alloc_tag: fix use-after-free in /proc/allocinfo after module unload
Subsystems (2)
PATCH 1/1 alloc_tag: fix use-after-free in /proc/allocinfo after module unload
allocinfo_start() only reinitializes the codetag iterator at position 0. For subsequent reads (position > 0), it reuses cached iterator state from the previous batch. allocinfo_stop() drops mod_lock between read batches, which allows module unload to complete and free the module memory that the cached iterator still references:
CPU0 (read) CPU1 (rmmod)
---- ----
allocinfo_start(pos=0)
down_read(mod_lock)
allocinfo_show()
...
allocinfo_stop()
up_read(mod_lock)
codetag_unload_module()
kfree(cmod)
release_module_tags()
...
free_mod_mem()
allocinfo_start(pos=N)
down_read(mod_lock)
// reuses cached iter, skips re-init
allocinfo_show()
ct->filename <-- UAFAfter free_mod_mem() frees the module's .rodata, allocinfo_show() dereferences ct->filename, ct->function which point there.
Fix by always reinitializing the iterator in allocinfo_start().
--- lib/alloc_tag.c | 13 ++++++++----- 1 file changed, 8 insertions(+), 5 deletions(-)
lib/alloc_tag.c +8 -5
@@ -51,16 +51,19 @@ struct allocinfo_private {static void *allocinfo_start (struct seq_file *m,loff_t *pos) {struct allocinfo_private *priv; +struct codetag *ct;loff_t node = *pos; priv = (struct allocinfo_private *)m->private;codetag_lock_module_list (alloc_tag_cttype, true); -if (node ==0 ) { +if (node ==0 ) priv->print_header = true; - priv->iter =codetag_get_ct_iter (alloc_tag_cttype); -codetag_next_ct (&priv->iter); - } -return priv->iter.ct ? priv :NULL ; + + priv->iter =codetag_get_ct_iter (alloc_tag_cttype); +while ((ct =codetag_next_ct (&priv->iter)) !=NULL && node) + node--; + +return ct ? priv :NULL ; }static void *allocinfo_next (struct seq_file *m,void *arg,loff_t *pos)
Signature · 1 line
2.25.1
Quoted 26 lines · Hao Ge
allocinfo_start()only reinitializes the codetag iterator at position 0. For subsequent reads (position > 0), it reuses cached iterator state from the previous batch.allocinfo_stop()dropsmod_lockbetween read batches, which allows module unload to complete and free the module memory that the cached iterator still references:CPU0 (read) CPU1 (rmmod)
---- ----
allocinfo_start(pos=0)down_read(mod_lock) allocinfo_show() ... allocinfo_stop() up_read(mod_lock) codetag_unload_module() kfree(cmod) release_module_tags() ... free_mod_mem() allocinfo_start(pos=N) down_read(mod_lock) // reuses cached iter, skips re-init allocinfo_show() ct->filename <-- UAF
Quoted 4 lines
After
free_mod_mem()frees the module's .rodata,allocinfo_show()dereferencesct->filename,ct->functionwhich point there.Fix by always reinitializing the iterator in
allocinfo_start().
Thanks for the fix!
Yeah, unfortunately with the way seq_read_iter() is implemented, the op->next() fetches the next element and then seq_has_overflowed() might cause op->stop() without printing that element. So, in the op->start() that follows we can't just call codetag_next_ct(), which would handle the module unloading case (see
https://elixir.bootlin.com/linux/v7.0.1/source/lib/codetag.c#L93).
Quoted 21 lines
Fixes 9f44df50fee4 ("alloc_tag: keep codetag iterator active between read()")Signed-off-by Hao Ge <[email protected]>--- lib/alloc_tag.c | 13 ++++++++----- 1 file changed, 8 insertions(+), 5 deletions(-)lib/alloc_tag.c +1
-1@@ -51,16 +51,19 @@ struct allocinfo_private {static void *allocinfo_start (struct seq_file *m,loff_t *pos) {struct allocinfo_private *priv; +struct codetag *ct;loff_t node = *pos; priv = (struct allocinfo_private *)m->private;codetag_lock_module_list (alloc_tag_cttype, true); -if (node ==0 ) {
lib/alloc_tag.c struct allocinfo_private { · line 64
Quoted 10 lines
lib/alloc_tag.c +5
-4+if (node ==0 ) priv->print_header = true; - priv->iter =codetag_get_ct_iter (alloc_tag_cttype); -codetag_next_ct (&priv->iter); - } -return priv->iter.ct ? priv :NULL ; + + priv->iter =codetag_get_ct_iter (alloc_tag_cttype); +while ((ct =codetag_next_ct (&priv->iter)) !=NULL && node) + node--;
I don't quite like this fix because if a module that we already passed gets unloaded too, then skipping "node" count of elements might skip some extra valid lines that we did not report yet. That's why codetag_next_ct() directly goes to the next module using idr_get_next_ul(). I would rather store the last codetag reported using allocinfo_show() (last_reported) and use codetag_next_ct(last_reported) in the allocinfo_start() to go to the next one (or possibly skip to the next module). Does that make sense?
Quoted 8 lines
lib/alloc_tag.c +2
-1+ +return ct ? priv :NULL ; }static void *allocinfo_next (struct seq_file *m,void *arg,loff_t *pos) --2.25.1
Quoted 68 lines · Suren Baghdasaryan
Quoted 25 lines · Hao Ge
allocinfo_start()only reinitializes the codetag iterator at position 0. For subsequent reads (position > 0), it reuses cached iterator state from the previous batch.allocinfo_stop()dropsmod_lockbetween read batches, which allows module unload to complete and free the module memory that the cached iterator still references:CPU0 (read) CPU1 (rmmod) ---- ---- allocinfo_start(pos=0) down_read(mod_lock) allocinfo_show() ... allocinfo_stop() up_read(mod_lock) codetag_unload_module() kfree(cmod) release_module_tags() ... free_mod_mem() allocinfo_start(pos=N) down_read(mod_lock) // reuses cached iter, skips re-init allocinfo_show() ct->filename <-- UAFQuoted 4 lines
After
free_mod_mem()frees the module's .rodata,allocinfo_show()dereferencesct->filename,ct->functionwhich point there.Fix by always reinitializing the iterator in
allocinfo_start().Thanks for the fix!
Yeah, unfortunately with the wayseq_read_iter()is implemented, theop->next()fetches the next element and thenseq_has_overflowed()might causeop->stop()without printing that element. So, in theop->start()that follows we can't just callcodetag_next_ct(), which would handle the module unloading case (see
https://elixir.bootlin.com/linux/v7.0.1/source/lib/codetag.c#L93).Quoted 20 lines
Fixes 9f44df50fee4 ("alloc_tag: keep codetag iterator active between read()")Signed-off-by Hao Ge <[email protected]>--- lib/alloc_tag.c | 13 ++++++++----- 1 file changed, 8 insertions(+), 5 deletions(-)lib/alloc_tag.c +1
-1@@ -51,16 +51,19 @@ struct allocinfo_private {static void *allocinfo_start (struct seq_file *m,loff_t *pos) {struct allocinfo_private *priv; +struct codetag *ct;loff_t node = *pos; priv = (struct allocinfo_private *)m->private;codetag_lock_module_list (alloc_tag_cttype, true); -if (node ==0 ) {lib/alloc_tag.c struct allocinfo_private { · line 64
Quoted 10 lines
lib/alloc_tag.c +5
-4+if (node ==0 ) priv->print_header = true; - priv->iter =codetag_get_ct_iter (alloc_tag_cttype); -codetag_next_ct (&priv->iter); - } -return priv->iter.ct ? priv :NULL ; + + priv->iter =codetag_get_ct_iter (alloc_tag_cttype); +while ((ct =codetag_next_ct (&priv->iter)) !=NULL && node) + node--;
Quoted 8 lines
I don't quite like this fix because if a module that we already passed gets unloaded too, then skipping "node" count of elements might skip some extra valid lines that we did not report yet. That's why
codetag_next_ct()directly goes to the next module usingidr_get_next_ul(). I would rather store the last codetag reported usingallocinfo_show()(last_reported) and usecodetag_next_ct(last_reported)in theallocinfo_start()to go to the next one (or possibly skip to the next module). Does that make sense?
Hi Suren
Thanks for the suggestion, I completely agree.
My patch has exactly the problem you described - reinitializing from scratch on every
start() also has an O(n) performance problem: each start(pos=N) needs to skip N
elements from the beginning, regardless of whether any module was unloaded.
One small difference from your suggestion: I save the iterator state in allocinfo_next() rather than allocinfo_show().
Since seq_read_iter() may roll back show()'s output on overflow
https://elixir.bootlin.com/linux/v7.1-rc6/source/fs/seq_file.c#L276
saving in show() could capture an element whose output was discarded, causing it to be skipped on the next start().
In the Fill loop, next() is not called again after a show() overflow since seq_read_iter() breaks out immediately.
Will send a v2 shortly.
Thanks
Best Regards
Hao
Quoted 8 lines
Quoted 8 lines
lib/alloc_tag.c +2
-1+ +return ct ? priv :NULL ; }static void *allocinfo_next (struct seq_file *m,void *arg,loff_t *pos) --2.25.1