Messages in this thread | | | Date | Sun, 12 Feb 2012 13:54:35 +0100 | Subject | Fwd: [PATCH 2/2] [RFC] fadvise: Add _VOLATILE,_ISVOLATILE, and _NONVOLATILE flags | From | Dmitry Adamushko <> |
| |
[ resent to lkml in 'plain-text' format ]
On 10 February 2012 01:16, John Stultz <john.stultz@linaro.org> wrote:
[ ... ]
> --- /dev/null > +++ b/mm/volatile.c > @@ -0,0 +1,314 @@ > +/* mm/volatile.c > + * > [ ... ] > > + > +#define range_on_lru(range) (!(range)->purged) > + > + > +static inline void volatile_range_shrink(struct volatile_range *range, > + pgoff_t start_index, pgoff_t end_index) > +{ > + size_t pre = range_size(range); > + > + range->range_node.start = start_index; > + range->range_node.end = end_index; > +
I guess, here we get a whole range of races with volatile_shrink(), which may see inconsistent (in-the-middle-of-update) ranges (e.g. .start and .end).
> > + if (range_on_lru(range)) {
here volatile_shrink() runs and sets range->purge to 1, then calls __lru_del() => lru_count gets updated.
> > + mutex_lock(&volatile_lru_mutex); > + lru_count -= pre - range_size(range); > + mutex_unlock(&volatile_lru_mutex);
and then lru_count gets updated once more - for the same 'range' object.
> > + } > +}
> > [ ... ]
> > > +static int volatile_shrink(struct shrinker *ignored, struct shrink_control *sc) > +{ > + struct volatile_range *range, *next; > + unsigned long nr_to_scan = sc->nr_to_scan; > + const gfp_t gfp_mask = sc->gfp_mask; > + > + /* We might recurse into filesystem code, so bail out if necessary */ > + if (nr_to_scan && !(gfp_mask & __GFP_FS)) > + return -1; > + if (!nr_to_scan) > + return lru_count;
So it's u64 -> int here, which is possibly 32 bits and signed. Can't it lead to inconsistent results on 32bit platforms?
> > + > + mutex_lock(&volatile_lru_mutex); > + list_for_each_entry_safe(range, next, &volatile_lru_list, lru) { > + struct inode *inode = range->mapping->host; > + loff_t start, end; > + > + > + start = range->range_node.start * PAGE_SIZE; > + end = (range->range_node.end + 1) * PAGE_SIZE - 1;
PAGE_CACHE_SHIFT was used in fadvise() to calculate .start and .end indexes, and here we use PAGE_SIZE to get back to 'normal' addresses. Isn't it inconsistent at the very least?
> > + > + /* > + * XXX - calling vmtruncate_range from a shrinker causes > + * lockdep warnings. Revisit this! > + */ > + vmtruncate_range(inode, start, end); > + range->purged = 1; > + __lru_del(range); > + > + nr_to_scan -= range_size(range);
hmm, unsigned long -= u64
> > + if (nr_to_scan <= 0)
nr_to_scan is "unsigned long" :-))
[ ... ]
> +arch_initcall(volatile_init); > -- > 1.7.3.2.146.gca209 >
--
-- Dmitry -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |