X-Git-Url: http://git.cascardo.info/?a=blobdiff_plain;f=mm%2Ffilemap.c;h=4a0f5fa79dbd552dd2ccd34480f494483646e339;hb=ebded02788b5;hp=3461d97ecb30bfe18d1ed50729e5147aa2ad7e6d;hpb=3b22371e202d2c2f48bb4192879cf611e91a97be;p=cascardo%2Flinux.git

diff --git a/mm/filemap.c b/mm/filemap.c
index 3461d97ecb30..4a0f5fa79dbd 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -195,6 +195,30 @@ void __delete_from_page_cache(struct page *page, void *shadow,
 	else
 		cleancache_invalidate_page(mapping, page);
 
+	VM_BUG_ON_PAGE(page_mapped(page), page);
+	if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
+		int mapcount;
+
+		pr_alert("BUG: Bad page cache in process %s  pfn:%05lx\n",
+			 current->comm, page_to_pfn(page));
+		dump_page(page, "still mapped when deleted");
+		dump_stack();
+		add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
+
+		mapcount = page_mapcount(page);
+		if (mapping_exiting(mapping) &&
+		    page_count(page) >= mapcount + 2) {
+			/*
+			 * All vmas have already been torn down, so it's
+			 * a good bet that actually the page is unmapped,
+			 * and we'd prefer not to leak it: if we're wrong,
+			 * some other bad page check should catch it later.
+			 */
+			page_mapcount_reset(page);
+			atomic_sub(mapcount, &page->_count);
+		}
+	}
+
 	page_cache_tree_delete(mapping, page, shadow);
 
 	page->mapping = NULL;
@@ -205,7 +229,6 @@ void __delete_from_page_cache(struct page *page, void *shadow,
 		__dec_zone_page_state(page, NR_FILE_PAGES);
 	if (PageSwapBacked(page))
 		__dec_zone_page_state(page, NR_SHMEM);
-	VM_BUG_ON_PAGE(page_mapped(page), page);
 
 	/*
 	 * At this point page must be either written or cleaned by truncate.
@@ -1645,6 +1668,15 @@ find_page:
 					index, last_index - index);
 		}
 		if (!PageUptodate(page)) {
+			/*
+			 * See comment in do_read_cache_page on why
+			 * wait_on_page_locked is used to avoid unnecessarily
+			 * serialisations and why it's safe.
+			 */
+			wait_on_page_locked_killable(page);
+			if (PageUptodate(page))
+				goto page_ok;
+
 			if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
 					!mapping->a_ops->is_partially_uptodate)
 				goto page_not_up_to_date;
@@ -2280,7 +2312,7 @@ static struct page *wait_on_page_read(struct page *page)
 	return page;
 }
 
-static struct page *__read_cache_page(struct address_space *mapping,
+static struct page *do_read_cache_page(struct address_space *mapping,
 				pgoff_t index,
 				int (*filler)(void *, struct page *),
 				void *data,
@@ -2302,53 +2334,74 @@ repeat:
 			/* Presumably ENOMEM for radix tree node */
 			return ERR_PTR(err);
 		}
+
+filler:
 		err = filler(data, page);
 		if (err < 0) {
 			page_cache_release(page);
-			page = ERR_PTR(err);
-		} else {
-			page = wait_on_page_read(page);
+			return ERR_PTR(err);
 		}
-	}
-	return page;
-}
-
-static struct page *do_read_cache_page(struct address_space *mapping,
-				pgoff_t index,
-				int (*filler)(void *, struct page *),
-				void *data,
-				gfp_t gfp)
 
-{
-	struct page *page;
-	int err;
+		page = wait_on_page_read(page);
+		if (IS_ERR(page))
+			return page;
+		goto out;
+	}
+	if (PageUptodate(page))
+		goto out;
 
-retry:
-	page = __read_cache_page(mapping, index, filler, data, gfp);
-	if (IS_ERR(page))
-		return page;
+	/*
+	 * Page is not up to date and may be locked due one of the following
+	 * case a: Page is being filled and the page lock is held
+	 * case b: Read/write error clearing the page uptodate status
+	 * case c: Truncation in progress (page locked)
+	 * case d: Reclaim in progress
+	 *
+	 * Case a, the page will be up to date when the page is unlocked.
+	 *    There is no need to serialise on the page lock here as the page
+	 *    is pinned so the lock gives no additional protection. Even if the
+	 *    the page is truncated, the data is still valid if PageUptodate as
+	 *    it's a race vs truncate race.
+	 * Case b, the page will not be up to date
+	 * Case c, the page may be truncated but in itself, the data may still
+	 *    be valid after IO completes as it's a read vs truncate race. The
+	 *    operation must restart if the page is not uptodate on unlock but
+	 *    otherwise serialising on page lock to stabilise the mapping gives
+	 *    no additional guarantees to the caller as the page lock is
+	 *    released before return.
+	 * Case d, similar to truncation. If reclaim holds the page lock, it
+	 *    will be a race with remove_mapping that determines if the mapping
+	 *    is valid on unlock but otherwise the data is valid and there is
+	 *    no need to serialise with page lock.
+	 *
+	 * As the page lock gives no additional guarantee, we optimistically
+	 * wait on the page to be unlocked and check if it's up to date and
+	 * use the page if it is. Otherwise, the page lock is required to
+	 * distinguish between the different cases. The motivation is that we
+	 * avoid spurious serialisations and wakeups when multiple processes
+	 * wait on the same page for IO to complete.
+	 */
+	wait_on_page_locked(page);
 	if (PageUptodate(page))
 		goto out;
 
+	/* Distinguish between all the cases under the safety of the lock */
 	lock_page(page);
+
+	/* Case c or d, restart the operation */
 	if (!page->mapping) {
 		unlock_page(page);
 		page_cache_release(page);
-		goto retry;
+		goto repeat;
 	}
+
+	/* Someone else locked and filled the page in a very small window */
 	if (PageUptodate(page)) {
 		unlock_page(page);
 		goto out;
 	}
-	err = filler(data, page);
-	if (err < 0) {
-		page_cache_release(page);
-		return ERR_PTR(err);
-	} else {
-		page = wait_on_page_read(page);
-		if (IS_ERR(page))
-			return page;
-	}
+	goto filler;
+
 out:
 	mark_page_accessed(page);
 	return page;