]> pilppa.org Git - linux-2.6-omap-h63xx.git/blobdiff - mm/filemap.c
lguest: Remove 'network: no dma buffer!' warning
[linux-2.6-omap-h63xx.git] / mm / filemap.c
index 2d3ec1ffc66e1621b3db99a4105fb92115804e82..5de7633e1dbe5c4da6b749f9fbd5bfd9841a382b 100644 (file)
 /*
  * Remove a page from the page cache and free it. Caller has to make
  * sure the page is locked and that nobody else uses it - or that usage
- * is safe.  The caller must hold a write_lock on the mapping's tree_lock.
+ * is safe.  The caller must hold the mapping's tree_lock.
  */
 void __remove_from_page_cache(struct page *page)
 {
@@ -141,9 +141,9 @@ void remove_from_page_cache(struct page *page)
 
        BUG_ON(!PageLocked(page));
 
-       write_lock_irq(&mapping->tree_lock);
+       spin_lock_irq(&mapping->tree_lock);
        __remove_from_page_cache(page);
-       write_unlock_irq(&mapping->tree_lock);
+       spin_unlock_irq(&mapping->tree_lock);
 }
 
 static int sync_page(void *word)
@@ -442,48 +442,52 @@ int filemap_write_and_wait_range(struct address_space *mapping,
 }
 
 /**
- * add_to_page_cache - add newly allocated pagecache pages
+ * add_to_page_cache_locked - add a locked page to the pagecache
  * @page:      page to add
  * @mapping:   the page's address_space
  * @offset:    page index
  * @gfp_mask:  page allocation mode
  *
- * This function is used to add newly allocated pagecache pages;
- * the page is new, so we can just run SetPageLocked() against it.
- * The other page state flags were set by rmqueue().
- *
+ * This function is used to add a page to the pagecache. It must be locked.
  * This function does not add the page to the LRU.  The caller must do that.
  */
-int add_to_page_cache(struct page *page, struct address_space *mapping,
+int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
                pgoff_t offset, gfp_t gfp_mask)
 {
-       int error = mem_cgroup_cache_charge(page, current->mm,
+       int error;
+
+       VM_BUG_ON(!PageLocked(page));
+
+       error = mem_cgroup_cache_charge(page, current->mm,
                                        gfp_mask & ~__GFP_HIGHMEM);
        if (error)
                goto out;
 
        error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
        if (error == 0) {
-               write_lock_irq(&mapping->tree_lock);
+               page_cache_get(page);
+               page->mapping = mapping;
+               page->index = offset;
+
+               spin_lock_irq(&mapping->tree_lock);
                error = radix_tree_insert(&mapping->page_tree, offset, page);
-               if (!error) {
-                       page_cache_get(page);
-                       SetPageLocked(page);
-                       page->mapping = mapping;
-                       page->index = offset;
+               if (likely(!error)) {
                        mapping->nrpages++;
                        __inc_zone_page_state(page, NR_FILE_PAGES);
-               } else
+               } else {
+                       page->mapping = NULL;
                        mem_cgroup_uncharge_cache_page(page);
+                       page_cache_release(page);
+               }
 
-               write_unlock_irq(&mapping->tree_lock);
+               spin_unlock_irq(&mapping->tree_lock);
                radix_tree_preload_end();
        } else
                mem_cgroup_uncharge_cache_page(page);
 out:
        return error;
 }
-EXPORT_SYMBOL(add_to_page_cache);
+EXPORT_SYMBOL(add_to_page_cache_locked);
 
 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
                                pgoff_t offset, gfp_t gfp_mask)
@@ -633,15 +637,35 @@ void __lock_page_nosync(struct page *page)
  * Is there a pagecache struct page at the given (mapping, offset) tuple?
  * If yes, increment its refcount and return it; if no, return NULL.
  */
-struct page * find_get_page(struct address_space *mapping, pgoff_t offset)
+struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
 {
+       void **pagep;
        struct page *page;
 
-       read_lock_irq(&mapping->tree_lock);
-       page = radix_tree_lookup(&mapping->page_tree, offset);
-       if (page)
-               page_cache_get(page);
-       read_unlock_irq(&mapping->tree_lock);
+       rcu_read_lock();
+repeat:
+       page = NULL;
+       pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
+       if (pagep) {
+               page = radix_tree_deref_slot(pagep);
+               if (unlikely(!page || page == RADIX_TREE_RETRY))
+                       goto repeat;
+
+               if (!page_cache_get_speculative(page))
+                       goto repeat;
+
+               /*
+                * Has the page moved?
+                * This is part of the lockless pagecache protocol. See
+                * include/linux/pagemap.h for details.
+                */
+               if (unlikely(page != *pagep)) {
+                       page_cache_release(page);
+                       goto repeat;
+               }
+       }
+       rcu_read_unlock();
+
        return page;
 }
 EXPORT_SYMBOL(find_get_page);
@@ -656,32 +680,22 @@ EXPORT_SYMBOL(find_get_page);
  *
  * Returns zero if the page was not present. find_lock_page() may sleep.
  */
-struct page *find_lock_page(struct address_space *mapping,
-                               pgoff_t offset)
+struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
 {
        struct page *page;
 
 repeat:
-       read_lock_irq(&mapping->tree_lock);
-       page = radix_tree_lookup(&mapping->page_tree, offset);
+       page = find_get_page(mapping, offset);
        if (page) {
-               page_cache_get(page);
-               if (TestSetPageLocked(page)) {
-                       read_unlock_irq(&mapping->tree_lock);
-                       __lock_page(page);
-
-                       /* Has the page been truncated while we slept? */
-                       if (unlikely(page->mapping != mapping)) {
-                               unlock_page(page);
-                               page_cache_release(page);
-                               goto repeat;
-                       }
-                       VM_BUG_ON(page->index != offset);
-                       goto out;
+               lock_page(page);
+               /* Has the page been truncated? */
+               if (unlikely(page->mapping != mapping)) {
+                       unlock_page(page);
+                       page_cache_release(page);
+                       goto repeat;
                }
+               VM_BUG_ON(page->index != offset);
        }
-       read_unlock_irq(&mapping->tree_lock);
-out:
        return page;
 }
 EXPORT_SYMBOL(find_lock_page);
@@ -747,13 +761,39 @@ unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
 {
        unsigned int i;
        unsigned int ret;
+       unsigned int nr_found;
+
+       rcu_read_lock();
+restart:
+       nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
+                               (void ***)pages, start, nr_pages);
+       ret = 0;
+       for (i = 0; i < nr_found; i++) {
+               struct page *page;
+repeat:
+               page = radix_tree_deref_slot((void **)pages[i]);
+               if (unlikely(!page))
+                       continue;
+               /*
+                * this can only trigger if nr_found == 1, making livelock
+                * a non issue.
+                */
+               if (unlikely(page == RADIX_TREE_RETRY))
+                       goto restart;
 
-       read_lock_irq(&mapping->tree_lock);
-       ret = radix_tree_gang_lookup(&mapping->page_tree,
-                               (void **)pages, start, nr_pages);
-       for (i = 0; i < ret; i++)
-               page_cache_get(pages[i]);
-       read_unlock_irq(&mapping->tree_lock);
+               if (!page_cache_get_speculative(page))
+                       goto repeat;
+
+               /* Has the page moved? */
+               if (unlikely(page != *((void **)pages[i]))) {
+                       page_cache_release(page);
+                       goto repeat;
+               }
+
+               pages[ret] = page;
+               ret++;
+       }
+       rcu_read_unlock();
        return ret;
 }
 
@@ -774,19 +814,44 @@ unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
 {
        unsigned int i;
        unsigned int ret;
+       unsigned int nr_found;
+
+       rcu_read_lock();
+restart:
+       nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
+                               (void ***)pages, index, nr_pages);
+       ret = 0;
+       for (i = 0; i < nr_found; i++) {
+               struct page *page;
+repeat:
+               page = radix_tree_deref_slot((void **)pages[i]);
+               if (unlikely(!page))
+                       continue;
+               /*
+                * this can only trigger if nr_found == 1, making livelock
+                * a non issue.
+                */
+               if (unlikely(page == RADIX_TREE_RETRY))
+                       goto restart;
 
-       read_lock_irq(&mapping->tree_lock);
-       ret = radix_tree_gang_lookup(&mapping->page_tree,
-                               (void **)pages, index, nr_pages);
-       for (i = 0; i < ret; i++) {
-               if (pages[i]->mapping == NULL || pages[i]->index != index)
+               if (page->mapping == NULL || page->index != index)
                        break;
 
-               page_cache_get(pages[i]);
+               if (!page_cache_get_speculative(page))
+                       goto repeat;
+
+               /* Has the page moved? */
+               if (unlikely(page != *((void **)pages[i]))) {
+                       page_cache_release(page);
+                       goto repeat;
+               }
+
+               pages[ret] = page;
+               ret++;
                index++;
        }
-       read_unlock_irq(&mapping->tree_lock);
-       return i;
+       rcu_read_unlock();
+       return ret;
 }
 EXPORT_SYMBOL(find_get_pages_contig);
 
@@ -806,15 +871,43 @@ unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
 {
        unsigned int i;
        unsigned int ret;
+       unsigned int nr_found;
+
+       rcu_read_lock();
+restart:
+       nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
+                               (void ***)pages, *index, nr_pages, tag);
+       ret = 0;
+       for (i = 0; i < nr_found; i++) {
+               struct page *page;
+repeat:
+               page = radix_tree_deref_slot((void **)pages[i]);
+               if (unlikely(!page))
+                       continue;
+               /*
+                * this can only trigger if nr_found == 1, making livelock
+                * a non issue.
+                */
+               if (unlikely(page == RADIX_TREE_RETRY))
+                       goto restart;
+
+               if (!page_cache_get_speculative(page))
+                       goto repeat;
+
+               /* Has the page moved? */
+               if (unlikely(page != *((void **)pages[i]))) {
+                       page_cache_release(page);
+                       goto repeat;
+               }
+
+               pages[ret] = page;
+               ret++;
+       }
+       rcu_read_unlock();
 
-       read_lock_irq(&mapping->tree_lock);
-       ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
-                               (void **)pages, *index, nr_pages, tag);
-       for (i = 0; i < ret; i++)
-               page_cache_get(pages[i]);
        if (ret)
                *index = pages[ret - 1]->index + 1;
-       read_unlock_irq(&mapping->tree_lock);
+
        return ret;
 }
 EXPORT_SYMBOL(find_get_pages_tag);
@@ -1665,8 +1758,9 @@ static int __remove_suid(struct dentry *dentry, int kill)
        return notify_change(dentry, &newattrs);
 }
 
-int remove_suid(struct dentry *dentry)
+int file_remove_suid(struct file *file)
 {
+       struct dentry *dentry = file->f_path.dentry;
        int killsuid = should_remove_suid(dentry);
        int killpriv = security_inode_need_killpriv(dentry);
        int error = 0;
@@ -1680,7 +1774,7 @@ int remove_suid(struct dentry *dentry)
 
        return error;
 }
-EXPORT_SYMBOL(remove_suid);
+EXPORT_SYMBOL(file_remove_suid);
 
 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
                        const struct iovec *iov, size_t base, size_t bytes)
@@ -2436,7 +2530,7 @@ __generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
        if (count == 0)
                goto out;
 
-       err = remove_suid(file->f_path.dentry);
+       err = file_remove_suid(file);
        if (err)
                goto out;