Re: binfmt_elf padzero problems
Andrew Morton writes: > I guess if the bss has zero length then we can skip the zeroing of the end > of the page at the end of bss, as long as we're dead sure that we didn't > accidentally instantiate a single page on behalf of that zero-length bss. There is another thing I noticed about the bss code, which is that it doesn't give the bss the permissions from the PT_LOAD segment, rather it just uses VM_DATA_DEFAULT_FLAGS. That doesn't matter at the moment but may matter in future for ppc32. Paul. - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: binfmt_elf padzero problems
Nir Tzachar <[EMAIL PROTECTED]> wrote: > > hello. > > i am seeing a problem(?) with the patch described at: > http://marc.theaimsgroup.com/?l=linux-kernel=109865760703851=2 > i'm using vanilla 2.6.11 (not .1/.2/.3/.4 ...) > > the short version: > padzero does not alway do the right thing (more correctly, it's caller, > load_elf_binary). > > the longer version: > > padzero calls clear_user. clear_user first checks if the address passed > is writable. if it is not, an error is returned. > the problem manifest itself when the area being cleared is not > writable... this should not normally happen in the context of > load_elf_binary, however it _can_ happen with the following assembly > code (intel syntax): > > section .text > global _start > _start: > mov eax,0x1 > mov ebx,0x0 > int 0x80 > hlt > > assembled with nasm -f elf, produces a binary with a bss segment of zero > size, aligned to 1, and one program header. > now, the when calling padzero, elf_bss holds an address which belongs > to .text (since no (fake)program header for .bss wad created), i.e; not > writable > when padzero is called, it tries to clean the rest of the .text section, > which clearly results with an error. > > thus, my (very) small binary always segfaults under 2.6.11+ > > on the other hand, i can be dead wrong.. if so, id like to know why... > Tricky. I guess if the bss has zero length then we can skip the zeroing of the end of the page at the end of bss, as long as we're dead sure that we didn't accidentally instantiate a single page on behalf of that zero-length bss. Something like this, perhaps? --- 25/fs/binfmt_elf.c~aThu Mar 17 14:47:35 2005 +++ 25-akpm/fs/binfmt_elf.c Thu Mar 17 14:48:44 2005 @@ -907,15 +907,17 @@ static int load_elf_binary(struct linux_ * mapping in the interpreter, to make sure it doesn't wind * up getting placed where the bss needs to go. */ - retval = set_brk(elf_bss, elf_brk); - if (retval) { - send_sig(SIGKILL, current, 0); - goto out_free_dentry; - } - if (padzero(elf_bss)) { - send_sig(SIGSEGV, current, 0); - retval = -EFAULT; /* Nobody gets to see this, but.. */ - goto out_free_dentry; + if (likely(elf_bss != elf_brk)) { /* Is there any bss at all? */ + retval = set_brk(elf_bss, elf_brk); + if (retval) { + send_sig(SIGKILL, current, 0); + goto out_free_dentry; + } + if (padzero(elf_bss)) { + send_sig(SIGSEGV, current, 0); + retval = -EFAULT; /* Nobody gets to see this, but.. */ + goto out_free_dentry; + } } if (elf_interpreter) { _ - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
binfmt_elf padzero problems
hello. i am seeing a problem(?) with the patch described at: http://marc.theaimsgroup.com/?l=linux-kernel=109865760703851=2 i'm using vanilla 2.6.11 (not .1/.2/.3/.4 ...) the short version: padzero does not alway do the right thing (more correctly, it's caller, load_elf_binary). the longer version: padzero calls clear_user. clear_user first checks if the address passed is writable. if it is not, an error is returned. the problem manifest itself when the area being cleared is not writable... this should not normally happen in the context of load_elf_binary, however it _can_ happen with the following assembly code (intel syntax): section .text global _start _start: mov eax,0x1 mov ebx,0x0 int 0x80 hlt assembled with nasm -f elf, produces a binary with a bss segment of zero size, aligned to 1, and one program header. now, the when calling padzero, elf_bss holds an address which belongs to .text (since no (fake)program header for .bss wad created), i.e; not writable when padzero is called, it tries to clean the rest of the .text section, which clearly results with an error. thus, my (very) small binary always segfaults under 2.6.11+ on the other hand, i can be dead wrong.. if so, id like to know why... p.s. please cc me, im not subscribed, -- = Nir Tzachar. signature.asc Description: This is a digitally signed message part
binfmt_elf padzero problems
hello. i am seeing a problem(?) with the patch described at: http://marc.theaimsgroup.com/?l=linux-kernelm=109865760703851w=2 i'm using vanilla 2.6.11 (not .1/.2/.3/.4 ...) the short version: padzero does not alway do the right thing (more correctly, it's caller, load_elf_binary). the longer version: padzero calls clear_user. clear_user first checks if the address passed is writable. if it is not, an error is returned. the problem manifest itself when the area being cleared is not writable... this should not normally happen in the context of load_elf_binary, however it _can_ happen with the following assembly code (intel syntax): section .text global _start _start: mov eax,0x1 mov ebx,0x0 int 0x80 hlt assembled with nasm -f elf, produces a binary with a bss segment of zero size, aligned to 1, and one program header. now, the when calling padzero, elf_bss holds an address which belongs to .text (since no (fake)program header for .bss wad created), i.e; not writable when padzero is called, it tries to clean the rest of the .text section, which clearly results with an error. thus, my (very) small binary always segfaults under 2.6.11+ on the other hand, i can be dead wrong.. if so, id like to know why... p.s. please cc me, im not subscribed, -- = Nir Tzachar. signature.asc Description: This is a digitally signed message part
Re: binfmt_elf padzero problems
Nir Tzachar [EMAIL PROTECTED] wrote: hello. i am seeing a problem(?) with the patch described at: http://marc.theaimsgroup.com/?l=linux-kernelm=109865760703851w=2 i'm using vanilla 2.6.11 (not .1/.2/.3/.4 ...) the short version: padzero does not alway do the right thing (more correctly, it's caller, load_elf_binary). the longer version: padzero calls clear_user. clear_user first checks if the address passed is writable. if it is not, an error is returned. the problem manifest itself when the area being cleared is not writable... this should not normally happen in the context of load_elf_binary, however it _can_ happen with the following assembly code (intel syntax): section .text global _start _start: mov eax,0x1 mov ebx,0x0 int 0x80 hlt assembled with nasm -f elf, produces a binary with a bss segment of zero size, aligned to 1, and one program header. now, the when calling padzero, elf_bss holds an address which belongs to .text (since no (fake)program header for .bss wad created), i.e; not writable when padzero is called, it tries to clean the rest of the .text section, which clearly results with an error. thus, my (very) small binary always segfaults under 2.6.11+ on the other hand, i can be dead wrong.. if so, id like to know why... Tricky. I guess if the bss has zero length then we can skip the zeroing of the end of the page at the end of bss, as long as we're dead sure that we didn't accidentally instantiate a single page on behalf of that zero-length bss. Something like this, perhaps? --- 25/fs/binfmt_elf.c~aThu Mar 17 14:47:35 2005 +++ 25-akpm/fs/binfmt_elf.c Thu Mar 17 14:48:44 2005 @@ -907,15 +907,17 @@ static int load_elf_binary(struct linux_ * mapping in the interpreter, to make sure it doesn't wind * up getting placed where the bss needs to go. */ - retval = set_brk(elf_bss, elf_brk); - if (retval) { - send_sig(SIGKILL, current, 0); - goto out_free_dentry; - } - if (padzero(elf_bss)) { - send_sig(SIGSEGV, current, 0); - retval = -EFAULT; /* Nobody gets to see this, but.. */ - goto out_free_dentry; + if (likely(elf_bss != elf_brk)) { /* Is there any bss at all? */ + retval = set_brk(elf_bss, elf_brk); + if (retval) { + send_sig(SIGKILL, current, 0); + goto out_free_dentry; + } + if (padzero(elf_bss)) { + send_sig(SIGSEGV, current, 0); + retval = -EFAULT; /* Nobody gets to see this, but.. */ + goto out_free_dentry; + } } if (elf_interpreter) { _ - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: binfmt_elf padzero problems
Andrew Morton writes: I guess if the bss has zero length then we can skip the zeroing of the end of the page at the end of bss, as long as we're dead sure that we didn't accidentally instantiate a single page on behalf of that zero-length bss. There is another thing I noticed about the bss code, which is that it doesn't give the bss the permissions from the PT_LOAD segment, rather it just uses VM_DATA_DEFAULT_FLAGS. That doesn't matter at the moment but may matter in future for ppc32. Paul. - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/