj.generate_boundary_Dstr_::set ; 0x2a2b80 ; "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789" │ ╎╎│ 0x0012cfc3 ff15afc51e00 call qword [reloc.strchr] ; [0x319578:8]=0 │ ╎╎│ 0x0012cfc9 4885c0 test rax, rax │ └───< 0x0012cfcc 74e3 je 0x12cfb1 │ ╎│ 0x0012cfce 89de mov esi, ebx ; Fl_Pixmap.H:1262 │ ╎│ 0x0012cfd0 4c89ef mov rdi, r13 │ ╎│ 0x0012cfd3 67e830570800 call sym.dStr_append_c │ ╎│ 0x0012cfd9 4183c401 add r12d, 1 ; Fl_Pixmap.H:1251 │ ╎│ ; CODE XREF from generate_boundary(Dstr*) @ 0x12cfaf(x) │ ╎└─> 0x0012cfdd 4183fc45 cmp r12d, 0x45 ; 'E' │ └──< 0x0012cfe1 7ece jle 0x12cfb1 │ 0x0012cfe3 4883c408 add rsp, 8 ; Fl_Pixmap.H:1264 │ 0x0012cfe7 5b pop rbx │ 0x0012cfe8 415c pop r12 │ 0x0012cfea 415d pop r13 │ 0x0012cfec 5d pop rbp └ 0x0012cfed c3 ret This method gives us a perfect uniform distribution when RAND_MAX is a power of 2, at the cost of executing rand() more times than required (70/(62/256) ≈ 289 on average). This other method has a slightly not uniform distribution, but only runs rand() 70 times: static void generate_boundary(Dstr *boundary) { /* Extracted from RFC 2046, section 5.1.1. */ static const char set[] = "abcdefghijklmnopqrstuvwxyz" "ABCDEFGHIJKLMNOPQRSTUVWXYZ" "0123456789"; static const int n = strlen(set); for (int i = 0; i < 70; i++) { int c = (unsigned char) set[rand() % n]; dStr_append_c(boundary, c); } } Which is probably okay for this case. Notice I haven't tested any of these methods yet, I'll need to add a form upload test case to be able to see them in action (or a unit test). >> If we only use alphanumeric characters, we can just use isalnum() right? > >According to POSIX.1-2017 [2], isalnum(3) depends on the current locale >configured by the system. For example, characters such as Ä or ú could >return non-zero. Oh right, for some reason I was thinking this was the other way around, and isalnum only worked with ASCII. I think we should review the other uses of isalnum and friends as I think there may be used under similar assumptions. > To avoid this, there are two possible solutions: > >1. Use isalnum_l(3) to specify a locale_t object corresponding to the >"POSIX" locale (equivalent to "C" [3]), which must be previously >allocated by the newlocale(3) function [3] and released by the >freelocal(3) function [4]. A minimalist example is shown below: > > locale_t l = newlocale(LC_CTYPE, "POSIX", NULL); > > for (unsigned char i = 0; i < 255; i++) > printf("hhu=%hhu, c=%c, isalnum=%d\n", i, i, >isalnum_l(i, l)); > > freelocale(l); > >2. Define a known subset from the portable character set defined by >POSIX.1-2017 [5] and use strspn(3), as already suggested by the patch. >IMHO this approach is better because: > - It does not deal with locales, so developers not familiar with them >would understand the code better. > - It is also portable outside a POSIX environment (not sure if this a >requirement, though). > - It does not require dynamic allication via newlocale(3). > - It is the only possible option if non-alnum characters, such as ':' >or '/', are appended to the boundary string. Then, I think using our own set may be the most readable solution here, which also allows the quick module method. Best, Rodrigo

Sept. 1, 2024

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

by Rodrigo Arias

Hi Alex, On Thu, Aug 29, 2024 at 12:04:12PM +0200, a1ex(a)dismail.de wrote: >Hi Rodrigo, > >On Wed, 28 Aug 2024 23:02:38 +0200 >Rodrigo Arias <rodarima(a)gmail.com> wrote: > >> I think your IP may be banned from narkive. > >Yeah, that does sound like it could be the case. Funny, because I've >never even used that site before. I have a static IP address, so that >might have something to do with it. I'm thinking if you ever ran an exit Tor node that could be a plausible explanation. >On the plus side, without this issue we wouldn't have found the crash >in Dillo, so, thank you narkive for inadvertently helping Dillo :) Yes, this is more a symptom that we don't have enough tests. I'm thinking how to improve the test infrastructure so we can include more complex tests like bad TLS responses from a https server, among others. Best, Rodrigo.

Sept. 1, 2024

Re: Firefox bookmarks plugin

by a1ex＠dismail.de

Hi, On Fri, 30 Aug 2024 10:58:57 +0200 Diego <darkspirit5000(a)gmail.com> wrote: > > It doesn't require describing the available dpis in the file. Not > > sure what would be the advantage of doing so. > > For reference. To explain to the user how to switch between a built-in > DPI and another DPI for the same service. I think this part in the docs[0] covers it: "Plugins are searched in the ~/.dillo/dpi/ directory first, and then in the system directory if not found. The dpi_dir option sets the system dpi directory and must be specified once." [0] https://dillo-browser.github.io/user_help.html#dpidrc It seems reasonable to me, but if you have a better way of wording it, that might be the best place to start. Regards, Alex

Aug. 30, 2024

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

by Xavier Del Campo Romero

Hi Rodrigo, > Dillo has a mechanism to read chunks of data from different sources as they are arriving and pass them to the next stage for processing. However, AFAIK it always reads a chunk and appends it to a large buffer. It doesn't free the processed part until is done with the whole thing. > > This would require a change in the way Dillo processes data, but I think it would be required for large files. There are more details in the > devdoc/CCCwork.txt file and in src/chain.c if you want to take a closer look. > > As I'm planning to change the design of the CCC, I think I can take this into account too so it would be doable. I'll add it to the list of shortcomings of the current design. Thank you. I am still unfamiliar with that part of Dillo, so please let me know about any progress. > Okay, I'll focus on the boundary patch first, which is the easiest to merge and then I'll take a closer look at the others. > > Yeah, I would assume a lot of implementations are broken, so we want to try to minimize the chances we run into problems. Limiting ourselves to a-z, A-Z and 0-9 would still account for 62 out of the 75 possible characters, so roughly 82% of the set. I think that removing the quoting in favour of the limited set reduce the risk for broken implementations, yet still provide a good amount of randomness. > Check sizeof " ": https://godbolt.org/z/7Tso8ooYz Interestingly, the " " character on your last email is not really a <space> (<U0020>): $ printf "%s" " " | hd 00000000 e2 80 88 |...| 00000003 Compared to an ASCII whitespace: $ printf "%s" " " | hd 00000000 20 | | 00000001 Both Godbolt and my editor also flag that multi-byte character with a yellow rectangle around it because it would be highly confusing otherwise. For example: printf("len=%zu\n", strlen(" ")); Confusingly returns "len=3". I am not sure whether this was an intentional modification from your side. My patch is adding a <space> as defined by POSIX.1-2017 [1], so that sizeof " " would always return 2. Was it your intention to flag this potential confusion? Also, there was not strict reason to use sizeof " ". Any other character would do e.g.: sizeof "x", sizeof "A", etc. > You can also use dStr_append_c() to only append one character, so you only need a single character. That would be an unnecessary use of the heap, because the size is static. > If we only use alphanumeric characters, we can just use isalnum() right? According to POSIX.1-2017 [2], isalnum(3) depends on the current locale configured by the system. For example, characters such as Ä or ú could return non-zero. To avoid this, there are two possible solutions: 1. Use isalnum_l(3) to specify a locale_t object corresponding to the "POSIX" locale (equivalent to "C" [3]), which must be previously allocated by the newlocale(3) function [3] and released by the freelocal(3) function [4]. A minimalist example is shown below: locale_t l = newlocale(LC_CTYPE, "POSIX", NULL); for (unsigned char i = 0; i < 255; i++) printf("hhu=%hhu, c=%c, isalnum=%d\n", i, i, isalnum_l(i, l)); freelocale(l); 2. Define a known subset from the portable character set defined by POSIX.1-2017 [5] and use strspn(3), as already suggested by the patch. IMHO this approach is better because: - It does not deal with locales, so developers not familiar with them would understand the code better. - It is also portable outside a POSIX environment (not sure if this a requirement, though). - It does not require dynamic allication via newlocale(3). - It is the only possible option if non-alnum characters, such as ':' or '/', are appended to the boundary string. [1]: https://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap06.html [2]: https://pubs.opengroup.org/onlinepubs/9699919799/functions/isalnum.html [3]: https://pubs.opengroup.org/onlinepubs/9699919799/functions/newlocale.html [4]: https://pubs.opengroup.org/onlinepubs/9699919799/functions/freelocale.html [5]: https://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap06.html > I meant when is the next KoVoꓘ concert :-) No gigs ahead, but I will keep you informed. :) Best regards, Xavi On 28/8/24 22:47, Rodrigo Arias wrote: > Hi Xavier, > > On Wed, Aug 28, 2024 at 01:04:04AM +0200, Xavier Del Campo Romero wrote: >> Hi Rodrigo, >> >>> Glad to read that you also consider Dillo for slcl, and thanks for >>> preparing the patches :-) >> >> Thank you! I want slcl to be useful to anyone, including users who care >> about minimalist software like Dillo. The web is already too crowded >> with bloated "webapps" and other terrible things. :) > > Agreed! > >>> Sounds good, not sure how complicated it would be to do this. >> >> I still need to investigate this further, but I assume this would >> require Dillo to at least implement a sink callback. >> >> In other words, the component responsible for transmitting the data >> (probably src/IO/IO.c) should trigger a user-defined callback with an >> arbitrarily-sized buffer (typically, of BUFSIZ bytes, as defined by >> stdio.h) that must filled with file data. Then, the user-defined >> callback can fill from zero up to BUFSIZ bytes, which are eventually >> trasmitted to the server. > > Dillo has a mechanism to read chunks of data from different sources as > they are arriving and pass them to the next stage for processing. > However, AFAIK it always reads a chunk and appends it to a large buffer. > It doesn't free the processed part until is done with the whole thing. > > This would require a change in the way Dillo processes data, but I think > it would be required for large files. There are more details in the > devdoc/CCCwork.txt file and in src/chain.c if you want to take a closer > look. > > As I'm planning to change the design of the CCC, I think I can take this > into account too so it would be doable. I'll add it to the list of > shortcomings of the current design. > >> That said, I am still not sure how much actual effort this would take. >> But I am glad to receive positive feedback so far - I will then continue >> to find a solution. >> >>> However, being able to upload multiple files at the same time sounds >>> reasonable, so feel free to try on your own in the meanwhile. >> >> Uploading multiple files at once seems doable - the patches I sent on my >> previous email are probably already doing most of the required work. >> Again, the trickiest task is to send data on-the-fly for each selected >> file. > > Okay, I'll focus on the boundary patch first, which is the easiest to > merge and then I'll take a closer look at the others. > >> >>> Shouldn't it be 68 then? >> >> I understand the opposite: the boundary string with the two leading >> dashes ("--") included can be up to 72 bytes long, and 74 bytes long for >> the ending boundary (which includes two more dashes after the boundary >> string). This is confirmed by reading the BNF defined by RFC 2046 (some >> bits omitted for simplicity), section 5.1.1 [1]: >> >>> boundary := 0*69<bchars> bcharsnospace >>> bchars := bcharsnospace / " " >>> bcharsnospace := DIGIT / ALPHA / "'" / "(" / ")" / >>> "+" / "_" / "," / "-" / "." / >>> "/" / ":" / "=" / "?" >>> dash-boundary := "--" boundary >>> ; boundary taken from the value of >>> ; boundary parameter of the >>> ; Content-Type field> multipart-body := >>> [preamble CRLF] >>> dash-boundary transport-padding CRLF >>> body-part *encapsulation >>> close-delimiter transport-padding >>> [CRLF epilogue] >>> delimiter := CRLF dash-boundary >>> close-delimiter := delimiter "--" > > Oh right! I see that we are already using 70 characters anyway. > >> Note: even if the specification tells receivers to handle transport >> padding, for the time being I am assuming "transport-padding" as zero >> length since composers must not generate non-zero length transport >> padding. I am still not sure where transport padding would apply, >> anyway. Probably outside web browsers? >> >>> I would leave out all the symbols to avoid quoting and only use A-Z >>> a-z and 0-9. >> >> Interestingly, Dillo would always quote boundary strings [2], even if >> only using A-Z, a-z and 0-9. In fact, this is one of the wrong >> assumptions I spotted when testing slcl against Dillo. > > Yeah, I would assume a lot of implementations are broken, so we want to > try to minimize the chances we run into problems. > > Apart from slcl we should also test this with some sites and see if they > continue to work okay. > > This will also increase the fingerprinting information to distinguish > Dillo among other browsers, but I think it is not more information that > the already leaked by the user agent. > >>> Which, if I computed it correctly, is still too small to worry about. >> >> Not only it is too small of a chance: if we really wanted to do "the >> right thing" and make Dillo absolutely sure the boundary string is not >> contained within the selected files, this would imply a noticeable >> performance impact when dealing with large files, much likely for a >> near-zero benefit. >> >> I have not inspected their source code yet (and I do not want to), but I >> understand both Gecko and Chromium are also making that assumption, >> because otherwise it would take them a lot of CPU time to upload large >> files. > > But then they would be doing such assumption with a "much larger" > probability it hits the file. > > Skipping it with 70 characters is safe for one file, but also probably > safe for all files ever uploaded with Dillo. > > Maybe curl or other small codebases are easier to read, but not really > needed. > >> >>> Why sizeof " " instead of just 2? >> >> Because, to my eyes, sizeof " " has more meaningful semantics, compared >> to a magic integer constant such as 2. However, for this simple >> scenario, I would still consider both acceptable. > > Check sizeof " ": https://godbolt.org/z/7Tso8ooYz > > You can also use dStr_append_c() to only append one character, so you > only need a single character. > > If we only use alphanumeric characters, we can just use isalnum() right? > >> I can replace it with 2 if you find the other construct unacceptable. >> >>> PS: When are you playing? >> >> Sorry, I did not understand your last sentence. Could you please give a >> bit more context? :) > > I meant when is the next KoVoꓘ concert :-) > > Best, > Rodrigo. > _______________________________________________ > Dillo-dev mailing list -- dillo-dev(a)mailman3.com > To unsubscribe send an email to dillo-dev-leave(a)mailman3.com

Aug. 30, 2024

Re: Firefox bookmarks plugin

by Diego

El mar, 27 ago 2024 a las 23:17, Rodrigo Arias (<rodarima(a)gmail.com>) escribió: > > 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. > I know. > It doesn't require describing the available dpis in the file. Not sure > what would be the advantage of doing so. For reference. To explain to the user how to switch between a built-in DPI and another DPI for the same service. Maybe a documentation patch better? I can try to write some documentation about it Diego.

Aug. 30, 2024

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

by a1ex＠dismail.de

Hi Rodrigo, On Wed, 28 Aug 2024 23:02:38 +0200 Rodrigo Arias <rodarima(a)gmail.com> wrote: > I think your IP may be banned from narkive. Yeah, that does sound like it could be the case. Funny, because I've never even used that site before. I have a static IP address, so that might have something to do with it. On the plus side, without this issue we wouldn't have found the crash in Dillo, so, thank you narkive for inadvertently helping Dillo :) Regards, Alex

Aug. 29, 2024

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

by Rodrigo Arias

Hi, >I have confirmed that this patch prevents Dillo from crashing. >Thanks! Nice! >> However, it doesn't explain the EOF you are getting in narkive. Are >> you using tor or some proxy? > >No vpn/tor/proxy. I wouldn't be surprised if my network provider was >doing some weird stuff though... Maybe thats an area for further study. Look at what I just found: hop% torsocks curl -v https://narkive.com * Host narkive.com:443 was resolved. * IPv6: (none) * IPv4: 149.248.211.108 * Trying 149.248.211.108:443... * Connected to narkive.com (149.248.211.108) port 443 * ALPN: curl offers h2,http/1.1 * TLSv1.3 (OUT), TLS handshake, Client hello (1): * CAfile: /etc/ssl/certs/ca-certificates.crt * CApath: none * OpenSSL SSL_connect: SSL_ERROR_SYSCALL in connection to narkive.com:443 * Closing connection curl: (35) OpenSSL SSL_connect: SSL_ERROR_SYSCALL in connection to narkive.com:443 Same if I pass Dillo traffic via tor. I think your IP may be banned from narkive. Try in some days or reset your router if you have a dynamic IP. Or mail them if persists. Best, Rodrigo.

Aug. 28, 2024

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

by Rodrigo Arias

Hi Xavier, On Wed, Aug 28, 2024 at 01:04:04AM +0200, Xavier Del Campo Romero wrote: >Hi Rodrigo, > >> Glad to read that you also consider Dillo for slcl, and thanks for preparing the patches :-) > >Thank you! I want slcl to be useful to anyone, including users who care >about minimalist software like Dillo. The web is already too crowded >with bloated "webapps" and other terrible things. :) Agreed! >> Sounds good, not sure how complicated it would be to do this. > >I still need to investigate this further, but I assume this would >require Dillo to at least implement a sink callback. > >In other words, the component responsible for transmitting the data >(probably src/IO/IO.c) should trigger a user-defined callback with an >arbitrarily-sized buffer (typically, of BUFSIZ bytes, as defined by >stdio.h) that must filled with file data. Then, the user-defined >callback can fill from zero up to BUFSIZ bytes, which are eventually >trasmitted to the server. Dillo has a mechanism to read chunks of data from different sources as they are arriving and pass them to the next stage for processing. However, AFAIK it always reads a chunk and appends it to a large buffer. It doesn't free the processed part until is done with the whole thing. This would require a change in the way Dillo processes data, but I think it would be required for large files. There are more details in the devdoc/CCCwork.txt file and in src/chain.c if you want to take a closer look. As I'm planning to change the design of the CCC, I think I can take this into account too so it would be doable. I'll add it to the list of shortcomings of the current design. >That said, I am still not sure how much actual effort this would take. >But I am glad to receive positive feedback so far - I will then continue >to find a solution. > >> However, being able to upload multiple files at the same time sounds reasonable, so feel free to try on your own in the meanwhile. > >Uploading multiple files at once seems doable - the patches I sent on my >previous email are probably already doing most of the required work. >Again, the trickiest task is to send data on-the-fly for each selected file. Okay, I'll focus on the boundary patch first, which is the easiest to merge and then I'll take a closer look at the others. > >> Shouldn't it be 68 then? > >I understand the opposite: the boundary string with the two leading >dashes ("--") included can be up to 72 bytes long, and 74 bytes long for >the ending boundary (which includes two more dashes after the boundary >string). This is confirmed by reading the BNF defined by RFC 2046 (some >bits omitted for simplicity), section 5.1.1 [1]: > >> boundary := 0*69<bchars> bcharsnospace >> bchars := bcharsnospace / " " >> bcharsnospace := DIGIT / ALPHA / "'" / "(" / ")" / >> "+" / "_" / "," / "-" / "." / >> "/" / ":" / "=" / "?" >> dash-boundary := "--" boundary >> ; boundary taken from the value of >> ; boundary parameter of the >> ; Content-Type field> multipart-body := [preamble CRLF] >> dash-boundary transport-padding CRLF >> body-part *encapsulation >> close-delimiter transport-padding >> [CRLF epilogue] >> delimiter := CRLF dash-boundary >> close-delimiter := delimiter "--" Oh right! I see that we are already using 70 characters anyway. >Note: even if the specification tells receivers to handle transport >padding, for the time being I am assuming "transport-padding" as zero >length since composers must not generate non-zero length transport >padding. I am still not sure where transport padding would apply, >anyway. Probably outside web browsers? > >> I would leave out all the symbols to avoid quoting and only use A-Z a-z and 0-9. > >Interestingly, Dillo would always quote boundary strings [2], even if >only using A-Z, a-z and 0-9. In fact, this is one of the wrong >assumptions I spotted when testing slcl against Dillo. Yeah, I would assume a lot of implementations are broken, so we want to try to minimize the chances we run into problems. Apart from slcl we should also test this with some sites and see if they continue to work okay. This will also increase the fingerprinting information to distinguish Dillo among other browsers, but I think it is not more information that the already leaked by the user agent. >> Which, if I computed it correctly, is still too small to worry about. > >Not only it is too small of a chance: if we really wanted to do "the >right thing" and make Dillo absolutely sure the boundary string is not >contained within the selected files, this would imply a noticeable >performance impact when dealing with large files, much likely for a >near-zero benefit. > >I have not inspected their source code yet (and I do not want to), but I >understand both Gecko and Chromium are also making that assumption, >because otherwise it would take them a lot of CPU time to upload large >files. But then they would be doing such assumption with a "much larger" probability it hits the file. Skipping it with 70 characters is safe for one file, but also probably safe for all files ever uploaded with Dillo. Maybe curl or other small codebases are easier to read, but not really needed. > >> Why sizeof " " instead of just 2? > >Because, to my eyes, sizeof " " has more meaningful semantics, compared >to a magic integer constant such as 2. However, for this simple >scenario, I would still consider both acceptable. Check sizeof " ": https://godbolt.org/z/7Tso8ooYz You can also use dStr_append_c() to only append one character, so you only need a single character. If we only use alphanumeric characters, we can just use isalnum() right? >I can replace it with 2 if you find the other construct unacceptable. > >> PS: When are you playing? > >Sorry, I did not understand your last sentence. Could you please give a >bit more context? :) I meant when is the next KoVoꓘ concert :-) Best, Rodrigo.

Aug. 28, 2024

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

by Theo Buehler

Thanks to both of you for tracking this down. I pulled the fix into the OpenBSD port for now.

Aug. 28, 2024