module: kallsyms: Ensure preemption in add_kallsyms() with PREEMPT_RT
PATCH v2 1/1 module: kallsyms: Ensure preemption in add_kallsyms() with PREEMPT_RT
To disable preemption in the context of add_kallsyms() is incorrect.
Before kallsyms-specific data is prepared/or set-up, we ensure that the unformed module is known to be unique i.e. does not already exist (see load_module()). Therefore, we can fix this by using the common RCU primitive as this section of code can be safely preempted.
--- kernel/module/kallsyms.c | 24 ++++++++++++------------ 1 file changed, 12 insertions(+), 12 deletions(-)
kernel/module/kallsyms.c +12 -12
@@ -174,14 +174,14 @@ void add_kallsyms(struct module *mod, const struct load_info *info) mod->kallsyms = (void __rcu *)mod->init_layout.base + info->mod_kallsyms_init_off; -preempt_disable (); +rcu_read_lock ();/* The following is safe since this pointer cannot change */ -rcu_dereference_sched (mod->kallsyms)->symtab = (void *)symsec->sh_addr; -rcu_dereference_sched (mod->kallsyms)->num_symtab = symsec->sh_size /sizeof (Elf_Sym); +rcu_dereference (mod->kallsyms)->symtab = (void *)symsec->sh_addr; +rcu_dereference (mod->kallsyms)->num_symtab = symsec->sh_size /sizeof (Elf_Sym);/* Make sure we get permanent strtab: don't use info->strtab. */ -rcu_dereference_sched (mod->kallsyms)->strtab = +rcu_dereference (mod->kallsyms)->strtab = (void *)info->sechdrs[info->index.str].sh_addr; -rcu_dereference_sched (mod->kallsyms)->typetab = mod->init_layout.base + info->init_typeoffs; +rcu_dereference (mod->kallsyms)->typetab = mod->init_layout.base + info->init_typeoffs;/* * Now populate the cut down core kallsyms for after init
@@ -190,22 +190,22 @@ void add_kallsyms(struct module *mod, const struct load_info *info) mod->core_kallsyms.symtab = dst = mod->data_layout.base + info->symoffs; mod->core_kallsyms.strtab = s = mod->data_layout.base + info->stroffs; mod->core_kallsyms.typetab = mod->data_layout.base + info->core_typeoffs; - src =rcu_dereference_sched (mod->kallsyms)->symtab; -for (ndst = i =0 ; i <rcu_dereference_sched (mod->kallsyms)->num_symtab; i++) { -rcu_dereference_sched (mod->kallsyms)->typetab[i] =elf_type (src + i, info); + src =rcu_dereference (mod->kallsyms)->symtab; +for (ndst = i =0 ; i <rcu_dereference (mod->kallsyms)->num_symtab; i++) { +rcu_dereference (mod->kallsyms)->typetab[i] =elf_type (src + i, info);if (i ==0 ||is_livepatch_module (mod) ||is_core_symbol (src + i, info->sechdrs, info->hdr->e_shnum, info->index.pcpu)) { mod->core_kallsyms.typetab[ndst] = -rcu_dereference_sched (mod->kallsyms)->typetab[i]; +rcu_dereference (mod->kallsyms)->typetab[i]; dst[ndst] = src[i]; dst[ndst++].st_name = s - mod->core_kallsyms.strtab; - s +=strscpy (s, - &rcu_dereference_sched (mod->kallsyms)->strtab[src[i].st_name], + s +=strlcpy (s, + &rcu_dereference (mod->kallsyms)->strtab[src[i].st_name],KSYM_NAME_LEN ) +1 ; } } -preempt_enable (); +rcu_read_unlock (); mod->core_kallsyms.num_symtab = ndst; }
Signature · 1 line
2.34.3
Luis Chamberlain
maintainerHey Aaron, thanks again!
Quoted 1 line · Aaron Tomlin
To disable preemption in the context of
add_kallsyms()is incorrect.
Why, what broke? Did this used to work? Was the commit in question a regression then? Clarifying all this will help a lot.
The commit log is better but yet doesn't make it easy for me to tell if I should send this to Linus as a fix in the rc series.
Luis
Quoted 69 lines
Before kallsyms-specific data is prepared/or set-up, we ensure that the unformed module is known to be unique i.e. does not already exist (see
load_module()). Therefore, we can fix this by using the common RCU primitive as this section of code can be safely preempted.Reported-by Steven Rostedt <[email protected]>Fixes 08126db5ff73 ("module: kallsyms: Fix suspicious rcu usage")Signed-off-by Aaron Tomlin <[email protected]>--- kernel/module/kallsyms.c | 24 ++++++++++++------------ 1 file changed, 12 insertions(+), 12 deletions(-)kernel/module/kallsyms.c +12
-12@@ -174,14 +174,14 @@ void add_kallsyms(struct module *mod, const struct load_info *info) mod->kallsyms = (void __rcu *)mod->init_layout.base + info->mod_kallsyms_init_off; -preempt_disable (); +rcu_read_lock ();/* The following is safe since this pointer cannot change */ -rcu_dereference_sched (mod->kallsyms)->symtab = (void *)symsec->sh_addr; -rcu_dereference_sched (mod->kallsyms)->num_symtab = symsec->sh_size /sizeof (Elf_Sym); +rcu_dereference (mod->kallsyms)->symtab = (void *)symsec->sh_addr; +rcu_dereference (mod->kallsyms)->num_symtab = symsec->sh_size /sizeof (Elf_Sym);/* Make sure we get permanent strtab: don't use info->strtab. */ -rcu_dereference_sched (mod->kallsyms)->strtab = +rcu_dereference (mod->kallsyms)->strtab = (void *)info->sechdrs[info->index.str].sh_addr; -rcu_dereference_sched (mod->kallsyms)->typetab = mod->init_layout.base + info->init_typeoffs; +rcu_dereference (mod->kallsyms)->typetab = mod->init_layout.base + info->init_typeoffs;/* * Now populate the cut down core kallsyms for after init @@ -190,22 +190,22 @@ void add_kallsyms(struct module *mod, const struct load_info *info) mod->core_kallsyms.symtab = dst = mod->data_layout.base + info->symoffs; mod->core_kallsyms.strtab = s = mod->data_layout.base + info->stroffs; mod->core_kallsyms.typetab = mod->data_layout.base + info->core_typeoffs; - src =rcu_dereference_sched (mod->kallsyms)->symtab; -for (ndst = i =0 ; i <rcu_dereference_sched (mod->kallsyms)->num_symtab; i++) { -rcu_dereference_sched (mod->kallsyms)->typetab[i] =elf_type (src + i, info); + src =rcu_dereference (mod->kallsyms)->symtab; +for (ndst = i =0 ; i <rcu_dereference (mod->kallsyms)->num_symtab; i++) { +rcu_dereference (mod->kallsyms)->typetab[i] =elf_type (src + i, info);if (i ==0 ||is_livepatch_module (mod) ||is_core_symbol (src + i, info->sechdrs, info->hdr->e_shnum, info->index.pcpu)) { mod->core_kallsyms.typetab[ndst] = -rcu_dereference_sched (mod->kallsyms)->typetab[i]; +rcu_dereference (mod->kallsyms)->typetab[i]; dst[ndst] = src[i]; dst[ndst++].st_name = s - mod->core_kallsyms.strtab; - s +=strscpy (s, - &rcu_dereference_sched (mod->kallsyms)->strtab[src[i].st_name], + s +=strlcpy (s, + &rcu_dereference (mod->kallsyms)->strtab[src[i].st_name],KSYM_NAME_LEN ) +1 ; } } -preempt_enable (); +rcu_read_unlock (); mod->core_kallsyms.num_symtab = ndst; }Signature · 1 line
2.34.3
Quoted 1 line · Luis Chamberlain
Hey Aaron, thanks again!
Hi Luis,
No problem :)
Quoted 5 lines · Aaron Tomlin
Quoted 1 line
To disable preemption in the context of
add_kallsyms()is incorrect.Why, what broke? Did this used to work? Was the commit in question a regression then? Clarifying all this will help a lot.
Sorry for the confusion! If I understand correctly, nothing broke intrinsically.
Rather with commit 08126db5ff73 ("module: kallsyms: Fix suspicious rcu usage") under PREEMPT_RT=y, by disabling preemption, I introduced an unbounded latency since the loop is not fixed which is generally frowned upon. So, I would say this was a regression since earlier preemption was not disabled and we would dereference RCU-protected pointers explicitly i.e. without using the more appropriate rcu_dereference() family of primitives. That being said, these pointers cannot change in this context as explained previously.
Would the above be suitable - just to confirm before I send another iteration?
Kind regards,
Signature · 1 line
Aaron Tomlin
Luis Chamberlain
maintainerQuoted 13 lines · Aaron Tomlin
Quoted 1 line · Luis Chamberlain
Hey Aaron, thanks again!
Hi Luis,
No problem :)
Quoted 5 lines · Aaron Tomlin
Quoted 1 line
To disable preemption in the context of
add_kallsyms()is incorrect.Why, what broke? Did this used to work? Was the commit in question a regression then? Clarifying all this will help a lot.
Quoted 7 lines
Sorry for the confusion! If I understand correctly, nothing broke intrinsically.
Rather with commit
08126db5ff73("module: kallsyms: Fix suspicious rcu usage") under PREEMPT_RT=y, by disabling preemption, I introduced an unbounded latency since the loop is not fixed which is generally frowned upon.
This is incredibly important information which should be added to the commit log, specialy as PREEMPT_RT=y becomes a first class citizen.
Quoted 8 lines
So, I would say this was a regression since earlier preemption was not disabled and we would dereference RCU-protected pointers explicitly i.e. without using the more appropriate
rcu_dereference()family of primitives. That being said, these pointers cannot change in this context as explained previously.Would the above be suitable - just to confirm before I send another iteration?
Yes, I would send this to Linus for the rc series. Please adjust the commit log with all this information.
BTW I think there is just one more fix pending from you right?
Luis
Aaron Tomlin
reviewerQuoted 2 lines · Luis Chamberlain
This is incredibly important information which should be added to the commit log, specialy as PREEMPT_RT=y becomes a first class citizen.
Understood.
Quoted 7 lines
Quoted 7 lines
So, I would say this was a regression since earlier preemption was not disabled and we would dereference RCU-protected pointers explicitly i.e. without using the more appropriate
rcu_dereference()family of primitives. That being said, these pointers cannot change in this context as explained previously.
Quoted 5 lines
Quoted 2 lines
Would the above be suitable - just to confirm before I send another iteration?
Yes, I would send this to Linus for the rc series. Please adjust the commit log with all this information.
Will do.
Quoted 1 line
BTW I think there is just one more fix pending from you right?
Yes - I will send/or prepare it as a partial revert:
's/strscpy/strlcpy/' with a brief explanation.
Kind regards,
Signature · 1 line
Aaron Tomlin
Luis Chamberlain
maintainerQuoted 19 lines · Aaron Tomlin
Quoted 2 lines · Luis Chamberlain
This is incredibly important information which should be added to the commit log, specialy as PREEMPT_RT=y becomes a first class citizen.
Understood.
Quoted 7 lines
Quoted 7 lines
So, I would say this was a regression since earlier preemption was not disabled and we would dereference RCU-protected pointers explicitly i.e. without using the more appropriate
rcu_dereference()family of primitives. That being said, these pointers cannot change in this context as explained previously.Quoted 5 lines
Quoted 2 lines
Would the above be suitable - just to confirm before I send another iteration?
Yes, I would send this to Linus for the rc series. Please adjust the commit log with all this information.
Quoted 6 lines
Will do.
Quoted 1 line
BTW I think there is just one more fix pending from you right?
Yes - I will send/or prepare it as a partial revert:
's/strscpy/strlcpy/' with a brief explanation.
If that is just the kallsyms fix Adrian Hunter already sent a fix:
https://lkml.kernel.org/r/[email protected]
Luis