summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorSamuel Thibault <samuel.thibault@ens-lyon.org>2020-04-01 14:10:42 +0200
committerSamuel Thibault <samuel.thibault@ens-lyon.org>2020-04-01 14:10:42 +0200
commit92dc9670123d712324d370337fea8796088214b6 (patch)
treea8d035ca710456e7146cee851129cfd1e5d75996
parentb25d79a92fe9105506363ebd59f2330bc1799814 (diff)
Add thinking about running rump for the / filesystem
-rw-r--r--open_issues/running_rump_for_slash.mdwn53
1 files changed, 53 insertions, 0 deletions
diff --git a/open_issues/running_rump_for_slash.mdwn b/open_issues/running_rump_for_slash.mdwn
new file mode 100644
index 00000000..52c97b5d
--- /dev/null
+++ b/open_issues/running_rump_for_slash.mdwn
@@ -0,0 +1,53 @@
+[[!meta copyright="Copyright © 2019 Free Software
+Foundation, Inc."]]
+
+[[!meta license="""[[!toggle id="license" text="GFDL 1.2+"]][[!toggleable
+id="license" text="Permission is granted to copy, distribute and/or modify this
+document under the terms of the GNU Free Documentation License, Version 1.2 or
+any later version published by the Free Software Foundation; with no Invariant
+Sections, no Front-Cover Texts, and no Back-Cover Texts. A copy of the license
+is included in the section entitled [[GNU Free Documentation
+License|/fdl]]."]]"""]]
+
+[[!tag open_issue_hurd]]
+
+I have been thinking about how to get rump running for the / filesystem.
+
+Looking at how things go between ext2fs and exec: in grub.cfg we have
+roughly:
+
+ module ext2fs --exec-server-task='${exec-task}' '$(task-create)' '$(task-resume)'
+ module exec '$(exec-task=task-create)'
+
+i.e. the kernel is told to create two tasks, to pass a reference to
+the exec task to the ext2fs task, and to let only the ext2fs task to
+run. What happens then is in `diskfs_start_bootstrap`, which calls
+`start_execserver`, which uses `task_set_special_port` to set the
+`TASK_BOOTSTRAP_PORT` special port to a send right to ext2fs, and resumes
+the exec task. I.e. basically ext2fs tells exec where it is so that exec
+can start the userland with `/` available.
+
+I'm thinking that the same can be used for the rump translator,
+something like:
+
+ module rump --fs-server-task='${fs-task}' '$(task-create)' '$(task-resume)'
+ module ext2fs --exec-server-task='${exec-task}' '$(fs-task=task-create)'
+ module exec '$(exec-task=task-create)'
+
+and we'd make rump's initialization use `task_set_special_port` to set
+the `TASK_BOOTSTRAP_PORT` special port of ext2fs to a send right to rump,
+and resume it. When ext2fs sees that this port is set, it would use it
+instead of the gnumach-provided `_hurd_device_master` port to open
+devices.
+
+And we can nest this yet more for the pci-arbiter:
+
+ module pci-arbiter --disk-server-task='${disk-task}' '$(task-create)' '$(task-resume)'
+ module rump --fs-server-task='${fs-task}' '$(disk-task=task-create)'
+ module ext2fs --exec-server-task='${exec-task}' '$(fs-task=task-create)'
+ module exec '$(exec-task=task-create)'
+
+and we'd make `pci-arbiter`'s initialization use `task_set_special_port`
+to set the `TASK_BOOTSTRAP_PORT` special port of rump to a send right to
+pci-arbiter and resume it. When `libpciaccess` sees that this port is set,
+it would use it instead of looking up `/server/bus/pci`.