Commit f0a8ac84 authored by Vikram Narayanan's avatar Vikram Narayanan

update notes and bad-patterns from pmfs tree

Signed-off-by: Vikram Narayanan's avatarVikram Narayanan <>
parent 1f87043c
1 - A pointer points to a struct embedded inside another struct. It's difficult
or impossible to wrap that other struct in a container, because it changes
the layout of the struct it's inside. (It only makes sense to have one
container per memory chunk.) However, it's then hard for glue to know
whether the struct is embedded or not.
2 - The callee might try to use "current", including mm related stuff. This
assumes that the calling thread is the same as the original thread that
started off the call sequence. This may not be true if we have other threads
handling calls/replies. It may not be simple to switch "current", so it might
require some code modification of callee.
3 - The caller may pass a global variable as an argument to a function. But
the global variable is defined on the callee side. Could add a "getter". But
when does callee insert global variable into caller's cspace? Perhaps it checks
if ref is non-null already (insert already done).
4 - Tracking the percolation of shared, stateful objects is hard.
5 - Handling error scenarios is hard. For example, consider LCD A and B. B
does an rpc to A, and something happens inside A that causes an error. So
A returns an error status to B. Perhaps B wants to do *additional* rpc's
on A because of the error. A needs to be ready to handle those error-scenario
rpc's, or else B will just sit there waiting.
......@@ -78,5 +78,6 @@ EXPORT_SYMBOL(_lcd_create_sync_endpoint);
U arch_get_unmapped_area_sz
U autoremove_wake_function
U boot_cpu_data
U call_rcu_sched
X REDEF DONE U call_rcu_sched
U capable
U clear_inode
X GLUE DONE U clear_inode
U clear_nlink
U __copy_user_nocache
U current_kernel_time
U d_instantiate
X GLUE U d_make_root
X GLUE DONE U d_make_root
U d_obtain_alias
U drop_nlink
U d_splice_alias
......@@ -27,8 +27,8 @@ X GLUE U d_make_root
X REDEF DONE U get_random_bytes
X REDEF DONE U get_seconds
U __get_user_4
X GLUE U iget_failed
X GLUE U iget_locked
X GLUE DONE U iget_failed
X GLUE DONE U iget_locked
U ihold
U inc_nlink
U init_special_inode
......@@ -45,9 +45,9 @@ X LIBLCD DONE U ioremap_cache_ro
X LIBLCD DONE U ioremap_hpage_cache
X LIBLCD DONE U ioremap_hpage_cache_ro
X GLUE U iput
U iput
U is_bad_inode
X GLUE U kill_anon_super
X GLUE DONE U kill_anon_super
X NO-OP DONE U kthread_create_on_node
X NO-OP DONE U kthread_should_stop
X NO-OP DONE U kthread_stop
......@@ -58,7 +58,7 @@ X SUCK DONE U match_token
X SUCK DONE U memparse
U mnt_drop_write_file
U mnt_want_write_file
X GLUE U mount_nodev
X GLUE DONE U mount_nodev
U new_inode
U noop_fsync
U panic
......@@ -69,7 +69,7 @@ X GLUE U mount_nodev
U ___ratelimit
U _raw_spin_lock
U _raw_spin_unlock
X NO-OP DONE U __release_region XXX: Non-isol will need to grant during mount
X NO-OP DONE U __release_region
X NO-OP DONE U __request_region
U __sb_end_write
U __sb_start_write
......@@ -82,8 +82,8 @@ X NO-OP DONE U __request_region
U set_nlink
U __supported_pte_mask
U touch_atime
U truncate_inode_pages
X GLUE U unlock_new_inode
X GLUE DONE U truncate_inode_pages (shouldn't do much)
X GLUE DONE U unlock_new_inode
U vfs_follow_link
U vfs_readlink
U vm_insert_mixed
......@@ -96,5 +96,33 @@ X NO-OP DONE U wake_up_process
current fsuid, fsgid, ns --> elide DONE
glue for super ops -> put_super
glue for mount fn pointer
\ No newline at end of file
glue for super ops
alloc_inode DONE
destroy_inode DONE
put_super DONE
evict_inode DONE
-- assume i links > 0, truncate list is empty
glue for fs type DONE
-- populate fn pointers in register_filesystem DONE
-- kill fn pointers in unregister_filesystem DONE
mount DONE
-- need to volunteer fs phys mem DONE
kill_sb DONE
-- this will unvolunteer fs phys mem DONE
-- kill super_block trampolines after kill_sb rpc DONE
glue for pmfs_fill_super DONE
-- this is where we'll populate super ops DONE
Also todo: make sure containers are set up properly
-- modify super_block alloc/free DONE
-- modify dentry alloc/free DONE
TODO: Need to marshal module crap over in register fs DONE
Not going to have user level do mount call, because too complicated
to interface user thread with the dispatch loop on non-isolated side.
Just have non-isolated side call mount system call or something, same
for umount. Maybe. May not work - maybe only real threads (non-kthreads)
are allowed to call mount, since kthreads may not have a root dir.
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment