No further comments, except sorry for taking so long to mention that.

In D47992#1097186, @markj wrote:

The new PR_VNET_ROOT flag is unneeded: it's equivalent to PR_VNET, just check
that instead.

There are currently a wide variety of strings in jail parameters, and for that matter many more elsewhere in the kernel. So far, we have gotten by with counting on administrators putting reasonable values in them.

Accepted with the caveat I mentioned in D48074.

I'd prefer it to take the slot number, and do the "slot - 1" translation in the macro. That came to mind looking at D48075 which passes in slot - 1.

In D47668#1095796, @igoro wrote:

• The allowed chars for each buffer are very limited by default, it covers Base64, k=v\n format, and some extra bytes. It can be changed via security.jail.meta_allowedchars sysctl. For convenience (as it seems to me for now), setting it to an empty string allows everything.

Longer term, I'd like to push some of vnet removal into the period between prison_cleanup() and the final jail destruction; interface removal is definitely part of that desire. But that's a mess of dependencies that hasn't found a solution yet, so I can see how we need this.

I'm surprised the wrong behavior existed for this long without being noticed. I guess that shows how little-used kill(-1) is in usual operation.

In D47668#1089529, @igoro wrote:

Yeah, it seems that on the Jail Call of 26-Nov we came to a conclusion that for now we would keep it very simple like two buffers per jail managed from the user-land side, while keeping a wide spectrum of opportunities to extend it in the future having a more specific production need in mind. Thus, we can postpone thinking about extra complexity on the kernel side.

In D47668#1087001, @igoro wrote:

• The current state of the patch allows reading meta by any user within a jail. Do we want to disable it by default and add something like allow.read_meta or allow.metadata parameter to control it per jail?

In D47668#1087018, @crest_freebsd_rlwinm.de wrote:

A flat list of key=value pairs (instead of a single value) would allow multiple users of this feature per jail e.g. multiple helper commands preparing just one aspect of a jail e.g. dynamic devfs ruleset loading, network setup/teardown, storage provisioning.

Not really part of this specific problem, but jail(8) should auto-recognize list parameters because their SYSCTL_JAIL specification has a type with ",a" appended. Ideally, it would build a list of such parameters it sees like that. For non-kernel parameters like vnet.interface and zfs.dataset, putting them in the array is probably as good as its going to get.

In D47332#1082956, @igoro wrote:

If we add CTLFLAG_PRISON to such sysctlS it will add some mess to the existing interface and behavior, and can add headache for future code. Having such flag added to these variables makes a feeling they can be changed but instead of a correct message that "Operation not permitted" we get something like this:

root@somejail:/ # sysctl security.jail.set_hostname_allowed=1
security.jail.set_hostname_allowed: 0 -> 0

I like the idea of flagging sysctls that hold values that can be read per jail. If jail(2) says otherwise, we can change that to match.

In D47107#1078238, @zlei wrote:

This change is focusing on VNET variables, but is open for -J ( CTLFLAG_PRISON ) if requested.

In D47027#1075329, @markj wrote:

But, the non-null-checked use of the pointer is harmless, as nothing actually dereferences it. Because pr_ip is an array, the value passed to vfs_setopt_part() is the value of the pointer plus the offset of the array within the structure being pointed to.

Sorry I'm late to the game. This actually seems like something should go in. The null dereference was introduced in 500f82d6c32ed, replacing earlier code that was unintuitive but notably didn't need to dereference that pointer. Unfortunately, the fix didn't include a null check, leading to a statement that has one non-null-checked use of the pointer, followed immediately by a properly checked use (because that one predated 500f82d6c32ed).

In D46899#1070761, @olce wrote:

In D46899#1070434, @jamie wrote:

Currently, it seems to suggest (being a jailsys parameter) that there's some sort of valid "new" or "deleted" state for MAC inside the jail.

Not sure what you mean here. SYSCTL_JAIL_PARAM_NODE() just declares the common MAC sub-node. mac_do(4) will then indeed use the new SYSCTL_JAIL_PARAM_SYS_SUBNODE() for the mac.do jail parameter "node", effectively intended to be a jailsys one.

It would make sense for the exec.clean parameter to apply to the config execution. Bit of a chicken and egg problem there, but there's still the "-l" flag.

I suppose this will make more sense once there's at least one MAC policy that has an associated jail parameter. Currently, it seems to suggest (being a jailsys parameter) that there's some sort of valid "new" or "deleted" state for MAC inside the jail.

This comes from discussion in Bug 277210.

In D45647#1041512, @zlei wrote:

I guess the change for sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vfsops.c go to upstream first. Will it ?

Is there any value in a virtual time, i.e. letting a jail have its own clock? Apart from a test framework, I can't think of any, but maybe someone else can. That was actually my first thought when I saw the title of this, rather than encapsulating ntpd.

Diff updated for libsys and other recent changes.

Yes, this is a better way.

I'll admit very little familiarity with the testing framework. If there's a standard to show a test as skipped that doesn't indicate a problem, that sounds best. But if it just claims that it's passed, it would seem better to just not run it. Though both is probably a good idea: still have it pass (like if the test was built at another time), but don't build it on a jail-less system.

Here's the latest diff to address concerns so far (except those that request proper documentation).

In D43696#996617, @bz wrote:

Also, can we please have a man page?

Sure, looks helpful for just such a situation.

Looks good to me!

c) Or something else what I have not spotted yet :)

c) Jamie wasn't thinking and of course you don't need it for read-only.

You'll want to add CTLFLAG_PRISON to the sysctl flags.

In D43476#992839, @igor.ostapenko_pm.me wrote:

@jamie, does it look as an acceptable feature to introduce security.jail.children.max RO sysctl which reflects the current prison's p_childmax? If it does then I will help to implement it.
P.S. Probably you know existing "legal" ways to retrieve current prison's children.max w/o additional code to introduce.

I'm not sure why this limit exists in the first place (it predates me). I suppose it's just for neatness' sake, with the idea no one would have more jails than that anyway. But as long as it's around, it might as well be known.

That's fine if there's a use for it, such as to quiet errors.

In D42672#973897, @rmacklem wrote:

This version of the patch acquires a shared lock on
allprison_lock (which looks sufficient to ensure the
jail does not go from alive to dying.
It also acquires pr_mtx to check for prison_isalive(),
although I am not 100% sure it is needed?

In D42672#973775, @rmacklem wrote:

In D42672#973753, @markj wrote:

Since vfs_exjail_delete() is called when there are no processes running in the jail, a caller of vfs_exjail_clone() cannot be running in that jail.

It's not clear to me that the first part is true. In particular, prison_deref() first calls prison_deref_kill(), which calls prison_cleanup() and thus vfs_exjail_delete(); later it kills processes within the jail.

Hmm, unless I'm missing something, this is at odds with the comment in vfs_exjail_delete(). Perhaps we need to check the prison state when setting mnt_exjail? i.e., refuse to set it if the jail is dying.

Well, the comment on vfs_exjail_delete() states that no processes are in the
prison, but maybe the comment is bogus?
Anyhow, I think this version might be ok, but hopefully you or jamie@ can
confirm this?

My understanding (which could be wrong) is that a prison cannot go from
active to dying when there is a p_uref held on it. If that is the case, then
prison_proc_hold()/prison_proc_free() should ensure that it remains alive
until after the cloning is done, I hope?

The command parameters (including "command" itself) are well established as being run during jail setup and teardown. I would expect a lot of existing configuration to have problems with the command being run when the jail has already started up. While it makes sense on the command line, I don't want to break the connection between file configuration and command line configuration (more than it's already broken).

Nothing new, though I just updated the diff against the latest sources.