1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
|
[[!meta copyright="Copyright © 2011, 2014, 2016 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]]."]]"""]]
Be it statically or dynamically linked, the *startup* of glibc-based programs
is quite hairy on GNU Hurd systems.
[[!taglink open_issue_documentation]], [[!taglink open_issue_glibc]]
* [[!message-id "200103081944.f28JiDk00232@delius.kettenis.local"]]
* [[!message-id "3B7BF2B1.1417CD84@alcor.concordia.ca"]]
* [[!message-id "871xc9qv6y.wl@ulysses.g10code.de"]]
# How libc startup in a process works
## Statically-linked program
* The ELF headers points program start at `_start`.
* `_start` (sysdeps/mach/hurd/i386/static-start.S) calls `_hurd_stack_setup`
* `_hurd_stack_setup` (sysdeps/mach/hurd/i386/init-first.c) calls `first_init` which calls `__mach_init` to initialize enough to run RPCs, then runs the `_hurd_preinit_hook` hooks, which initialize global variables of libc.
* `_hurd_stack_setup` (sysdeps/mach/hurd/i386/init-first.c) calls `_hurd_startup`.
* `_hurd_startup` (hurd/hurdstartup.c) gets hurdish information from servers and calls its `main` parameter.
* the `main` parameter was actually `doinit`, which mangles the stack and calls `doinit1` which calls `init`.
* `init` sets threadvars, tries to initialize threads (and perhaps switches to the new stack) and gets to call `init1`.
* `init1` checks the new stack, calls `_hurd_init` (hurd/hurdinit.c)
* `_hurd_init` initializes initial ports, runs the `_hurd_subinit` hooks (`init_dtable` hurd/dtable.c notably initializes the FD table and the `_hurd_fd_subinit` hooks, which notably checks `std*`).
* We are back to `_start`, which jumps to `_start1` which is the normal libc startup which calls `__libc_start_main`
* `__libc_start_main` (actually called `LIBC_START_MAIN` in csu/libc-start.c) initializes libc, tls, libpthread, atexit
* `__libc_start_main` calls initialization function given as parameter `__libc_csu_init`,
* `__libc_csu_init` (csu/elf-init.c) calls `preinit_array_start` functions
* `__libc_csu_init` calls `_init`
* `_init` (sysdeps/i386/crti.S) calls `PREINIT_FUNCTION`, (actually libpthread on Linux, `__gmon_start__` on hurd)
* back to `__libc_csu_init` (csu/elf-init.c) calls `init_array_start` functions
* back to `__libc_start_main`, it calls calls application's `main`, then `exit`.
## dynamically-linked program
* dl.so ELF headers point its start at `_start`.
* `_start` (sysdeps/i386/dl-machine.h) calls `_dl_start`.
* `_dl_start` (elf/rtld.c) initializes `bootstrap_map`, calls `_dl_start_final`
* `_dl_start_final` calls `_dl_sysdep_start`.
* `_dl_sysdep_start` calls `__mach_init` to initialize enough to run RPCs, then calls `_hurd_startup`.
* `_hurd_startup` (hurd/hurdstartup.c) gets hurdish information from servers and calls its `main` parameter.
* the `main` parameter was actually `go`, which calls `dl_main`.
* `dl_main` (elf/rtld.c) interprets ld.so parameters, loads the binary and libraries, calls `_dl_allocate_tls_init`.
* we are back to `go`, which branches to `_dl_start_user`.
* `_dl_start_user` (./sysdeps/i386/dl-machine.h) runs `RTLD_START_SPECIAL_INIT` (sysdeps/mach/hurd/i386/dl-machine.h) which calls `_dl_init_first`.
* `_dl_init_first` calls `first_init` which calls `__mach_init` to initialize enough to run RPCs, then runs the `_hurd_preinit_hook` hooks, which initialize global variables of libc.
* `_dl_init_first` calls `init`.
* `init` sets threadvars, tries to initialize threads (and perhaps switches to the new stack) and gets to call `init1`.
* `init1` checks the new stack, calls `_hurd_init` (hurd/hurdinit.c)
* `_hurd_init` initializes initial ports, runs the `_hurd_subinit` hooks (`init_dtable` hurd/dtable.c notably initializes the FD table and the `_hurd_fd_subinit` hooks, which notably checks `std*`).
* we are back to `_dl_start_user`, which calls `_dl_init` (elf/dl-init.c) which calls application initializers.
* `_dl_start_user` jumps to the application's entry point, `_start`
* `_start` (sysdeps/i386/start.S) calls `__libc_start_main`
* `__libc_start_main` (actually called `LIBC_START_MAIN` in csu/libc-start.c) initializes libc, atexit,
* `__libc_start_main` calls initialization function given as parameter `__libc_csu_init`,
* `__libc_csu_init` calls `_init`
* `_init` (sysdeps/i386/crti.S) calls `PREINIT_FUNCTION`, (actually libpthread on Linux, `__gmon_start__` on hurd)
* back to `__libc_csu_init` (csu/elf-init.c) calls `init_array_start` functions
* back to `__libc_start_main`, it calls application's `main`, then `exit`.
# IRC, freenode, #hurd, 2013-12-31
<youpi> braunr: btw, after patching glibc and trying installing it, do you
try rebuilding hurd against it? ext2fs.static often breaks due to details
<braunr> youpi: yes
<braunr> youpi: always
<youpi> ok, good :)
<braunr> i want the rootfs to benefit from it too :)
<youpi> heh :)
<braunr> and yes, there were issues that occurred only in the rootfs
<braunr> but because of the special early state, not because of static
linking
|
