** Description changed: gnome-terminal-server (PID 2700) received SIGSEGV while rendering glyphs on Ubuntu 22.04.5. systemd-coredump captured a 3.4 MB core. With libcairo2 debug symbols the fault is in _emit_glyphs_chunk() in src/cairo-xlib-render-compositor.c, writing past a 64-entry XGlyphElt8 stack array. Environment ----------- - OS: Ubuntu 22.04.5 LTS (jammy), X11 session (XDG_SESSION_TYPE=x11), GNOME - Kernel at crash: 6.8.0-138-generic - CPU: AMD Ryzen 5 3600X (12 logical CPUs) - GPU: NVIDIA TU102 [GeForce RTX 2080 Ti Rev. A], driver 550.163.01 - Packages: - libcairo2 1.16.0-5ubuntu2.1 - build-id 92b77d9d96478523ebb65dad17a9373571f4bbe7 - gnome-terminal 3.44.0-1ubuntu1 - libvte-2.91-0 0.68.0-1ubuntu0.1 - Displays: HP 22m on HDMI-0; a portable panel on USB-C-0 (NVIDIA DFP-7, - shown by Xorg as "California institute of Technology TYPE-C"). The USB-C - output re-probes about every 5 seconds. + OS: Ubuntu 22.04.5 LTS (jammy), X11 session (XDG_SESSION_TYPE=x11), GNOME + Kernel at crash: 6.8.0-138-generic + CPU: AMD Ryzen 5 3600X (12 logical CPUs) + GPU: NVIDIA TU102 [GeForce RTX 2080 Ti Rev. A], driver 550.163.01 + Packages: + libcairo2 1.16.0-5ubuntu2.1 + build-id 92b77d9d96478523ebb65dad17a9373571f4bbe7 + gnome-terminal 3.44.0-1ubuntu1 + libvte-2.91-0 0.68.0-1ubuntu0.1 + Displays: HP 22m on HDMI-0; a portable panel on USB-C-0 (NVIDIA DFP-7, + shown by Xorg as "California institute of Technology TYPE-C"). The USB-C + output re-probes about every 5 seconds. Crash ----- - Time: 2026-11-08 08:02:30.918 CET (07:02:30Z) - Process: /usr/libexec/gnome-terminal-server - Signal: 11 (SEGV) - Kernel line: gnome-terminal-[2700]: segfault at 7fff378d8000 - ip 000079c36e376080 sp 00007fff378d0e30 error 4 in - libcairo.so.2.11600.0 + Time: 2026-11-08 08:02:30.918 CET (07:02:30Z) + Process: /usr/libexec/gnome-terminal-server + Signal: 11 (SEGV) + Kernel line: gnome-terminal-[2700]: segfault at 7fff378d8000 + ip 000079c36e376080 sp 00007fff378d0e30 error 4 in + libcairo.so.2.11600.0 With libcairo2 debug symbols: - #0 _emit_glyphs_chunk (...) at src/cairo-xlib-render-compositor.c:1477 - if (_start_new_glyph_elt (j, &glyphs[i])) + #0 _emit_glyphs_chunk (...) at src/cairo-xlib-render-compositor.c:1477 + if (_start_new_glyph_elt (j, &glyphs[i])) Registers at the fault: - r9 0x1000030 num_glyphs = 16,777,264 - r12 0x7fff378d2160 glyphs pointer (consistent with cairo's - 85-entry stack_transformed_glyphs array) - r10/r11 0x28b nelt = 651 elements written - rbp 0x7fff378d0e70 elts = stack_elts (64-entry array) - rcx 0x7fff378d7ff8 current glyph; the +8 read is the fault - fault 0x7fff378d8000 first unmapped byte above the main-thread stack + r9 0x1000030 num_glyphs = 16,777,264 + r12 0x7fff378d2160 glyphs pointer (consistent with cairo's + 85-entry stack_transformed_glyphs array) + r10/r11 0x28b nelt = 651 elements written + rbp 0x7fff378d0e70 elts = stack_elts (64-entry array) + rcx 0x7fff378d7ff8 current glyph; the +8 read is the fault + fault 0x7fff378d8000 first unmapped byte above the main-thread stack Why this looks like a cairo defect ---------------------------------- _emit_glyphs_chunk declares XGlyphElt8 stack_elts[64] (2048 bytes) and uses it when num_elts <= 64 (line 1458). The loop iterates num_glyphs and writes elts[nelt] for every new element (line 1483). nelt is not bounded by the array length. Here num_glyphs was 16,777,264 while num_elts selected the stack path; by the fault the loop had written 651 elements (20,832 bytes) into a 2,048-byte array. The saved registers and saved return address at rsp+0x858..0x888 were overwritten; the saved RIP is stack data (0x7fff378d21e6), so the caller frame cannot be unwound from the core. assert(nelt == num_elts) (line 1507) is compiled out in release builds and would not prevent the writes anyway. The glyph pointer lies in the stack region used for cairo's 85-entry stack_transformed_glyphs array (2048 / sizeof(cairo_glyph_t) = 85), so the original run was likely <= 85 glyphs. The 16.7 M count is inconsistent with that array and is the immediate defect. The XGlyphElt8 records left on the stack contain glyphset=0x1e000a6, the value read from info->glyphset, confirming the writes come from this function. I also checked the cairo 1.18.6 source tarball: _emit_glyphs_chunk still has the same unbounded elts[nelt] loop and no length check against stack_elts. Trigger ------- Xorg re-probed every NVIDIA output at 08:02:30.122-30.171; the HP 22m (DFP-2) and TYPE-C (DFP-7) outputs were re-reported as connected. The crash followed about 0.75 s later. The same re-probe repeats every ~5 s (08:10:39-49, 08:13:59-14:04, ...), so the trigger is still active and recurrence risk remains. The terminal had a heavy VTE child (9 min 45 s CPU in its scope, terminated with the server) and had been open since 07:33. A full redraw during the display reconfiguration is the likely trigger. Reproduction ------------ 1. Use an X11 session with NVIDIA 550.163.01 and a DisplayPort output that - re-probes every few seconds (the USB-C-0 / DFP-7 output on this host). + re-probes every few seconds (the USB-C-0 / DFP-7 output on this host). 2. Run gnome-terminal with continuous output or several tabs and force a - redraw while the output re-probes. + redraw while the output re-probes. 3. Capture the core and symbolicate libcairo2 (debuginfod or - libcairo2-dbgsym), then inspect _emit_glyphs_chunk and the elts array. + libcairo2-dbgsym), then inspect _emit_glyphs_chunk and the elts array. Not deterministic: 2 crashes in about 14 hours. The second occurrence is this report. Evidence -------- Attached: backtrace.txt, crash-analysis.md, hotplug-window.txt, packages.txt, core-sha256.txt. The raw core is not attached because it contains terminal contents and process memory. Requested action ---------------- cairo: confirm the failure mode and whether _emit_glyphs_chunk should bound nelt by the element-array length and/or validate num_glyphs against the caller-provided array. Advise whether a defensive fix is warranted for 1.16.x and Ubuntu 22.04. GNOME Terminal / VTE: check DrawingContext::draw_text_internal (cr_glyphs[MAX_RUN_LENGTH], MAX_RUN_LENGTH=100) for a path that can pass an inconsistent count after a display reconfiguration. Ubuntu: evaluate a backport if upstream identifies a fix; advise on a workaround. Workaround: remove or replace the flapping USB-C output; keep the core for triage. Reporter -------- Gerard Sala https://www.linkedin.com/in/gerard-sala/ https://github.com/scripting-drafts
** Description changed: + I'm afraid I couldn't post this in a single round + gnome-terminal-server (PID 2700) received SIGSEGV while rendering glyphs on Ubuntu 22.04.5. systemd-coredump captured a 3.4 MB core. With libcairo2 debug symbols the fault is in _emit_glyphs_chunk() in src/cairo-xlib-render-compositor.c, writing past a 64-entry XGlyphElt8 stack array. Environment ----------- OS: Ubuntu 22.04.5 LTS (jammy), X11 session (XDG_SESSION_TYPE=x11), GNOME Kernel at crash: 6.8.0-138-generic CPU: AMD Ryzen 5 3600X (12 logical CPUs) GPU: NVIDIA TU102 [GeForce RTX 2080 Ti Rev. A], driver 550.163.01 Packages: libcairo2 1.16.0-5ubuntu2.1 build-id 92b77d9d96478523ebb65dad17a9373571f4bbe7 gnome-terminal 3.44.0-1ubuntu1 libvte-2.91-0 0.68.0-1ubuntu0.1 Displays: HP 22m on HDMI-0; a portable panel on USB-C-0 (NVIDIA DFP-7, shown by Xorg as "California institute of Technology TYPE-C"). The USB-C output re-probes about every 5 seconds. Crash ----- Time: 2026-11-08 08:02:30.918 CET (07:02:30Z) Process: /usr/libexec/gnome-terminal-server Signal: 11 (SEGV) Kernel line: gnome-terminal-[2700]: segfault at 7fff378d8000 ip 000079c36e376080 sp 00007fff378d0e30 error 4 in libcairo.so.2.11600.0 With libcairo2 debug symbols: #0 _emit_glyphs_chunk (...) at src/cairo-xlib-render-compositor.c:1477 if (_start_new_glyph_elt (j, &glyphs[i])) Registers at the fault: r9 0x1000030 num_glyphs = 16,777,264 r12 0x7fff378d2160 glyphs pointer (consistent with cairo's 85-entry stack_transformed_glyphs array) r10/r11 0x28b nelt = 651 elements written rbp 0x7fff378d0e70 elts = stack_elts (64-entry array) rcx 0x7fff378d7ff8 current glyph; the +8 read is the fault fault 0x7fff378d8000 first unmapped byte above the main-thread stack Why this looks like a cairo defect ---------------------------------- _emit_glyphs_chunk declares XGlyphElt8 stack_elts[64] (2048 bytes) and uses it when num_elts <= 64 (line 1458). The loop iterates num_glyphs and writes elts[nelt] for every new element (line 1483). nelt is not bounded by the array length. Here num_glyphs was 16,777,264 while num_elts selected the stack path; by the fault the loop had written 651 elements (20,832 bytes) into a 2,048-byte array. The saved registers and saved return address at rsp+0x858..0x888 were overwritten; the saved RIP is stack data (0x7fff378d21e6), so the caller frame cannot be unwound from the core. assert(nelt == num_elts) (line 1507) is compiled out in release builds and would not prevent the writes anyway. The glyph pointer lies in the stack region used for cairo's 85-entry stack_transformed_glyphs array (2048 / sizeof(cairo_glyph_t) = 85), so the original run was likely <= 85 glyphs. The 16.7 M count is inconsistent with that array and is the immediate defect. The XGlyphElt8 records left on the stack contain glyphset=0x1e000a6, the value read from info->glyphset, confirming the writes come from this function. I also checked the cairo 1.18.6 source tarball: _emit_glyphs_chunk still has the same unbounded elts[nelt] loop and no length check against stack_elts. Trigger ------- Xorg re-probed every NVIDIA output at 08:02:30.122-30.171; the HP 22m (DFP-2) and TYPE-C (DFP-7) outputs were re-reported as connected. The crash followed about 0.75 s later. The same re-probe repeats every ~5 s (08:10:39-49, 08:13:59-14:04, ...), so the trigger is still active and recurrence risk remains. The terminal had a heavy VTE child (9 min 45 s CPU in its scope, terminated with the server) and had been open since 07:33. A full redraw during the display reconfiguration is the likely trigger. Reproduction ------------ 1. Use an X11 session with NVIDIA 550.163.01 and a DisplayPort output that re-probes every few seconds (the USB-C-0 / DFP-7 output on this host). 2. Run gnome-terminal with continuous output or several tabs and force a redraw while the output re-probes. 3. Capture the core and symbolicate libcairo2 (debuginfod or libcairo2-dbgsym), then inspect _emit_glyphs_chunk and the elts array. Not deterministic: 2 crashes in about 14 hours. The second occurrence is this report. Evidence -------- Attached: backtrace.txt, crash-analysis.md, hotplug-window.txt, packages.txt, core-sha256.txt. The raw core is not attached because it contains terminal contents and process memory. Requested action ---------------- cairo: confirm the failure mode and whether _emit_glyphs_chunk should bound nelt by the element-array length and/or validate num_glyphs against the caller-provided array. Advise whether a defensive fix is warranted for 1.16.x and Ubuntu 22.04. GNOME Terminal / VTE: check DrawingContext::draw_text_internal (cr_glyphs[MAX_RUN_LENGTH], MAX_RUN_LENGTH=100) for a path that can pass an inconsistent count after a display reconfiguration. Ubuntu: evaluate a backport if upstream identifies a fix; advise on a workaround. Workaround: remove or replace the flapping USB-C output; keep the core for triage. Reporter -------- Gerard Sala https://www.linkedin.com/in/gerard-sala/ https://github.com/scripting-drafts ** Changed in: cairo (Ubuntu) Status: New => Incomplete -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/2174600 Title: libcairo2 _emit_glyphs_chunk overflows XGlyphElt8 stack_elts (gnome- terminal-server SIGSEGV) To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/cairo/+bug/2174600/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
