1===========================
2Livepatch module Elf format
3===========================
4
5This document outlines the Elf format requirements that livepatch modules must follow.
6
7-----------------
8Table of Contents
9-----------------
100. Background and motivation
111. Livepatch modinfo field
122. Livepatch relocation sections
13   2.1 What are livepatch relocation sections?
14   2.2 Livepatch relocation section format
15       2.2.1 Required flags
16       2.2.2 Required name format
17       2.2.3 Example livepatch relocation section names
18       2.2.4 Example `readelf --sections` output
19       2.2.5 Example `readelf --relocs` output
203. Livepatch symbols
21   3.1 What are livepatch symbols?
22   3.2 A livepatch module's symbol table
23   3.3 Livepatch symbol format
24       3.3.1 Required flags
25       3.3.2 Required name format
26       3.3.3 Example livepatch symbol names
27       3.3.4 Example `readelf --symbols` output
284. Architecture-specific sections
295. Symbol table and Elf section access
30
31----------------------------
320. Background and motivation
33----------------------------
34
35Formerly, livepatch required separate architecture-specific code to write
36relocations. However, arch-specific code to write relocations already
37exists in the module loader, so this former approach produced redundant
38code. So, instead of duplicating code and re-implementing what the module
39loader can already do, livepatch leverages existing code in the module
40loader to perform the all the arch-specific relocation work. Specifically,
41livepatch reuses the apply_relocate_add() function in the module loader to
42write relocations. The patch module Elf format described in this document
43enables livepatch to be able to do this. The hope is that this will make
44livepatch more easily portable to other architectures and reduce the amount
45of arch-specific code required to port livepatch to a particular
46architecture.
47
48Since apply_relocate_add() requires access to a module's section header
49table, symbol table, and relocation section indices, Elf information is
50preserved for livepatch modules (see section 5). Livepatch manages its own
51relocation sections and symbols, which are described in this document. The
52Elf constants used to mark livepatch symbols and relocation sections were
53selected from OS-specific ranges according to the definitions from glibc.
54
550.1 Why does livepatch need to write its own relocations?
56---------------------------------------------------------
57A typical livepatch module contains patched versions of functions that can
58reference non-exported global symbols and non-included local symbols.
59Relocations referencing these types of symbols cannot be left in as-is
60since the kernel module loader cannot resolve them and will therefore
61reject the livepatch module. Furthermore, we cannot apply relocations that
62affect modules not yet loaded at patch module load time (e.g. a patch to a
63driver that is not loaded). Formerly, livepatch solved this problem by
64embedding special "dynrela" (dynamic rela) sections in the resulting patch
65module Elf output. Using these dynrela sections, livepatch could resolve
66symbols while taking into account its scope and what module the symbol
67belongs to, and then manually apply the dynamic relocations. However this
68approach required livepatch to supply arch-specific code in order to write
69these relocations. In the new format, livepatch manages its own SHT_RELA
70relocation sections in place of dynrela sections, and the symbols that the
71relas reference are special livepatch symbols (see section 2 and 3). The
72arch-specific livepatch relocation code is replaced by a call to
73apply_relocate_add().
74
75================================
76PATCH MODULE FORMAT REQUIREMENTS
77================================
78
79--------------------------
801. Livepatch modinfo field
81--------------------------
82
83Livepatch modules are required to have the "livepatch" modinfo attribute.
84See the sample livepatch module in samples/livepatch/ for how this is done.
85
86Livepatch modules can be identified by users by using the 'modinfo' command
87and looking for the presence of the "livepatch" field. This field is also
88used by the kernel module loader to identify livepatch modules.
89
90Example modinfo output:
91-----------------------
92% modinfo livepatch-meminfo.ko
93filename:		livepatch-meminfo.ko
94livepatch:		Y
95license:		GPL
96depends:
97vermagic:		4.3.0+ SMP mod_unload
98
99--------------------------------
1002. Livepatch relocation sections
101--------------------------------
102
103-------------------------------------------
1042.1 What are livepatch relocation sections?
105-------------------------------------------
106A livepatch module manages its own Elf relocation sections to apply
107relocations to modules as well as to the kernel (vmlinux) at the
108appropriate time. For example, if a patch module patches a driver that is
109not currently loaded, livepatch will apply the corresponding livepatch
110relocation section(s) to the driver once it loads.
111
112Each "object" (e.g. vmlinux, or a module) within a patch module may have
113multiple livepatch relocation sections associated with it (e.g. patches to
114multiple functions within the same object). There is a 1-1 correspondence
115between a livepatch relocation section and the target section (usually the
116text section of a function) to which the relocation(s) apply. It is
117also possible for a livepatch module to have no livepatch relocation
118sections, as in the case of the sample livepatch module (see
119samples/livepatch).
120
121Since Elf information is preserved for livepatch modules (see Section 5), a
122livepatch relocation section can be applied simply by passing in the
123appropriate section index to apply_relocate_add(), which then uses it to
124access the relocation section and apply the relocations.
125
126Every symbol referenced by a rela in a livepatch relocation section is a
127livepatch symbol. These must be resolved before livepatch can call
128apply_relocate_add(). See Section 3 for more information.
129
130---------------------------------------
1312.2 Livepatch relocation section format
132---------------------------------------
133
1342.2.1 Required flags
135--------------------
136Livepatch relocation sections must be marked with the SHF_RELA_LIVEPATCH
137section flag. See include/uapi/linux/elf.h for the definition. The module
138loader recognizes this flag and will avoid applying those relocation sections
139at patch module load time. These sections must also be marked with SHF_ALLOC,
140so that the module loader doesn't discard them on module load (i.e. they will
141be copied into memory along with the other SHF_ALLOC sections).
142
1432.2.2 Required name format
144--------------------------
145The name of a livepatch relocation section must conform to the following format:
146
147.klp.rela.objname.section_name
148^        ^^     ^ ^          ^
149|________||_____| |__________|
150   [A]      [B]        [C]
151
152[A] The relocation section name is prefixed with the string ".klp.rela."
153[B] The name of the object (i.e. "vmlinux" or name of module) to
154    which the relocation section belongs follows immediately after the prefix.
155[C] The actual name of the section to which this relocation section applies.
156
1572.2.3 Example livepatch relocation section names:
158-------------------------------------------------
159.klp.rela.ext4.text.ext4_attr_store
160.klp.rela.vmlinux.text.cmdline_proc_show
161
1622.2.4 Example `readelf --sections` output for a patch
163module that patches vmlinux and modules 9p, btrfs, ext4:
164--------------------------------------------------------
165  Section Headers:
166  [Nr] Name                          Type                    Address          Off    Size   ES Flg Lk Inf Al
167  [ snip ]
168  [29] .klp.rela.9p.text.caches.show RELA                    0000000000000000 002d58 0000c0 18 AIo 64   9  8
169  [30] .klp.rela.btrfs.text.btrfs.feature.attr.show RELA     0000000000000000 002e18 000060 18 AIo 64  11  8
170  [ snip ]
171  [34] .klp.rela.ext4.text.ext4.attr.store RELA              0000000000000000 002fd8 0000d8 18 AIo 64  13  8
172  [35] .klp.rela.ext4.text.ext4.attr.show RELA               0000000000000000 0030b0 000150 18 AIo 64  15  8
173  [36] .klp.rela.vmlinux.text.cmdline.proc.show RELA         0000000000000000 003200 000018 18 AIo 64  17  8
174  [37] .klp.rela.vmlinux.text.meminfo.proc.show RELA         0000000000000000 003218 0000f0 18 AIo 64  19  8
175  [ snip ]                                       ^                                             ^
176                                                 |                                             |
177                                                [*]                                           [*]
178[*] Livepatch relocation sections are SHT_RELA sections but with a few special
179characteristics. Notice that they are marked SHF_ALLOC ("A") so that they will
180not be discarded when the module is loaded into memory, as well as with the
181SHF_RELA_LIVEPATCH flag ("o" - for OS-specific).
182
1832.2.5 Example `readelf --relocs` output for a patch module:
184-----------------------------------------------------------
185Relocation section '.klp.rela.btrfs.text.btrfs_feature_attr_show' at offset 0x2ba0 contains 4 entries:
186    Offset             Info             Type               Symbol's Value  Symbol's Name + Addend
187000000000000001f  0000005e00000002 R_X86_64_PC32          0000000000000000 .klp.sym.vmlinux.printk,0 - 4
1880000000000000028  0000003d0000000b R_X86_64_32S           0000000000000000 .klp.sym.btrfs.btrfs_ktype,0 + 0
1890000000000000036  0000003b00000002 R_X86_64_PC32          0000000000000000 .klp.sym.btrfs.can_modify_feature.isra.3,0 - 4
190000000000000004c  0000004900000002 R_X86_64_PC32          0000000000000000 .klp.sym.vmlinux.snprintf,0 - 4
191[ snip ]                                                                   ^
192                                                                           |
193                                                                          [*]
194[*] Every symbol referenced by a relocation is a livepatch symbol.
195
196--------------------
1973. Livepatch symbols
198--------------------
199
200-------------------------------
2013.1 What are livepatch symbols?
202-------------------------------
203Livepatch symbols are symbols referred to by livepatch relocation sections.
204These are symbols accessed from new versions of functions for patched
205objects, whose addresses cannot be resolved by the module loader (because
206they are local or unexported global syms). Since the module loader only
207resolves exported syms, and not every symbol referenced by the new patched
208functions is exported, livepatch symbols were introduced. They are used
209also in cases where we cannot immediately know the address of a symbol when
210a patch module loads. For example, this is the case when livepatch patches
211a module that is not loaded yet. In this case, the relevant livepatch
212symbols are resolved simply when the target module loads. In any case, for
213any livepatch relocation section, all livepatch symbols referenced by that
214section must be resolved before livepatch can call apply_relocate_add() for
215that reloc section.
216
217Livepatch symbols must be marked with SHN_LIVEPATCH so that the module
218loader can identify and ignore them. Livepatch modules keep these symbols
219in their symbol tables, and the symbol table is made accessible through
220module->symtab.
221
222-------------------------------------
2233.2 A livepatch module's symbol table
224-------------------------------------
225Normally, a stripped down copy of a module's symbol table (containing only
226"core" symbols) is made available through module->symtab (See layout_symtab()
227in kernel/module.c). For livepatch modules, the symbol table copied into memory
228on module load must be exactly the same as the symbol table produced when the
229patch module was compiled. This is because the relocations in each livepatch
230relocation section refer to their respective symbols with their symbol indices,
231and the original symbol indices (and thus the symtab ordering) must be
232preserved in order for apply_relocate_add() to find the right symbol.
233
234For example, take this particular rela from a livepatch module:
235Relocation section '.klp.rela.btrfs.text.btrfs_feature_attr_show' at offset 0x2ba0 contains 4 entries:
236    Offset             Info             Type               Symbol's Value  Symbol's Name + Addend
237000000000000001f  0000005e00000002 R_X86_64_PC32          0000000000000000 .klp.sym.vmlinux.printk,0 - 4
238
239This rela refers to the symbol '.klp.sym.vmlinux.printk,0', and the symbol index is encoded
240in 'Info'. Here its symbol index is 0x5e, which is 94 in decimal, which refers to the
241symbol index 94.
242And in this patch module's corresponding symbol table, symbol index 94 refers to that very symbol:
243[ snip ]
24494: 0000000000000000     0 NOTYPE  GLOBAL DEFAULT OS [0xff20] .klp.sym.vmlinux.printk,0
245[ snip ]
246
247---------------------------
2483.3 Livepatch symbol format
249---------------------------
250
2513.3.1 Required flags
252--------------------
253Livepatch symbols must have their section index marked as SHN_LIVEPATCH, so
254that the module loader can identify them and not attempt to resolve them.
255See include/uapi/linux/elf.h for the actual definitions.
256
2573.3.2 Required name format
258--------------------------
259Livepatch symbol names must conform to the following format:
260
261.klp.sym.objname.symbol_name,sympos
262^       ^^     ^ ^         ^ ^
263|_______||_____| |_________| |
264   [A]     [B]       [C]    [D]
265
266[A] The symbol name is prefixed with the string ".klp.sym."
267[B] The name of the object (i.e. "vmlinux" or name of module) to
268    which the symbol belongs follows immediately after the prefix.
269[C] The actual name of the symbol.
270[D] The position of the symbol in the object (as according to kallsyms)
271    This is used to differentiate duplicate symbols within the same
272    object. The symbol position is expressed numerically (0, 1, 2...).
273    The symbol position of a unique symbol is 0.
274
2753.3.3 Example livepatch symbol names:
276-------------------------------------
277.klp.sym.vmlinux.snprintf,0
278.klp.sym.vmlinux.printk,0
279.klp.sym.btrfs.btrfs_ktype,0
280
2813.3.4 Example `readelf --symbols` output for a patch module:
282------------------------------------------------------------
283Symbol table '.symtab' contains 127 entries:
284   Num:    Value          Size Type    Bind   Vis     Ndx         Name
285   [ snip ]
286    73: 0000000000000000     0 NOTYPE  GLOBAL DEFAULT OS [0xff20] .klp.sym.vmlinux.snprintf,0
287    74: 0000000000000000     0 NOTYPE  GLOBAL DEFAULT OS [0xff20] .klp.sym.vmlinux.capable,0
288    75: 0000000000000000     0 NOTYPE  GLOBAL DEFAULT OS [0xff20] .klp.sym.vmlinux.find_next_bit,0
289    76: 0000000000000000     0 NOTYPE  GLOBAL DEFAULT OS [0xff20] .klp.sym.vmlinux.si_swapinfo,0
290  [ snip ]                                               ^
291                                                         |
292                                                        [*]
293[*] Note that the 'Ndx' (Section index) for these symbols is SHN_LIVEPATCH (0xff20).
294    "OS" means OS-specific.
295
296---------------------------------
2974. Architecture-specific sections
298---------------------------------
299Architectures may override arch_klp_init_object_loaded() to perform
300additional arch-specific tasks when a target module loads, such as applying
301arch-specific sections. On x86 for example, we must apply per-object
302.altinstructions and .parainstructions sections when a target module loads.
303These sections must be prefixed with ".klp.arch.$objname." so that they can
304be easily identified when iterating through a patch module's Elf sections
305(See arch/x86/kernel/livepatch.c for a complete example).
306
307--------------------------------------
3085. Symbol table and Elf section access
309--------------------------------------
310A livepatch module's symbol table is accessible through module->symtab.
311
312Since apply_relocate_add() requires access to a module's section headers,
313symbol table, and relocation section indices, Elf information is preserved for
314livepatch modules and is made accessible by the module loader through
315module->klp_info, which is a klp_modinfo struct. When a livepatch module loads,
316this struct is filled in by the module loader. Its fields are documented below:
317
318struct klp_modinfo {
319	Elf_Ehdr hdr; /* Elf header */
320	Elf_Shdr *sechdrs; /* Section header table */
321	char *secstrings; /* String table for the section headers */
322	unsigned int symndx; /* The symbol table section index */
323};
324