]> git.hungrycats.org Git - linux/commitdiff
mm + fs: prepare for non-page entries in page cache radix trees
authorJohannes Weiner <hannes@cmpxchg.org>
Thu, 28 Aug 2014 18:34:54 +0000 (19:34 +0100)
committerJiri Slaby <jslaby@suse.cz>
Fri, 26 Sep 2014 09:51:58 +0000 (11:51 +0200)
commit 0cd6144aadd2afd19d1aca880153530c52957604 upstream.

shmem mappings already contain exceptional entries where swap slot
information is remembered.

To be able to store eviction information for regular page cache, prepare
every site dealing with the radix trees directly to handle entries other
than pages.

The common lookup functions will filter out non-page entries and return
NULL for page cache holes, just as before.  But provide a raw version of
the API which returns non-page entries as well, and switch shmem over to
use it.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Minchan Kim <minchan@kernel.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Bob Liu <bob.liu@oracle.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jan Kara <jack@suse.cz>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Luigi Semenzato <semenzato@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Metin Doslu <metin@citusdata.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Ozgun Erdogan <ozgun@citusdata.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Roman Gushchin <klamm@yandex-team.ru>
Cc: Ryan Mallon <rmallon@gmail.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
fs/btrfs/compression.c
include/linux/mm.h
include/linux/pagemap.h
include/linux/pagevec.h
include/linux/shmem_fs.h
mm/filemap.c
mm/mincore.c
mm/readahead.c
mm/shmem.c
mm/swap.c
mm/truncate.c

index 6e9ff8fac75a1ba56d5199f491f6d402ad30cbc1..6357298932bffdd7d544309e119a0f35dc0d545e 100644 (file)
@@ -474,7 +474,7 @@ static noinline int add_ra_bio_pages(struct inode *inode,
                rcu_read_lock();
                page = radix_tree_lookup(&mapping->page_tree, pg_index);
                rcu_read_unlock();
-               if (page) {
+               if (page && !radix_tree_exceptional_entry(page)) {
                        misses++;
                        if (misses > 4)
                                break;
index 073734339583e2f67f63a6327ea3a8aee15009e4..96fc7ddd307f0f38f0f7f4ff79b7ed426214b7b9 100644 (file)
@@ -919,6 +919,14 @@ extern void show_free_areas(unsigned int flags);
 extern bool skip_free_areas_node(unsigned int flags, int nid);
 
 int shmem_zero_setup(struct vm_area_struct *);
+#ifdef CONFIG_SHMEM
+bool shmem_mapping(struct address_space *mapping);
+#else
+static inline bool shmem_mapping(struct address_space *mapping)
+{
+       return false;
+}
+#endif
 
 extern int can_do_mlock(void);
 extern int user_shm_lock(size_t, struct user_struct *);
index c73130c607c4eb040a731fefcb4c2327b214ed54..76aaeaad37be3fb01119c23eed3147e04b5ee8f6 100644 (file)
@@ -248,12 +248,15 @@ pgoff_t page_cache_next_hole(struct address_space *mapping,
 pgoff_t page_cache_prev_hole(struct address_space *mapping,
                             pgoff_t index, unsigned long max_scan);
 
-extern struct page * find_get_page(struct address_space *mapping,
-                               pgoff_t index);
-extern struct page * find_lock_page(struct address_space *mapping,
-                               pgoff_t index);
-extern struct page * find_or_create_page(struct address_space *mapping,
-                               pgoff_t index, gfp_t gfp_mask);
+struct page *find_get_entry(struct address_space *mapping, pgoff_t offset);
+struct page *find_get_page(struct address_space *mapping, pgoff_t offset);
+struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset);
+struct page *find_lock_page(struct address_space *mapping, pgoff_t offset);
+struct page *find_or_create_page(struct address_space *mapping, pgoff_t index,
+                                gfp_t gfp_mask);
+unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
+                         unsigned int nr_entries, struct page **entries,
+                         pgoff_t *indices);
 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
                        unsigned int nr_pages, struct page **pages);
 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t start,
index e4dbfab377296765e6190d7d6f1ead9bf53245ec..b45d391b454053c5af635db7992c1e12532f0a8e 100644 (file)
@@ -22,6 +22,11 @@ struct pagevec {
 
 void __pagevec_release(struct pagevec *pvec);
 void __pagevec_lru_add(struct pagevec *pvec);
+unsigned pagevec_lookup_entries(struct pagevec *pvec,
+                               struct address_space *mapping,
+                               pgoff_t start, unsigned nr_entries,
+                               pgoff_t *indices);
+void pagevec_remove_exceptionals(struct pagevec *pvec);
 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
                pgoff_t start, unsigned nr_pages);
 unsigned pagevec_lookup_tag(struct pagevec *pvec,
index 30aa0dc60d75786287226bf3cbedffd9f1b266a3..deb49609cd36b45b260daff79de66892dd10b666 100644 (file)
@@ -49,6 +49,7 @@ extern struct file *shmem_file_setup(const char *name,
                                        loff_t size, unsigned long flags);
 extern int shmem_zero_setup(struct vm_area_struct *);
 extern int shmem_lock(struct file *file, int lock, struct user_struct *user);
+extern bool shmem_mapping(struct address_space *mapping);
 extern void shmem_unlock_mapping(struct address_space *mapping);
 extern struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
                                        pgoff_t index, gfp_t gfp_mask);
index a47d0a72d3a2c323450cf64cd0eb2c5b7f65f712..5d57a78bf0211120b97d811c9ca128f730f2d44d 100644 (file)
@@ -448,6 +448,29 @@ int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
 }
 EXPORT_SYMBOL_GPL(replace_page_cache_page);
 
+static int page_cache_tree_insert(struct address_space *mapping,
+                                 struct page *page)
+{
+       void **slot;
+       int error;
+
+       slot = radix_tree_lookup_slot(&mapping->page_tree, page->index);
+       if (slot) {
+               void *p;
+
+               p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
+               if (!radix_tree_exceptional_entry(p))
+                       return -EEXIST;
+               radix_tree_replace_slot(slot, page);
+               mapping->nrpages++;
+               return 0;
+       }
+       error = radix_tree_insert(&mapping->page_tree, page->index, page);
+       if (!error)
+               mapping->nrpages++;
+       return error;
+}
+
 /**
  * add_to_page_cache_locked - add a locked page to the pagecache
  * @page:      page to add
@@ -482,11 +505,10 @@ int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
        page->index = offset;
 
        spin_lock_irq(&mapping->tree_lock);
-       error = radix_tree_insert(&mapping->page_tree, offset, page);
+       error = page_cache_tree_insert(mapping, page);
        radix_tree_preload_end();
        if (unlikely(error))
                goto err_insert;
-       mapping->nrpages++;
        __inc_zone_page_state(page, NR_FILE_PAGES);
        spin_unlock_irq(&mapping->tree_lock);
        trace_mm_filemap_add_to_page_cache(page);
@@ -714,7 +736,10 @@ pgoff_t page_cache_next_hole(struct address_space *mapping,
        unsigned long i;
 
        for (i = 0; i < max_scan; i++) {
-               if (!radix_tree_lookup(&mapping->page_tree, index))
+               struct page *page;
+
+               page = radix_tree_lookup(&mapping->page_tree, index);
+               if (!page || radix_tree_exceptional_entry(page))
                        break;
                index++;
                if (index == 0)
@@ -752,7 +777,10 @@ pgoff_t page_cache_prev_hole(struct address_space *mapping,
        unsigned long i;
 
        for (i = 0; i < max_scan; i++) {
-               if (!radix_tree_lookup(&mapping->page_tree, index))
+               struct page *page;
+
+               page = radix_tree_lookup(&mapping->page_tree, index);
+               if (!page || radix_tree_exceptional_entry(page))
                        break;
                index--;
                if (index == ULONG_MAX)
@@ -764,14 +792,19 @@ pgoff_t page_cache_prev_hole(struct address_space *mapping,
 EXPORT_SYMBOL(page_cache_prev_hole);
 
 /**
- * find_get_page - find and get a page reference
+ * find_get_entry - find and get a page cache entry
  * @mapping: the address_space to search
- * @offset: the page index
+ * @offset: the page cache index
+ *
+ * Looks up the page cache slot at @mapping & @offset.  If there is a
+ * page cache page, it is returned with an increased refcount.
  *
- * Is there a pagecache struct page at the given (mapping, offset) tuple?
- * If yes, increment its refcount and return it; if no, return NULL.
+ * If the slot holds a shadow entry of a previously evicted page, it
+ * is returned.
+ *
+ * Otherwise, %NULL is returned.
  */
-struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
+struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
 {
        void **pagep;
        struct page *page;
@@ -812,24 +845,50 @@ out:
 
        return page;
 }
-EXPORT_SYMBOL(find_get_page);
+EXPORT_SYMBOL(find_get_entry);
 
 /**
- * find_lock_page - locate, pin and lock a pagecache page
+ * find_get_page - find and get a page reference
  * @mapping: the address_space to search
  * @offset: the page index
  *
- * Locates the desired pagecache page, locks it, increments its reference
- * count and returns its address.
+ * Looks up the page cache slot at @mapping & @offset.  If there is a
+ * page cache page, it is returned with an increased refcount.
  *
- * Returns zero if the page was not present. find_lock_page() may sleep.
+ * Otherwise, %NULL is returned.
  */
-struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
+struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
+{
+       struct page *page = find_get_entry(mapping, offset);
+
+       if (radix_tree_exceptional_entry(page))
+               page = NULL;
+       return page;
+}
+EXPORT_SYMBOL(find_get_page);
+
+/**
+ * find_lock_entry - locate, pin and lock a page cache entry
+ * @mapping: the address_space to search
+ * @offset: the page cache index
+ *
+ * Looks up the page cache slot at @mapping & @offset.  If there is a
+ * page cache page, it is returned locked and with an increased
+ * refcount.
+ *
+ * If the slot holds a shadow entry of a previously evicted page, it
+ * is returned.
+ *
+ * Otherwise, %NULL is returned.
+ *
+ * find_lock_entry() may sleep.
+ */
+struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
 {
        struct page *page;
 
 repeat:
-       page = find_get_page(mapping, offset);
+       page = find_get_entry(mapping, offset);
        if (page && !radix_tree_exception(page)) {
                lock_page(page);
                /* Has the page been truncated? */
@@ -842,6 +901,29 @@ repeat:
        }
        return page;
 }
+EXPORT_SYMBOL(find_lock_entry);
+
+/**
+ * find_lock_page - locate, pin and lock a pagecache page
+ * @mapping: the address_space to search
+ * @offset: the page index
+ *
+ * Looks up the page cache slot at @mapping & @offset.  If there is a
+ * page cache page, it is returned locked and with an increased
+ * refcount.
+ *
+ * Otherwise, %NULL is returned.
+ *
+ * find_lock_page() may sleep.
+ */
+struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
+{
+       struct page *page = find_lock_entry(mapping, offset);
+
+       if (radix_tree_exceptional_entry(page))
+               page = NULL;
+       return page;
+}
 EXPORT_SYMBOL(find_lock_page);
 
 /**
@@ -850,16 +932,18 @@ EXPORT_SYMBOL(find_lock_page);
  * @index: the page's index into the mapping
  * @gfp_mask: page allocation mode
  *
- * Locates a page in the pagecache.  If the page is not present, a new page
- * is allocated using @gfp_mask and is added to the pagecache and to the VM's
- * LRU list.  The returned page is locked and has its reference count
- * incremented.
+ * Looks up the page cache slot at @mapping & @offset.  If there is a
+ * page cache page, it is returned locked and with an increased
+ * refcount.
+ *
+ * If the page is not present, a new page is allocated using @gfp_mask
+ * and added to the page cache and the VM's LRU list.  The page is
+ * returned locked and with an increased refcount.
  *
- * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
- * allocation!
+ * On memory exhaustion, %NULL is returned.
  *
- * find_or_create_page() returns the desired page's address, or zero on
- * memory exhaustion.
+ * find_or_create_page() may sleep, even if @gfp_flags specifies an
+ * atomic allocation!
  */
 struct page *find_or_create_page(struct address_space *mapping,
                pgoff_t index, gfp_t gfp_mask)
@@ -891,6 +975,76 @@ repeat:
 }
 EXPORT_SYMBOL(find_or_create_page);
 
+/**
+ * find_get_entries - gang pagecache lookup
+ * @mapping:   The address_space to search
+ * @start:     The starting page cache index
+ * @nr_entries:        The maximum number of entries
+ * @entries:   Where the resulting entries are placed
+ * @indices:   The cache indices corresponding to the entries in @entries
+ *
+ * find_get_entries() will search for and return a group of up to
+ * @nr_entries entries in the mapping.  The entries are placed at
+ * @entries.  find_get_entries() takes a reference against any actual
+ * pages it returns.
+ *
+ * The search returns a group of mapping-contiguous page cache entries
+ * with ascending indexes.  There may be holes in the indices due to
+ * not-present pages.
+ *
+ * Any shadow entries of evicted pages are included in the returned
+ * array.
+ *
+ * find_get_entries() returns the number of pages and shadow entries
+ * which were found.
+ */
+unsigned find_get_entries(struct address_space *mapping,
+                         pgoff_t start, unsigned int nr_entries,
+                         struct page **entries, pgoff_t *indices)
+{
+       void **slot;
+       unsigned int ret = 0;
+       struct radix_tree_iter iter;
+
+       if (!nr_entries)
+               return 0;
+
+       rcu_read_lock();
+restart:
+       radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
+               struct page *page;
+repeat:
+               page = radix_tree_deref_slot(slot);
+               if (unlikely(!page))
+                       continue;
+               if (radix_tree_exception(page)) {
+                       if (radix_tree_deref_retry(page))
+                               goto restart;
+                       /*
+                        * Otherwise, we must be storing a swap entry
+                        * here as an exceptional entry: so return it
+                        * without attempting to raise page count.
+                        */
+                       goto export;
+               }
+               if (!page_cache_get_speculative(page))
+                       goto repeat;
+
+               /* Has the page moved? */
+               if (unlikely(page != *slot)) {
+                       page_cache_release(page);
+                       goto repeat;
+               }
+export:
+               indices[ret] = iter.index;
+               entries[ret] = page;
+               if (++ret == nr_entries)
+                       break;
+       }
+       rcu_read_unlock();
+       return ret;
+}
+
 /**
  * find_get_pages - gang pagecache lookup
  * @mapping:   The address_space to search
index da2be56a7b8fdd3c8391fd79e758f394dd63157a..06cb81005c771a5d40d907b70300691984e5fae4 100644 (file)
@@ -70,13 +70,21 @@ static unsigned char mincore_page(struct address_space *mapping, pgoff_t pgoff)
         * any other file mapping (ie. marked !present and faulted in with
         * tmpfs's .fault). So swapped out tmpfs mappings are tested here.
         */
-       page = find_get_page(mapping, pgoff);
 #ifdef CONFIG_SWAP
-       /* shmem/tmpfs may return swap: account for swapcache page too. */
-       if (radix_tree_exceptional_entry(page)) {
-               swp_entry_t swap = radix_to_swp_entry(page);
-               page = find_get_page(swap_address_space(swap), swap.val);
-       }
+       if (shmem_mapping(mapping)) {
+               page = find_get_entry(mapping, pgoff);
+               /*
+                * shmem/tmpfs may return swap: account for swapcache
+                * page too.
+                */
+               if (radix_tree_exceptional_entry(page)) {
+                       swp_entry_t swp = radix_to_swp_entry(page);
+                       page = find_get_page(swap_address_space(swp), swp.val);
+               }
+       } else
+               page = find_get_page(mapping, pgoff);
+#else
+       page = find_get_page(mapping, pgoff);
 #endif
        if (page) {
                present = PageUptodate(page);
index 6db5d5603c1ced51d5e681e981e227527e2b0566..4ab0c93bc2b3f82e8722bd74f7dded142a68bdaa 100644 (file)
@@ -179,7 +179,7 @@ __do_page_cache_readahead(struct address_space *mapping, struct file *filp,
                rcu_read_lock();
                page = radix_tree_lookup(&mapping->page_tree, page_offset);
                rcu_read_unlock();
-               if (page)
+               if (page && !radix_tree_exceptional_entry(page))
                        continue;
 
                page = page_cache_alloc_readahead(mapping);
index 1460b201c0b36aca13b6ba14b7666762673b50e8..15e5527934ad52f3107b8491fc972b71a2063e3a 100644 (file)
@@ -329,56 +329,6 @@ static void shmem_delete_from_page_cache(struct page *page, void *radswap)
        BUG_ON(error);
 }
 
-/*
- * Like find_get_pages, but collecting swap entries as well as pages.
- */
-static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
-                                       pgoff_t start, unsigned int nr_pages,
-                                       struct page **pages, pgoff_t *indices)
-{
-       void **slot;
-       unsigned int ret = 0;
-       struct radix_tree_iter iter;
-
-       if (!nr_pages)
-               return 0;
-
-       rcu_read_lock();
-restart:
-       radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
-               struct page *page;
-repeat:
-               page = radix_tree_deref_slot(slot);
-               if (unlikely(!page))
-                       continue;
-               if (radix_tree_exception(page)) {
-                       if (radix_tree_deref_retry(page))
-                               goto restart;
-                       /*
-                        * Otherwise, we must be storing a swap entry
-                        * here as an exceptional entry: so return it
-                        * without attempting to raise page count.
-                        */
-                       goto export;
-               }
-               if (!page_cache_get_speculative(page))
-                       goto repeat;
-
-               /* Has the page moved? */
-               if (unlikely(page != *slot)) {
-                       page_cache_release(page);
-                       goto repeat;
-               }
-export:
-               indices[ret] = iter.index;
-               pages[ret] = page;
-               if (++ret == nr_pages)
-                       break;
-       }
-       rcu_read_unlock();
-       return ret;
-}
-
 /*
  * Remove swap entry from radix tree, free the swap and its page cache.
  */
@@ -396,21 +346,6 @@ static int shmem_free_swap(struct address_space *mapping,
        return 0;
 }
 
-/*
- * Pagevec may contain swap entries, so shuffle up pages before releasing.
- */
-static void shmem_deswap_pagevec(struct pagevec *pvec)
-{
-       int i, j;
-
-       for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
-               struct page *page = pvec->pages[i];
-               if (!radix_tree_exceptional_entry(page))
-                       pvec->pages[j++] = page;
-       }
-       pvec->nr = j;
-}
-
 /*
  * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
  */
@@ -429,12 +364,12 @@ void shmem_unlock_mapping(struct address_space *mapping)
                 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
                 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
                 */
-               pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
-                                       PAGEVEC_SIZE, pvec.pages, indices);
+               pvec.nr = find_get_entries(mapping, index,
+                                          PAGEVEC_SIZE, pvec.pages, indices);
                if (!pvec.nr)
                        break;
                index = indices[pvec.nr - 1] + 1;
-               shmem_deswap_pagevec(&pvec);
+               pagevec_remove_exceptionals(&pvec);
                check_move_unevictable_pages(pvec.pages, pvec.nr);
                pagevec_release(&pvec);
                cond_resched();
@@ -466,9 +401,9 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
        pagevec_init(&pvec, 0);
        index = start;
        while (index < end) {
-               pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
-                               min(end - index, (pgoff_t)PAGEVEC_SIZE),
-                                                       pvec.pages, indices);
+               pvec.nr = find_get_entries(mapping, index,
+                       min(end - index, (pgoff_t)PAGEVEC_SIZE),
+                       pvec.pages, indices);
                if (!pvec.nr)
                        break;
                mem_cgroup_uncharge_start();
@@ -497,7 +432,7 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
                        }
                        unlock_page(page);
                }
-               shmem_deswap_pagevec(&pvec);
+               pagevec_remove_exceptionals(&pvec);
                pagevec_release(&pvec);
                mem_cgroup_uncharge_end();
                cond_resched();
@@ -535,9 +470,10 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
        index = start;
        while (index < end) {
                cond_resched();
-               pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
+
+               pvec.nr = find_get_entries(mapping, index,
                                min(end - index, (pgoff_t)PAGEVEC_SIZE),
-                                                       pvec.pages, indices);
+                               pvec.pages, indices);
                if (!pvec.nr) {
                        /* If all gone or hole-punch or unfalloc, we're done */
                        if (index == start || end != -1)
@@ -580,7 +516,7 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
                        }
                        unlock_page(page);
                }
-               shmem_deswap_pagevec(&pvec);
+               pagevec_remove_exceptionals(&pvec);
                pagevec_release(&pvec);
                mem_cgroup_uncharge_end();
                index++;
@@ -1089,7 +1025,7 @@ static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
                return -EFBIG;
 repeat:
        swap.val = 0;
-       page = find_lock_page(mapping, index);
+       page = find_lock_entry(mapping, index);
        if (radix_tree_exceptional_entry(page)) {
                swap = radix_to_swp_entry(page);
                page = NULL;
@@ -1484,6 +1420,11 @@ static struct inode *shmem_get_inode(struct super_block *sb, const struct inode
        return inode;
 }
 
+bool shmem_mapping(struct address_space *mapping)
+{
+       return mapping->backing_dev_info == &shmem_backing_dev_info;
+}
+
 #ifdef CONFIG_TMPFS
 static const struct inode_operations shmem_symlink_inode_operations;
 static const struct inode_operations shmem_short_symlink_operations;
@@ -1796,7 +1737,7 @@ static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
        pagevec_init(&pvec, 0);
        pvec.nr = 1;            /* start small: we may be there already */
        while (!done) {
-               pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
+               pvec.nr = find_get_entries(mapping, index,
                                        pvec.nr, pvec.pages, indices);
                if (!pvec.nr) {
                        if (whence == SEEK_DATA)
@@ -1823,7 +1764,7 @@ static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
                                break;
                        }
                }
-               shmem_deswap_pagevec(&pvec);
+               pagevec_remove_exceptionals(&pvec);
                pagevec_release(&pvec);
                pvec.nr = PAGEVEC_SIZE;
                cond_resched();
index aa4da5d9401d0937c007a4e2ccf95c2604847304..2d9ee765102a8756429a7646651e073355f0f484 100644 (file)
--- a/mm/swap.c
+++ b/mm/swap.c
@@ -915,6 +915,57 @@ void __pagevec_lru_add(struct pagevec *pvec)
 }
 EXPORT_SYMBOL(__pagevec_lru_add);
 
+/**
+ * pagevec_lookup_entries - gang pagecache lookup
+ * @pvec:      Where the resulting entries are placed
+ * @mapping:   The address_space to search
+ * @start:     The starting entry index
+ * @nr_entries:        The maximum number of entries
+ * @indices:   The cache indices corresponding to the entries in @pvec
+ *
+ * pagevec_lookup_entries() will search for and return a group of up
+ * to @nr_entries pages and shadow entries in the mapping.  All
+ * entries are placed in @pvec.  pagevec_lookup_entries() takes a
+ * reference against actual pages in @pvec.
+ *
+ * The search returns a group of mapping-contiguous entries with
+ * ascending indexes.  There may be holes in the indices due to
+ * not-present entries.
+ *
+ * pagevec_lookup_entries() returns the number of entries which were
+ * found.
+ */
+unsigned pagevec_lookup_entries(struct pagevec *pvec,
+                               struct address_space *mapping,
+                               pgoff_t start, unsigned nr_pages,
+                               pgoff_t *indices)
+{
+       pvec->nr = find_get_entries(mapping, start, nr_pages,
+                                   pvec->pages, indices);
+       return pagevec_count(pvec);
+}
+
+/**
+ * pagevec_remove_exceptionals - pagevec exceptionals pruning
+ * @pvec:      The pagevec to prune
+ *
+ * pagevec_lookup_entries() fills both pages and exceptional radix
+ * tree entries into the pagevec.  This function prunes all
+ * exceptionals from @pvec without leaving holes, so that it can be
+ * passed on to page-only pagevec operations.
+ */
+void pagevec_remove_exceptionals(struct pagevec *pvec)
+{
+       int i, j;
+
+       for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
+               struct page *page = pvec->pages[i];
+               if (!radix_tree_exceptional_entry(page))
+                       pvec->pages[j++] = page;
+       }
+       pvec->nr = j;
+}
+
 /**
  * pagevec_lookup - gang pagecache lookup
  * @pvec:      Where the resulting pages are placed
index 353b683afd6ef538ab3318f5c40ad6502789227e..2e84fe59190b7e8073fe96c25fc2e9212c5b9da7 100644 (file)
 #include <linux/cleancache.h>
 #include "internal.h"
 
+static void clear_exceptional_entry(struct address_space *mapping,
+                                   pgoff_t index, void *entry)
+{
+       /* Handled by shmem itself */
+       if (shmem_mapping(mapping))
+               return;
+
+       spin_lock_irq(&mapping->tree_lock);
+       /*
+        * Regular page slots are stabilized by the page lock even
+        * without the tree itself locked.  These unlocked entries
+        * need verification under the tree lock.
+        */
+       radix_tree_delete_item(&mapping->page_tree, index, entry);
+       spin_unlock_irq(&mapping->tree_lock);
+}
 
 /**
  * do_invalidatepage - invalidate part or all of a page
@@ -208,6 +224,7 @@ void truncate_inode_pages_range(struct address_space *mapping,
        unsigned int    partial_start;  /* inclusive */
        unsigned int    partial_end;    /* exclusive */
        struct pagevec  pvec;
+       pgoff_t         indices[PAGEVEC_SIZE];
        pgoff_t         index;
        int             i;
 
@@ -238,17 +255,23 @@ void truncate_inode_pages_range(struct address_space *mapping,
 
        pagevec_init(&pvec, 0);
        index = start;
-       while (index < end && pagevec_lookup(&pvec, mapping, index,
-                       min(end - index, (pgoff_t)PAGEVEC_SIZE))) {
+       while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
+                       min(end - index, (pgoff_t)PAGEVEC_SIZE),
+                       indices)) {
                mem_cgroup_uncharge_start();
                for (i = 0; i < pagevec_count(&pvec); i++) {
                        struct page *page = pvec.pages[i];
 
                        /* We rely upon deletion not changing page->index */
-                       index = page->index;
+                       index = indices[i];
                        if (index >= end)
                                break;
 
+                       if (radix_tree_exceptional_entry(page)) {
+                               clear_exceptional_entry(mapping, index, page);
+                               continue;
+                       }
+
                        if (!trylock_page(page))
                                continue;
                        WARN_ON(page->index != index);
@@ -259,6 +282,7 @@ void truncate_inode_pages_range(struct address_space *mapping,
                        truncate_inode_page(mapping, page);
                        unlock_page(page);
                }
+               pagevec_remove_exceptionals(&pvec);
                pagevec_release(&pvec);
                mem_cgroup_uncharge_end();
                cond_resched();
@@ -307,14 +331,16 @@ void truncate_inode_pages_range(struct address_space *mapping,
        index = start;
        for ( ; ; ) {
                cond_resched();
-               if (!pagevec_lookup(&pvec, mapping, index,
-                       min(end - index, (pgoff_t)PAGEVEC_SIZE))) {
+               if (!pagevec_lookup_entries(&pvec, mapping, index,
+                       min(end - index, (pgoff_t)PAGEVEC_SIZE),
+                       indices)) {
                        if (index == start)
                                break;
                        index = start;
                        continue;
                }
-               if (index == start && pvec.pages[0]->index >= end) {
+               if (index == start && indices[0] >= end) {
+                       pagevec_remove_exceptionals(&pvec);
                        pagevec_release(&pvec);
                        break;
                }
@@ -323,16 +349,22 @@ void truncate_inode_pages_range(struct address_space *mapping,
                        struct page *page = pvec.pages[i];
 
                        /* We rely upon deletion not changing page->index */
-                       index = page->index;
+                       index = indices[i];
                        if (index >= end)
                                break;
 
+                       if (radix_tree_exceptional_entry(page)) {
+                               clear_exceptional_entry(mapping, index, page);
+                               continue;
+                       }
+
                        lock_page(page);
                        WARN_ON(page->index != index);
                        wait_on_page_writeback(page);
                        truncate_inode_page(mapping, page);
                        unlock_page(page);
                }
+               pagevec_remove_exceptionals(&pvec);
                pagevec_release(&pvec);
                mem_cgroup_uncharge_end();
                index++;
@@ -375,6 +407,7 @@ EXPORT_SYMBOL(truncate_inode_pages);
 unsigned long invalidate_mapping_pages(struct address_space *mapping,
                pgoff_t start, pgoff_t end)
 {
+       pgoff_t indices[PAGEVEC_SIZE];
        struct pagevec pvec;
        pgoff_t index = start;
        unsigned long ret;
@@ -390,17 +423,23 @@ unsigned long invalidate_mapping_pages(struct address_space *mapping,
         */
 
        pagevec_init(&pvec, 0);
-       while (index <= end && pagevec_lookup(&pvec, mapping, index,
-                       min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
+       while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
+                       min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
+                       indices)) {
                mem_cgroup_uncharge_start();
                for (i = 0; i < pagevec_count(&pvec); i++) {
                        struct page *page = pvec.pages[i];
 
                        /* We rely upon deletion not changing page->index */
-                       index = page->index;
+                       index = indices[i];
                        if (index > end)
                                break;
 
+                       if (radix_tree_exceptional_entry(page)) {
+                               clear_exceptional_entry(mapping, index, page);
+                               continue;
+                       }
+
                        if (!trylock_page(page))
                                continue;
                        WARN_ON(page->index != index);
@@ -414,6 +453,7 @@ unsigned long invalidate_mapping_pages(struct address_space *mapping,
                                deactivate_page(page);
                        count += ret;
                }
+               pagevec_remove_exceptionals(&pvec);
                pagevec_release(&pvec);
                mem_cgroup_uncharge_end();
                cond_resched();
@@ -481,6 +521,7 @@ static int do_launder_page(struct address_space *mapping, struct page *page)
 int invalidate_inode_pages2_range(struct address_space *mapping,
                                  pgoff_t start, pgoff_t end)
 {
+       pgoff_t indices[PAGEVEC_SIZE];
        struct pagevec pvec;
        pgoff_t index;
        int i;
@@ -491,17 +532,23 @@ int invalidate_inode_pages2_range(struct address_space *mapping,
        cleancache_invalidate_inode(mapping);
        pagevec_init(&pvec, 0);
        index = start;
-       while (index <= end && pagevec_lookup(&pvec, mapping, index,
-                       min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
+       while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
+                       min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
+                       indices)) {
                mem_cgroup_uncharge_start();
                for (i = 0; i < pagevec_count(&pvec); i++) {
                        struct page *page = pvec.pages[i];
 
                        /* We rely upon deletion not changing page->index */
-                       index = page->index;
+                       index = indices[i];
                        if (index > end)
                                break;
 
+                       if (radix_tree_exceptional_entry(page)) {
+                               clear_exceptional_entry(mapping, index, page);
+                               continue;
+                       }
+
                        lock_page(page);
                        WARN_ON(page->index != index);
                        if (page->mapping != mapping) {
@@ -539,6 +586,7 @@ int invalidate_inode_pages2_range(struct address_space *mapping,
                                ret = ret2;
                        unlock_page(page);
                }
+               pagevec_remove_exceptionals(&pvec);
                pagevec_release(&pvec);
                mem_cgroup_uncharge_end();
                cond_resched();