Messages in this thread | | | From | David Howells <> | Subject | Re: [RFC PATCH 54/61] afs: Wait on PG_fscache before modifying/releasing a page | Date | Fri, 08 May 2020 15:39:48 +0100 |
| |
David Howells <dhowells@redhat.com> wrote:
> ITER_MAPPING relies on the mapping to maintain the pointers to the pages so > that it can find them rather than being like ITER_BVEC where there's a > separate list. > > Truncate removes the pages from the mapping - at which point ITER_MAPPING can > no longer find them.
It looks like ITER_MAPPING is fine with truncate, provided the invalidation waits for the iterator to complete first:
int truncate_inode_page(struct address_space *mapping, struct page *page) { VM_BUG_ON_PAGE(PageTail(page), page);
if (page->mapping != mapping) return -EIO;
truncate_cleanup_page(mapping, page); delete_from_page_cache(page); return 0; }
In which case, ->invalidatepage() needs to wait for PG_fscache.
Similarly, it looks like ->releasepage() is fine, provided it waits for PG_fscache also.
If I have to use ITER_BVEC, what's the advisability of using vmalloc() to allocate the bio_vec array for a transient op? Such an array can reference up to 1MiB on a 64-bit machine with 4KiB non-compound pages if it only allocates up to a single page. I'm wondering what the teardown cost is, though, if all the corresponding PTEs have to be erased from all CPUs.
David
| |