Skip to contentSkip to conversation

x86/build32: Discard all orphaned sections

PATCH In review Andrew Cooper 3 participants 6 messages May 12, 2020 – May 13, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86/build32: Discard all orphaned sections

Linkers may put orphaned sections ahead of .text, which breaks the calling requirements. A concrete example is Ubuntu's GCC-9 default of enabling -fcf-protection which causes us to try and execute .note.gnu.properties during Xen's boot.

Put .got.plt in its own section as it specifically needs preserving from the linkers point of view, and discard everything else. This will hopefully be more robust to other unexpected toolchain properties.

Fixes boot from an Ubuntu build of Xen.

---
CC: Jan Beulich <[email protected]>
CC: Wei Liu <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Jason Andryuk <[email protected]>
CC: Stefan Bader <[email protected]>

---
 xen/arch/x86/boot/build32.lds | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)
xen/arch/x86/boot/build32.lds +10 -2
index da35aee910..97454b40ff 100644
@@ -31,7 +31,7 @@ SECTIONS
         *(.bss.*)
   }
 
-  /DISCARD/ : {
+  .got.plt : {
         /*
          * PIC/PIE executable contains .got.plt section even if it is not linked
          * with dynamic libraries. In such case it is just placeholder for
@@ -47,6 +47,14 @@ SECTIONS
          *
          * Please check build32.mk for more details.
          */
-        /* *(.got.plt) */
+        *(.got.plt)
+  }
+
+  /DISCARD/ : {
+        /*
+         * Discard everything else, to prevent linkers from putting
+         * orphaned sections ahead of .text, which needs to be first.
+         */
+        *(*)
   }
 }
Signature · 1 line
2.11.0
Tested-by on v1 1/1
#1 lore raw
Quoted 6 lines · Andrew Cooper

Linkers may put orphaned sections ahead of .text, which breaks the calling requirements. A concrete example is Ubuntu's GCC-9 default of enabling -fcf-protection which causes us to try and execute .note.gnu.properties during Xen's boot.

Quoted 8 lines

Put .got.plt in its own section as it specifically needs preserving from the linkers point of view, and discard everything else. This will hopefully be more robust to other unexpected toolchain properties.

Fixes boot from an Ubuntu build of Xen.

Thanks

#2 lore raw
In reply to Andrew Cooper
Quoted 6 lines · Andrew Cooper
xen/arch/x86/boot/build32.lds +1 -1
@@ -47,6 +47,14 @@ SECTIONS
          *
          * Please check build32.mk for more details.
          */
-        /* *(.got.plt) */
+        *(.got.plt)

xen/arch/x86/boot/build32.lds SECTIONS · line 60

Quoted 10 lines
xen/arch/x86/boot/build32.lds +8 -0
+  }
+
+  /DISCARD/ : {
+        /*
+         * Discard everything else, to prevent linkers from putting
+         * orphaned sections ahead of .text, which needs to be first.
+         */
+        *(*)
   }
 }

To be honest I'm not sure if this isn't going too far. Much depends on what would happen if a new real section appeared that needs retaining. Granted the linker may then once again put it at the beginning of the image, and we'll be screwed again, just like we'll be screwed if a section gets discarded by mistake. But it would be really nice if we had a way to flag the need to play with the linker script. Hence perhaps on new enough tool chains we indeed may want to make use of --orphan-handling= ? And then discard just .note and .note.* here?

Jan

#3 lore raw
Quoted 14 lines · Jan Beulich

[CAUTION - EXTERNAL EMAIL] DO NOT reply, click links, or open attachments unless you have verified the sender and know the content is safe.

Quoted 11 lines · Andrew Cooper
xen/arch/x86/boot/build32.lds +6 -1
@@ -47,6 +47,14 @@ SECTIONS
          *
          * Please check build32.mk for more details.
          */
-        /* *(.got.plt) */
+        *(.got.plt)
+  }
+
+  /DISCARD/ : {
+        /*
+         * Discard everything else, to prevent linkers from putting
Quoted 8 lines

xen/arch/x86/boot/build32.lds SECTIONS · line 60

Quoted 5 lines
xen/arch/x86/boot/build32.lds +3 -0
+         * orphaned sections ahead of .text, which needs to be first.
+         */
+        *(*)
   }
 }

To be honest I'm not sure if this isn't going too far. Much depends on what would happen if a new real section appeared that needs retaining.

Anything which is important enough will result in a linker error.

Quoted 4 lines

Granted the linker may then once again
put it at the beginning of the image, and we'll be screwed again, just like we'll be screwed if a section gets discarded by mistake.

This way is more likely to result in a build failure than an inability to boot the resulting build of Xen.

Quoted 5 lines

But it would be really nice if we had a way to flag the need to play with the linker script. Hence perhaps on new enough tool chains we indeed may want to make use of --orphan-handling= ? And then discard just .note and .note.* here?

The only valid option would be =error, but experimenting with that yields

ld: error: unplaced orphan section .comment' from cmdline.o'
ld: error: unplaced orphan section .note.GNU-stack' from cmdline.o'
ld: error: unplaced orphan section .note.gnu.property' from cmdline.o'
ld: error: unplaced orphan section .rel.got' from cmdline.o'
ld: error: unplaced orphan section .got' from cmdline.o'
ld: error: unplaced orphan section .got.plt' from cmdline.o'
ld: error: unplaced orphan section .iplt' from cmdline.o'
ld: error: unplaced orphan section .rel.iplt' from cmdline.o'
ld: error: unplaced orphan section .igot.plt' from cmdline.o'

which I think is going to get us massively bogged down in toolchain specifics.  I'm not entirely convinced this would be a good move.

~Andrew

Acked-by
#4 lore raw
Quoted 52 lines · Andrew Cooper
Quoted 14 lines · Jan Beulich

[CAUTION - EXTERNAL EMAIL] DO NOT reply, click links, or open attachments unless you have verified the sender and know the content is safe.

Quoted 11 lines · Andrew Cooper
xen/arch/x86/boot/build32.lds +6 -1
@@ -47,6 +47,14 @@ SECTIONS
          *
          * Please check build32.mk for more details.
          */
-        /* *(.got.plt) */
+        *(.got.plt)
+  }
+
+  /DISCARD/ : {
+        /*
+         * Discard everything else, to prevent linkers from putting
Quoted 8 lines

xen/arch/x86/boot/build32.lds SECTIONS · line 60

Quoted 5 lines
xen/arch/x86/boot/build32.lds +3 -0
+         * orphaned sections ahead of .text, which needs to be first.
+         */
+        *(*)
   }
 }

To be honest I'm not sure if this isn't going too far. Much depends on what would happen if a new real section appeared that needs retaining.

Anything which is important enough will result in a linker error.

Quoted 4 lines

Granted the linker may then once again
put it at the beginning of the image, and we'll be screwed again, just like we'll be screwed if a section gets discarded by mistake.

This way is more likely to result in a build failure than an inability to boot the resulting build of Xen.

Quoted 5 lines

But it would be really nice if we had a way to flag the need to play with the linker script. Hence perhaps on new enough tool chains we indeed may want to make use of --orphan-handling= ? And then discard just .note and .note.* here?

The only valid option would be =error, but experimenting with that yields

ld: error: unplaced orphan section .comment' from cmdline.o'
ld: error: unplaced orphan section .note.GNU-stack' from cmdline.o'
ld: error: unplaced orphan section .note.gnu.property' from cmdline.o'
ld: error: unplaced orphan section .rel.got' from cmdline.o'
ld: error: unplaced orphan section .got' from cmdline.o'
ld: error: unplaced orphan section .got.plt' from cmdline.o'
ld: error: unplaced orphan section .iplt' from cmdline.o'
ld: error: unplaced orphan section .rel.iplt' from cmdline.o'
ld: error: unplaced orphan section .igot.plt' from cmdline.o'

Quoted 2 lines

which I think is going to get us massively bogged down in toolchain specifics.  I'm not entirely convinced this would be a good move.

That's ugly indeed; especially the .rel.* sections are worrying to appear there. Hence patch as is

Jan

#5 lore raw
Quoted 17 lines · Jan Beulich
Quoted 17 lines
Quoted 5 lines

But it would be really nice if we had a way to flag the need to play with the linker script. Hence perhaps on new enough tool chains we indeed may want to make use of --orphan-handling= ? And then discard just .note and .note.* here?

The only valid option would be =error, but experimenting with that yields

ld: error: unplaced orphan section .comment' from cmdline.o'
ld: error: unplaced orphan section .note.GNU-stack' from cmdline.o'
ld: error: unplaced orphan section .note.gnu.property' from cmdline.o'
ld: error: unplaced orphan section .rel.got' from cmdline.o'
ld: error: unplaced orphan section .got' from cmdline.o'
ld: error: unplaced orphan section .got.plt' from cmdline.o'
ld: error: unplaced orphan section .iplt' from cmdline.o'
ld: error: unplaced orphan section .rel.iplt' from cmdline.o'
ld: error: unplaced orphan section .igot.plt' from cmdline.o'

Quoted 4 lines
Quoted 2 lines

which I think is going to get us massively bogged down in toolchain specifics.  I'm not entirely convinced this would be a good move.

That's ugly indeed; especially the .rel.* sections are worrying to appear there.

What is even more curious, most of them don't exist in cmdine.o

Section Headers:
  [Nr] Name              Type            Addr     Off    Size   ES Flg Lk Inf Al
  [ 0]                   NULL            00000000 000000 000000 00 0   0  0
  [ 1] .group            GROUP           00000000 000034 000008 04 14  38  4
  [ 2] .group            GROUP           00000000 00003c 000008 04 14  40  4
  [ 3] .text             PROGBITS        00000000 000044 00094a 00  AX 0   0  1
  [ 4] .rel.text         REL             00000000 000e88 0001e8 08   I 14   3  4
  [ 5] .data             PROGBITS        00000000 00098e 000000 00  WA 0   0  1
  [ 6] .bss              NOBITS          00000000 00098e 000000 00  WA 0   0  1
  [ 7] .rodata           PROGBITS        00000000 000990 0000f3 00   A 0   0  4
  [ 8] .rel.rodata       REL             00000000 001070 000120 08   I 14   7  4
  [ 9] .text.__x86.get_pc_thunk.ax PROGBITS        00000000 000a83 000004 00 AXG  0   0  1
  [10] .text.__x86.get_pc_thunk.bx PROGBITS        00000000 000a87 000004 00 AXG  0   0  1
  [11] .comment          PROGBITS        00000000 000a8b 00002d 01  MS 0   0  1
  [12] .note.GNU-stack   PROGBITS        00000000 000ab8 000000 00 0   0  1
  [13] .note.gnu.property NOTE            00000000 000ab8 00001c 00   A 0   0  4
  [14] .symtab           SYMTAB          00000000 000ad4 000290 10 15  36  4
  [15] .strtab           STRTAB          00000000 000d64 000124 00 0   0  1
  [16] .shstrtab         STRTAB          00000000 001190 0000a7 00 0   0  1

I suspect they are inserted by default as part of processing the relocations, and end up empty.

With =warn rather than =error, we instead get

ld: warning: orphan section .comment' from cmdline.o' being placed in section `.comment'
ld: warning: orphan section .note.GNU-stack' from cmdline.o' being placed in section `.note.GNU-stack'
ld: warning: orphan section .note.gnu.property' from cmdline.o' being placed in section `.note.gnu.property'
ld: warning: orphan section .rel.got' from cmdline.o' being placed in section `.rel.dyn'
ld: warning: orphan section .got' from cmdline.o' being placed in section `.got'
ld: warning: orphan section .got.plt' from cmdline.o' being placed in section `.got.plt'
ld: warning: orphan section .iplt' from cmdline.o' being placed in section `.iplt'
ld: warning: orphan section .rel.iplt' from cmdline.o' being placed in section `.rel.dyn'
ld: warning: orphan section .igot.plt' from cmdline.o' being placed in section `.igot.plt'

and

Section Headers:
  [Nr] Name              Type            Addr     Off    Size   ES Flg Lk Inf Al
  [ 0]                   NULL            00000000 000000 000000 00 0   0  0
  [ 1] .note.gnu.property NOTE            00000000 0000b4 00001c 00   A 0   0  4
  [ 2] .text             PROGBITS        0000001c 0000d0 000a47 00 WAX 0   0  4
  [ 3] .got.plt          PROGBITS        00000a64 000b18 00000c 04  WA 0   0  4
  [ 4] .comment          PROGBITS        00000000 000b24 00002c 01  MS 0   0  1
  [ 5] .symtab           SYMTAB          00000000 000b50 000230 10 6  31  4
  [ 6] .strtab           STRTAB          00000000 000d80 000124 00 0   0  1
  [ 7] .shstrtab         STRTAB          00000000 000ea4 000046 00 0   0  1

in cmdline.lnk, so the .rel.* sections have been dropped overall.  I think the =error logic is simply at an unhelpful position during processing.

Quoted 1 line

Hence patch as is Acked-by: Jan Beulich <[email protected]>

Thanks,

As I say, I have a plan to replace all of this completely when a bit more of kbuild is in place.

~Andrew