Linux kernel mirror (for testing) git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel os linux

Documentation: livepatch: outline Elf format and requirements for patch modules

Document livepatch module requirements and the special Elf constants patch
modules use.

Signed-off-by: Jessica Yu <jeyu@redhat.com>
Acked-by: Miroslav Benes <mbenes@suse.cz>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>

authored by

Jessica Yu and committed by
Jiri Kosina
3b812ecc 425595a7

+311
+311
Documentation/livepatch/module-elf-format.txt
··· 1 + =========================== 2 + Livepatch module Elf format 3 + =========================== 4 + 5 + This document outlines the Elf format requirements that livepatch modules must follow. 6 + 7 + ----------------- 8 + Table of Contents 9 + ----------------- 10 + 0. Background and motivation 11 + 1. Livepatch modinfo field 12 + 2. 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 20 + 3. 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 28 + 4. Symbol table and Elf section access 29 + 30 + ---------------------------- 31 + 0. Background and motivation 32 + ---------------------------- 33 + 34 + Formerly, livepatch required separate architecture-specific code to write 35 + relocations. However, arch-specific code to write relocations already 36 + exists in the module loader, so this former approach produced redundant 37 + code. So, instead of duplicating code and re-implementing what the module 38 + loader can already do, livepatch leverages existing code in the module 39 + loader to perform the all the arch-specific relocation work. Specifically, 40 + livepatch reuses the apply_relocate_add() function in the module loader to 41 + write relocations. The patch module Elf format described in this document 42 + enables livepatch to be able to do this. The hope is that this will make 43 + livepatch more easily portable to other architectures and reduce the amount 44 + of arch-specific code required to port livepatch to a particular 45 + architecture. 46 + 47 + Since apply_relocate_add() requires access to a module's section header 48 + table, symbol table, and relocation section indices, Elf information is 49 + preserved for livepatch modules (see section 4). Livepatch manages its own 50 + relocation sections and symbols, which are described in this document. The 51 + Elf constants used to mark livepatch symbols and relocation sections were 52 + selected from OS-specific ranges according to the definitions from glibc. 53 + 54 + 0.1 Why does livepatch need to write its own relocations? 55 + --------------------------------------------------------- 56 + A typical livepatch module contains patched versions of functions that can 57 + reference non-exported global symbols and non-included local symbols. 58 + Relocations referencing these types of symbols cannot be left in as-is 59 + since the kernel module loader cannot resolve them and will therefore 60 + reject the livepatch module. Furthermore, we cannot apply relocations that 61 + affect modules not yet loaded at patch module load time (e.g. a patch to a 62 + driver that is not loaded). Formerly, livepatch solved this problem by 63 + embedding special "dynrela" (dynamic rela) sections in the resulting patch 64 + module Elf output. Using these dynrela sections, livepatch could resolve 65 + symbols while taking into account its scope and what module the symbol 66 + belongs to, and then manually apply the dynamic relocations. However this 67 + approach required livepatch to supply arch-specific code in order to write 68 + these relocations. In the new format, livepatch manages its own SHT_RELA 69 + relocation sections in place of dynrela sections, and the symbols that the 70 + relas reference are special livepatch symbols (see section 2 and 3). The 71 + arch-specific livepatch relocation code is replaced by a call to 72 + apply_relocate_add(). 73 + 74 + ================================ 75 + PATCH MODULE FORMAT REQUIREMENTS 76 + ================================ 77 + 78 + -------------------------- 79 + 1. Livepatch modinfo field 80 + -------------------------- 81 + 82 + Livepatch modules are required to have the "livepatch" modinfo attribute. 83 + See the sample livepatch module in samples/livepatch/ for how this is done. 84 + 85 + Livepatch modules can be identified by users by using the 'modinfo' command 86 + and looking for the presence of the "livepatch" field. This field is also 87 + used by the kernel module loader to identify livepatch modules. 88 + 89 + Example modinfo output: 90 + ----------------------- 91 + % modinfo livepatch-meminfo.ko 92 + filename: livepatch-meminfo.ko 93 + livepatch: Y 94 + license: GPL 95 + depends: 96 + vermagic: 4.3.0+ SMP mod_unload 97 + 98 + -------------------------------- 99 + 2. Livepatch relocation sections 100 + -------------------------------- 101 + 102 + ------------------------------------------- 103 + 2.1 What are livepatch relocation sections? 104 + ------------------------------------------- 105 + A livepatch module manages its own Elf relocation sections to apply 106 + relocations to modules as well as to the kernel (vmlinux) at the 107 + appropriate time. For example, if a patch module patches a driver that is 108 + not currently loaded, livepatch will apply the corresponding livepatch 109 + relocation section(s) to the driver once it loads. 110 + 111 + Each "object" (e.g. vmlinux, or a module) within a patch module may have 112 + multiple livepatch relocation sections associated with it (e.g. patches to 113 + multiple functions within the same object). There is a 1-1 correspondence 114 + between a livepatch relocation section and the target section (usually the 115 + text section of a function) to which the relocation(s) apply. It is 116 + also possible for a livepatch module to have no livepatch relocation 117 + sections, as in the case of the sample livepatch module (see 118 + samples/livepatch). 119 + 120 + Since Elf information is preserved for livepatch modules (see Section 4), a 121 + livepatch relocation section can be applied simply by passing in the 122 + appropriate section index to apply_relocate_add(), which then uses it to 123 + access the relocation section and apply the relocations. 124 + 125 + Every symbol referenced by a rela in a livepatch relocation section is a 126 + livepatch symbol. These must be resolved before livepatch can call 127 + apply_relocate_add(). See Section 3 for more information. 128 + 129 + --------------------------------------- 130 + 2.2 Livepatch relocation section format 131 + --------------------------------------- 132 + 133 + 2.2.1 Required flags 134 + -------------------- 135 + Livepatch relocation sections must be marked with the SHF_RELA_LIVEPATCH 136 + section flag. See include/uapi/linux/elf.h for the definition. The module 137 + loader recognizes this flag and will avoid applying those relocation sections 138 + at patch module load time. These sections must also be marked with SHF_ALLOC, 139 + so that the module loader doesn't discard them on module load (i.e. they will 140 + be copied into memory along with the other SHF_ALLOC sections). 141 + 142 + 2.2.2 Required name format 143 + -------------------------- 144 + The name of a livepatch relocation section must conform to the following format: 145 + 146 + .klp.rela.objname.section_name 147 + ^ ^^ ^ ^ ^ 148 + |________||_____| |__________| 149 + [A] [B] [C] 150 + 151 + [A] The relocation section name is prefixed with the string ".klp.rela." 152 + [B] The name of the object (i.e. "vmlinux" or name of module) to 153 + which the relocation section belongs follows immediately after the prefix. 154 + [C] The actual name of the section to which this relocation section applies. 155 + 156 + 2.2.3 Example livepatch relocation section names: 157 + ------------------------------------------------- 158 + .klp.rela.ext4.text.ext4_attr_store 159 + .klp.rela.vmlinux.text.cmdline_proc_show 160 + 161 + 2.2.4 Example `readelf --sections` output for a patch 162 + module that patches vmlinux and modules 9p, btrfs, ext4: 163 + -------------------------------------------------------- 164 + Section Headers: 165 + [Nr] Name Type Address Off Size ES Flg Lk Inf Al 166 + [ snip ] 167 + [29] .klp.rela.9p.text.caches.show RELA 0000000000000000 002d58 0000c0 18 AIo 64 9 8 168 + [30] .klp.rela.btrfs.text.btrfs.feature.attr.show RELA 0000000000000000 002e18 000060 18 AIo 64 11 8 169 + [ snip ] 170 + [34] .klp.rela.ext4.text.ext4.attr.store RELA 0000000000000000 002fd8 0000d8 18 AIo 64 13 8 171 + [35] .klp.rela.ext4.text.ext4.attr.show RELA 0000000000000000 0030b0 000150 18 AIo 64 15 8 172 + [36] .klp.rela.vmlinux.text.cmdline.proc.show RELA 0000000000000000 003200 000018 18 AIo 64 17 8 173 + [37] .klp.rela.vmlinux.text.meminfo.proc.show RELA 0000000000000000 003218 0000f0 18 AIo 64 19 8 174 + [ snip ] ^ ^ 175 + | | 176 + [*] [*] 177 + [*] Livepatch relocation sections are SHT_RELA sections but with a few special 178 + characteristics. Notice that they are marked SHF_ALLOC ("A") so that they will 179 + not be discarded when the module is loaded into memory, as well as with the 180 + SHF_RELA_LIVEPATCH flag ("o" - for OS-specific). 181 + 182 + 2.2.5 Example `readelf --relocs` output for a patch module: 183 + ----------------------------------------------------------- 184 + Relocation section '.klp.rela.btrfs.text.btrfs_feature_attr_show' at offset 0x2ba0 contains 4 entries: 185 + Offset Info Type Symbol's Value Symbol's Name + Addend 186 + 000000000000001f 0000005e00000002 R_X86_64_PC32 0000000000000000 .klp.sym.vmlinux.printk,0 - 4 187 + 0000000000000028 0000003d0000000b R_X86_64_32S 0000000000000000 .klp.sym.btrfs.btrfs_ktype,0 + 0 188 + 0000000000000036 0000003b00000002 R_X86_64_PC32 0000000000000000 .klp.sym.btrfs.can_modify_feature.isra.3,0 - 4 189 + 000000000000004c 0000004900000002 R_X86_64_PC32 0000000000000000 .klp.sym.vmlinux.snprintf,0 - 4 190 + [ snip ] ^ 191 + | 192 + [*] 193 + [*] Every symbol referenced by a relocation is a livepatch symbol. 194 + 195 + -------------------- 196 + 3. Livepatch symbols 197 + -------------------- 198 + 199 + ------------------------------- 200 + 3.1 What are livepatch symbols? 201 + ------------------------------- 202 + Livepatch symbols are symbols referred to by livepatch relocation sections. 203 + These are symbols accessed from new versions of functions for patched 204 + objects, whose addresses cannot be resolved by the module loader (because 205 + they are local or unexported global syms). Since the module loader only 206 + resolves exported syms, and not every symbol referenced by the new patched 207 + functions is exported, livepatch symbols were introduced. They are used 208 + also in cases where we cannot immediately know the address of a symbol when 209 + a patch module loads. For example, this is the case when livepatch patches 210 + a module that is not loaded yet. In this case, the relevant livepatch 211 + symbols are resolved simply when the target module loads. In any case, for 212 + any livepatch relocation section, all livepatch symbols referenced by that 213 + section must be resolved before livepatch can call apply_relocate_add() for 214 + that reloc section. 215 + 216 + Livepatch symbols must be marked with SHN_LIVEPATCH so that the module 217 + loader can identify and ignore them. Livepatch modules keep these symbols 218 + in their symbol tables, and the symbol table is made accessible through 219 + module->symtab. 220 + 221 + ------------------------------------- 222 + 3.2 A livepatch module's symbol table 223 + ------------------------------------- 224 + Normally, a stripped down copy of a module's symbol table (containing only 225 + "core" symbols) is made available through module->symtab (See layout_symtab() 226 + in kernel/module.c). For livepatch modules, the symbol table copied into memory 227 + on module load must be exactly the same as the symbol table produced when the 228 + patch module was compiled. This is because the relocations in each livepatch 229 + relocation section refer to their respective symbols with their symbol indices, 230 + and the original symbol indices (and thus the symtab ordering) must be 231 + preserved in order for apply_relocate_add() to find the right symbol. 232 + 233 + For example, take this particular rela from a livepatch module: 234 + Relocation section '.klp.rela.btrfs.text.btrfs_feature_attr_show' at offset 0x2ba0 contains 4 entries: 235 + Offset Info Type Symbol's Value Symbol's Name + Addend 236 + 000000000000001f 0000005e00000002 R_X86_64_PC32 0000000000000000 .klp.sym.vmlinux.printk,0 - 4 237 + 238 + This rela refers to the symbol '.klp.sym.vmlinux.printk,0', and the symbol index is encoded 239 + in 'Info'. Here its symbol index is 0x5e, which is 94 in decimal, which refers to the 240 + symbol index 94. 241 + And in this patch module's corresponding symbol table, symbol index 94 refers to that very symbol: 242 + [ snip ] 243 + 94: 0000000000000000 0 NOTYPE GLOBAL DEFAULT OS [0xff20] .klp.sym.vmlinux.printk,0 244 + [ snip ] 245 + 246 + --------------------------- 247 + 3.3 Livepatch symbol format 248 + --------------------------- 249 + 250 + 3.3.1 Required flags 251 + -------------------- 252 + Livepatch symbols must have their section index marked as SHN_LIVEPATCH, so 253 + that the module loader can identify them and not attempt to resolve them. 254 + See include/uapi/linux/elf.h for the actual definitions. 255 + 256 + 3.3.2 Required name format 257 + -------------------------- 258 + Livepatch symbol names must conform to the following format: 259 + 260 + .klp.sym.objname.symbol_name,sympos 261 + ^ ^^ ^ ^ ^ ^ 262 + |_______||_____| |_________| | 263 + [A] [B] [C] [D] 264 + 265 + [A] The symbol name is prefixed with the string ".klp.sym." 266 + [B] The name of the object (i.e. "vmlinux" or name of module) to 267 + which the symbol belongs follows immediately after the prefix. 268 + [C] The actual name of the symbol. 269 + [D] The position of the symbol in the object (as according to kallsyms) 270 + This is used to differentiate duplicate symbols within the same 271 + object. The symbol position is expressed numerically (0, 1, 2...). 272 + The symbol position of a unique symbol is 0. 273 + 274 + 3.3.3 Example livepatch symbol names: 275 + ------------------------------------- 276 + .klp.sym.vmlinux.snprintf,0 277 + .klp.sym.vmlinux.printk,0 278 + .klp.sym.btrfs.btrfs_ktype,0 279 + 280 + 3.3.4 Example `readelf --symbols` output for a patch module: 281 + ------------------------------------------------------------ 282 + Symbol table '.symtab' contains 127 entries: 283 + Num: Value Size Type Bind Vis Ndx Name 284 + [ snip ] 285 + 73: 0000000000000000 0 NOTYPE GLOBAL DEFAULT OS [0xff20] .klp.sym.vmlinux.snprintf,0 286 + 74: 0000000000000000 0 NOTYPE GLOBAL DEFAULT OS [0xff20] .klp.sym.vmlinux.capable,0 287 + 75: 0000000000000000 0 NOTYPE GLOBAL DEFAULT OS [0xff20] .klp.sym.vmlinux.find_next_bit,0 288 + 76: 0000000000000000 0 NOTYPE GLOBAL DEFAULT OS [0xff20] .klp.sym.vmlinux.si_swapinfo,0 289 + [ snip ] ^ 290 + | 291 + [*] 292 + [*] Note that the 'Ndx' (Section index) for these symbols is SHN_LIVEPATCH (0xff20). 293 + "OS" means OS-specific. 294 + 295 + -------------------------------------- 296 + 4. Symbol table and Elf section access 297 + -------------------------------------- 298 + A livepatch module's symbol table is accessible through module->symtab. 299 + 300 + Since apply_relocate_add() requires access to a module's section headers, 301 + symbol table, and relocation section indices, Elf information is preserved for 302 + livepatch modules and is made accessible by the module loader through 303 + module->klp_info, which is a klp_modinfo struct. When a livepatch module loads, 304 + this struct is filled in by the module loader. Its fields are documented below: 305 + 306 + struct klp_modinfo { 307 + Elf_Ehdr hdr; /* Elf header */ 308 + Elf_Shdr *sechdrs; /* Section header table */ 309 + char *secstrings; /* String table for the section headers */ 310 + unsigned int symndx; /* The symbol table section index */ 311 + };