Re: two approaches to file tree recursion with *at() syscalls

On Sun, 20 Jul 2008 06:06:26 -0700 John Reiser <jreiser@xxxxxxxxxxxx> wrote:
|> | Is the file graph contained in a single block device (no mounts, no
|> | off-device symlinks)?
|>
|> No guarantees. [For this or any other "nasty" condition.]
|
| In such cases then neither of the two proposed recursion solutions (keep open
| directories on path from root; use just one open directory, plus "..")
| is safe to walk the graph. Because the graph may span devices,
| may contain arbitrary loops, may be mal-formed, may be on a filesystem
| with a different notion of "directory", and may be changing as you walk,
| then you must keep a history of the device number, inode number, name,
| and creation+modification times of each entry, to help you navigate and
| detect and avoid getting lost in loops. Even then, a clever concurrent
| adversary can force an infinite traversal within a finite graph,
| with probability greater than zero.

I know that neither method is perfect. If one of them was, I'd use that.
What I want to figure out is which method would be more appropriate to use
in general for most cases, or in a specific case.

I'm writing a program to do file copying of a tree to another. Mostly it is
intended as a fast and ordered copy program. But I am also making it handle
target replacement, which can involve deleting a directory subtree at the
target, as well. So the code could have tree walk both for copying source
to target, as well as deletion of (parts of) the target.

Maybe the specific case(s) will be better served by one of these methods.

The uses I would apply this program to would not have the tree changing
during the copying (or of the deletion of a target). Someone else might
try to do that. It is possible that the source tree could span filesystem
mounts. I will be tracking device and inode already as part of replicating
the source's hardlink relationships into the target. It seems that mounts
simulate fake ".." entries that always go to the parent of the mount point,
or to "/" itself for the "/" mount point.

I'm contemplating some kind of hybrid method that uses an open descriptor of
each directory level, then switch to the other method of closing directories
and reopen them coming back up when descriptors run out. The catch with that
is I need a few reserve descriptors, so I need to transition between methods
before actually running out. I'll need at least 3 reserve descriptors. I'd
have to trust sysconf(_SC_OPEN_MAX) or something to count them on my own.

--
|WARNING: Due to extreme spam, googlegroups.com is blocked. Due to ignorance |
| by the abuse department, bellsouth.net is blocked. If you post to |
| Usenet from these places, find another Usenet provider ASAP. |
| Phil Howard KA9WGN (email for humans: first name in lower case at ipal.net) |
.

Relevant Pages

  • Re: SNMP and Cisco
    ... There are a couple of things you should do on MRTG. ... # * Lines which follow a keyword line which do start ... # Configuration for each Target you want to monitor ... MRTG needs to graph two values, ...
    (comp.dcom.sys.cisco)
  • Re: Naval Gunfire Question
    ... You may be thinking of the open 5 inch mounts fitted to the CG ... weapons additions, nor, come to think of it, was there such a chse, at ... Burkes "on the fly" for small surface target hunting and arming the ... Well, we did a bunch of that in SEA (and learned that the little cutters, ...
    (sci.military.naval)
  • Re: special graph format
    ... As long as target doesn't change to much per day, ... track this post so I can learn from any responses from the graph experts). ... Change series 1: Format data series: Patterns: No ... If T-line < Q-line surface between Q and T must be red ...
    (microsoft.public.excel.programming)
  • Re: HK Kormoran + HMAS Sydney
    ... one or more decks below the mounts. ... When both ships are hove to at point blank range its not a big deal to ... target sleds, towed sleeves and "Delmars", I'll be quick to assure ViKnce ... KORMORAN turned away for a tail chase, ...
    (sci.military.naval)
  • Re: HK Kormoran + HMAS Sydney
    ... shielded mounts. ... gun fires so rapidly, because we can get on the target before ... When both ships are hove to at point blank range its not a big ... variety of target sleds, towed sleeves and "Delmars", I'll be quick ...
    (sci.military.naval)