summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorJeremie Koenig <jk@jk.fr.eu.org>2010-08-25 19:15:35 +0000
committerJeremie Koenig <jk@jk.fr.eu.org>2010-08-30 15:14:35 +0200
commite25cfb3f3a843122df7e70e97992a97ffe81717a (patch)
tree99433172d3e05e2f3b1a77e4f46c49901c6928b4
parent0f3c9c2640ffd76f1d60f16dce6d16c47aac98b4 (diff)
Update the to-do list.
* TODO: Add more entries.
-rw-r--r--TODO25
1 files changed, 24 insertions, 1 deletions
diff --git a/TODO b/TODO
index 241c61a..952d67b 100644
--- a/TODO
+++ b/TODO
@@ -1 +1,24 @@
-* Threads in /proc/[pid]/task/[n]
+Known bugs to be fixed
+----------------------
+
+* The non-owned processes sometimes show up with INT_MAX as their owner,
+ instead of opt_anon_uid. This is likely to be a libps problem.
+
+Improvements and new features
+-----------------------------
+
+* There is a lot of dynamic memory allocation going on and it comes with a
+ cost in performance. We could try to limit such allocation, as long as it
+ keeps the inner interface simple and preserves the read/readdir semantics
+ (performance is probably not critical for a proc filesystem.)
+ One way would be to add an (optional) "needed_length" field to
+ procfs_node_ops, and arrange to pass a sufficent buffer in (*contents,
+ *contents_len) when get_contents is called. Then the user-provided buffer
+ might be used directly under some circumstances.
+
+* Add thread directories as [pid]/task/[n]. This shouldn't be too hard if we
+ use "process" nodes for threads, and provide an "exists" hook for the "task"
+ entry itself so that it's disabled in thread nodes. It might prove necessary
+ to have "optional" libps flags for some content generators, though, since
+ some of them might be missing for threads.
+