only create nsh database in privileged mode Avoids noisy warnings on startup as a regular user when the database does not exist yet.

ensure that build timestamp is kept up-to-date on each build

make -current builds display build timestamp and compilation info in greeting Suggested by Tom

generate man command tab-completion table with a script at build-time This prevents the table from becoming out of date going forward. There were a few tags in nsh.8 which weren't yet being displayed: + { "inet6", "Search for tag inet6", CMPL0 NULL, 0 }, + { "powerdown", "Search for tag powerdown", CMPL0 NULL, 0 }, + { "saveenv", "Search for tag saveenv", CMPL0 NULL, 0 }, And an error has been corrected: - { "unsetenv", "Search for tag setenv", CMPL0 NULL, 0 }, + { "unsetenv", "Search for tag unsetenv", CMPL0 NULL, 0 },

add a dhcp restart command; requested by Tom

fix sqlite API error handling - call sqlite3_errmsg(db) before closing the db, not after - sqlite3_errmsg() can only be used when sqlite3_step() returns SQLITE_ERROR - check for errors from sqlite3_finalize() - close db handle even when an error occurred during sqlite3_open() with suggestions from chris testing + ok tom

Fix dhcpd handling with respect to rdoamins Run dhcpd in the correct rdomain when rdomain is != 0 and start dhcpd with a list of interfaces on its command line. All interfaces must be in the same rdomain; dhcpd will then set its own rdomain accordingly. Use a separate DHCP lease database per rdomain just in case a given MAC and IP tuple appears in multiple rdomains. earlier version ok + testing by Tom

allow T_HANDLER to process any number of arguments between 0 and 6 This way we can write handlers that accept all arguments listed in their ctl table as actual arguments, rather than passing unused parameters which contain ununitialized garbage. The benefit is better clarity about the expected values of arguments passed on by handler functions. Adjust motd and crontab handlers accordingly which use T_HANDLER. All instances of call_editor() now use T_HANDLER_FILL1. ok chris

fix conversion of struct ctl2->table to struct ctl->table The old code would only convert the first command table entry to a dummy struct ctl sitting on the stack. This confuses code which loops over the command table and requires a sentinel to break out of this loop. Segfault found by Tom

detect ambiguous matches from genget on a ctl2 table Cannot happen right now since the table has only one entry but better be prepared for a future where we would crash otherwise.

repair "write-config" command which broke with introduction of struct ctl2 Writing any changes to dhcp config was causing "genget internal failure". Regression found by Tom.

always keep basic environment variables intact after 'enable' command Make nshdoas preserve some important environment variables even when there is no rule in /etc/doas.conf. Without a matching rule in doas.conf we ended up running the privileged shell in an empty environment, which leads to tools such as vi(4) erroring out for lack of a terminal type. As before, the keepenv option in /etc/doas.conf can be used to preserve the entire user environment. We now preserve the following variables by default: DOAS_USER=nsh # added by nshdoas HOME=/root # overridden by nshdoas LOGNAME=root # overridden by nshdoas PATH=/usr/bin:/bin:/usr/sbin:/sbin:/usr/X11R6/bin:/usr/local/bin:/usr/local/sbin SHELL=/usr/local/bin/nsh TERM=xterm USER=root # overridden by nshdoas Problem noticed by Tom.

introduce struct ctl2 with backwards compat wrappers The new struct avoids the necessity to cast configuration file test handler arguments from char ** to a char *, such that the char ** pointer can be stored in ctl->args[1]. Such shadowing of variable types results in code that is hard to follow. We can convert daemons one-by-one to fully replace struct ctl eventually. As an example, convert dhcpd handling to struct ctl2 already. Ok tom, chris

Merge pull request #143 from smytht/master Improve Readme by converting to markdown and adding a quick installation guide for nsh

Update README.md use email addresses proper Update README.md use email addresses proper

backout previous dhcp rdomain commits I forgot that these diffs only work correctly with the &fillargs patch which has been backed out. This needs more work before it can come back...

Update README.md fix typo fix typo screen not scree

use a separate DHCP lease database per rdomain ok Tom

Update README.md fix number formats Update README.md fix number formats

run dhcpd in the correct rdomain when rdomain is != 0 test + ok Tom

Update README.md fix typos in and after section 7 Update README.md fix typos in and after section 7

Update and rename nsh-openbsd-integration.sh to extensive-nsh-openbsd-integration.sh Add in extensive warnings to indicate this is a more invasive installation script, (for users who mainly want to manage systems with nsh) (moving config files for import not copying to avoid confusion between which config files are the source of truth, added in a write configuration command at end of nsh configuration to capture the config before reboot.

Update README.md to include the detail of requirements Update README.md to include the detail of requirements for nsh to control the system

Update motd to explain how nsh has integrated with OpenBSD add

back out "fix bogus pointer being passed to fill_tmpfile() for T_HANDLER_FILL1" This backs out commit 1bb61b1da84f5690f3125d4f6595b833632622b4: The commit broke configuration file syntax checks. It triggeed errors from exec since we were trying to exec a list of garbage arguments. I misunderstood what the fillargs[1] pointer represents. Even though it is a char * the value stored in case of FILLARGS1 is in fact a char *[].