• EdMaste
• ToDo

Current list

• Modify Clang/LLD/etc. build to generate Version.inc files or similar rather than hardcoding (e.g. r305088)

Outstanding - work ($)

Tags of the form reqst######## are bug IDs in $oldwork's bug tracking system.

Old current list

1. Performance test and commit netmap stats change (netperf cluster)
2. Netmap locking issue reported by witness, on detach
3. FreeBSD/NetBSD diff reduction, for OVS port
   1. route.h RT_ADVANCE etc.

Doc issues

1. exit(3) man page return value guidance - too much exit(-1) in proprietary code

Solution is understood, work just needs to be done

1. BSD licensed smarthost delivery mini MTA

   • Most likely case - dma from DragonFly

     • https://github.com/corecode/dma (current master repo, github)

     • http://gitorious.org/dma (old gitorious repo)

     • http://gitweb.dragonflybsd.org/dragonfly.git/tree/master:/libexec/dma (in DragonFlyBSD)

     • Outdated version currently in ports

       • http://www.freshports.org/mail/dma/

     • Current version compiles on FreeBSD -CURRENT with no change
       • needs trivial stat mtim diff for older releases
     • dma does local delivery, smarthost, or direct MX delivery
       • retry handled for the latter two cases
     • bz@ is working on it
2. Indexed arrays for sh(1) a la ksh/bash -- var[index]=val; echo ${var[index]}
   • Some opposition to this idea - don't want to "bloat" sh(1)
   • I originally had "associative arrays" listed, but indexed arrays are simpler and sufficient for many purposes. Also Bash only added associative arrays recently.
   • zsh supports indexed arrays (but not associative)

3. lagg failover discarding input -> Tushar

   • disconnect master link
   • switch's mac table assigns mac to backup link
   • reconnect master link
   • switch continues to send frames to backup link
   • lagg discards the frames
   • easy hack: allow accept on any port via sysctl setting
   • better fix: allow lagg to send gratuitous ARP when failover active port changes
   • need link state notification input to lagg to do the latter
4. bootcode / vfs.root.mountfrom cleanup
   • patch is waiting in my tree, reviewed by Attilio
     • (phk) [rootdevnames] should be removable by now.
     • dev/md/md.c: rootdevnames[0] = "ufs:/dev/md0";
     • ia64/ia64/sscdisk.c: rootdevnames[0] = "ufs:/dev/sscdisk0";
     • kern/vfs_mount.c:char *rootdevnames[2] = {NULL, NULL};
     • kern/vfs_mount.c: if (!vfs_mountroot_try(rootdevnames[0]))
     • kern/vfs_mount.c: if (!vfs_mountroot_try(rootdevnames[1]))
     • nfsclient/bootp_subr.c: rootdevnames[0] = "nfs:";
     • nfsclient/nfs_diskless.c: rootdevnames[0] = "nfs:";

     • sys/systm.h:extern char *rootdevnames[2]; names of possible root devices

5. TCP support for syslogd
   • plain syslog format over TCP
   • as implemented in syslog-ng
6. schedgraph.py
   • thread names from ps -axH -o paddr,tdaddr,pcomm,tdnam
   • sort by thread name
   • tree view of threads in a process
   • show/hide threads or processes
   • lines between processes in right-hand window don't display in some cases (towards bottom)

   • < > buttons in event window - move to top so you can keep clicking next without moving mouse

   • allow event zoom (magnify selected thread around current time w/o changing overall ticks/pixel)
   • allow absolute ticks, relative ticks, relative time as timestamps
   • add a legend window
   • lend prio event - to thread shows up with just proc name it appears
7. Allow tzsetup to pick UTC or UTC +/- offset zones

   • reqst00227912

8. Kernel core handling infrastructure
   • Need a method to determine, from a kernel core,
     • Path to the kernel that was loaded at the time (i.e., sysctl kern.bootfile)
     • Whether or not the kernel now at that path is identical to the one that was running (e.g. md5 hash of kernel file)

   • http://fxr.watson.org/fxr/ident?im=excerpts;i=kernelname

   • reqst00220678

Further investigation or scoping is necessary before beginning work

1. Teach {k,}gdb to read gzipped symbols files (idea from brooks)

   • Support --compress-debug-sections

2. sysv semaphore locking issues
   • does it need a rewrite?
   • some cleanup has happened here, it might be fine now?
3. Big MD root crashes kernel or loader

   • Make big ramdisk: dd if=/dev/zero of=bsdimage bs=1k count=128k

   • /boot/loader.rc
   • load /boot/kernel/kernel
   • load -t mfs_root /bsdimage

   • JohnBaldwin: fixed by bumping up NKPT usually, and now NKPT on i386 is a real kernel option so you can change it more easily

4. other compat32 issues: (maybe just clean up and contribute from $WORKBSD)
   • libcam
   • libgeom / devstat
   • sys/dev/smbus/smb.c
   • PCIOCGETCONF
5. PR kern/40021 patch to use ld(1) to include a prebuilt md root filesystem instead of writing to a preallocated region

6. serial gdb & conserver improvements

   • conserver, or pty<->conserver connector

   • conserver probably needs a 'passthrough' mode; ^Ec parsing etc. is handled by the server
7. valgrind support

8. hwpmc new CPUs, update/release/incorporate into pmcannotate $WORKBSD profiler

   • adrian looking at our profiler src
   • Ryan also

   • http://code.google.com/p/sprofiler/

9. mount behaviour

   • at mountroot> ufs:ufs/freebsdvm mount -uw / filesystem becomes dirty but does not go rw

   • looks like update changes clean flag and last-mounted string, but then fails when ufs/freebsdvm instead of /dev/ufs/freebsdvm shows up
10. spb perf / FCAPS / 1UX7
11. Audit sys/net/if.c:if_grow() and callers for potential locking issues/race conditions
    • had been fixed in $WORKBSD in the 5.x timeframe, but maybe the race is gone or much less likely now?
12. caching DNS proxy and small DNS server
    • provide hostnames for local network
    • basically what you might expect on a small home router
    • like dnsmasq
    • BSD license would be a plus

13. <@JohnBaldwin> emaste: we probably need a separate BSD.32.dist

14. Conflict between utrace records in ld-elf32 and 64-bit kdump
15. pam_diameter - use diameter for AAA, like pam_radius/pam_tacacs
    • Does anyone do this?
16. bsdtar -U option behaviour
    • reqst00174629

    • create a symlink dirlink -> /some/other/path

    • create a tar file containing directory dirlink

    • tar -x -P -U file.tar
    • 6.1 left the symlink in place; 6.3 replaces with a directory
17. ifconfig w/ ether and inet config

    • bin/123633 http://www.freebsd.org/cgi/query-pr.cgi?pr=123633&cat=bin

18. em(4) + vlan + polling issue

    • kern/118695 http://www.freebsd.org/cgi/query-pr.cgi?pr=kern/118695

19. lagg / IPv6

    • kern/122954 http://www.freebsd.org/cgi/query-pr.cgi?pr=kern/122954

    • kern/137089 http://www.freebsd.org/cgi/query-pr.cgi?pr=kern/137089

20. First line of vmstat output is incorrect (reqst00198101)
21. Audit ifnet reference counting -- if_ref / vlan / lagg / carp / ...
22. WITNESS for userland (pthreads)

23. kexec (named kload in FreeBSD: http://lists.freebsd.org/pipermail/freebsd-current/2012-November/037906.html)

24. ksplice (http://www.ksplice.com/doc/ksplice.pdf)

Outstanding - Somewhat work related

1. Add a nodiskless keyworkd to startup scripts and have /etc/rc not run those for diskless machines
2. Multi device support for loader(8) -- allow simultaneous disk, tftp, nfs, httpfs, etc.
   • Sam said he might look at this
3. HTTP support for pxe loader
   • integrate taleks' work from previous GSoC
4. Database-backed or in memory DHCP server
   • for booting diskless machines in a test lab
   • be able to change next-server, root-path etc. per host, without rewriting config file and restarting
5. Add support for loading FreeBSD kernels to QEMU, like it does with Linux kernels
   • direct loading, not from disk image or PXE
6. Automatically unset the boot.nfsroot variables,
   • whenever the loader is compiled with LOADER_TFTP_SUPPORT?
   • if a root md(4) filesystem is loaded?
   • if vfs.root.mountfrom is set and doesn't start with nfs:?
   • via the default loader.conf on the installation media?
   • I don't recall what I was trying to solve here
7. Add SNMP v3 support to bsnmpd
8. Add AgentX support (RFC2741) to bsnmpd
9. Automatically set vfs.root.mountfrom=ufs:/dev/md0 if a root md(4) filesystem is loaded.

Outstanding - personal interest

1. SIGINFO handler for newfs_msdos (percentage progress indicator)
2. vnode backed md is in raw format - allow vmdk (vmware) / qcow2(qemu) etc. format as well
   • or perhaps geom_vmdk, geom_qcow2?
3. CUPS / devfs conflict with lpt0
4. media with sector size != 512 (e.g. 2048 byte USB stick)
5. Offer a prebuilt pxeboot_tftp binary compiled with LOADER_TFTP_SUPPORT=YES so that the loader doesn't need to be recompiled.
   • No point if multi-device support gets done.
6. Add twiddle to fsck
   • jhb mentioned this
   • I thought he had an implementation, but apparently not so
7. Kernel options and modules
   • Can't build ipfw module with forwarding or default to accept
   • sys/netinet/ipfw/ip_fw2.c - #if !defined(KLD_MODULE)
   • looks like a generic problem

Todo after 8.0 branch

1. Clean up src/lib/libstand/nfs.c vs. netbsd src/sys/lib/libsa/nfs.c

   • http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/lib/libsa/nfs.c?only_with_tag=MAIN

   • also rpc.c, net.c http://www.openbsd.org/cgi-bin/cvsweb/src/sys/lib/libsa/net.c?only_with_tag=MAIN

   • and openbsd
2. Add debug message to src/sys/boot/common/load_elf.c (symbols)

Done

1. NFS truncation with non-power-of-two block size (reqst00031351)

   • mount -t nfs -orsize=8000,wsize=8000 localhost:/mnt/src /mnt/dst
     jot -b 1234567890 1000 > /mnt/src/foo

   • wc -l /mnt/src/foo -- returns 1000
   • wc -l /mnt/dst/foo -- returns 698

   • Issue is uio->uio_offset & (biosize - 1) used as modulo for non-power-of-two

   • Committed as r248500

2. SIOCGIFADDR
3. pivot_root (reroot)
   • To support deployable VMs

Abandoned / Overcome by Events

1. dump_write API change to take offset (and add di->mediaoffset itself)

   • makes netdumping slightly cleaner

   • Attilio to include in cleanup of netdump to HEAD

2. NFS client improvements against SOSS
   • SOSS is a Windows NFS v2 server
3. pxeboot / loader / serial console (RELENG_6 fails, HEAD ok)
4. CPU affinity and binding
   • probably the support in 7.x/8.x is all we want
5. Scheduler work for CPU affinity
6. handle panics from within X "somehow"