ceph: address space operations

The ceph address space methods are concerned primarily with managing
the dirty page accounting in the inode, which (among other things)
must keep track of which snapshot context each page was dirtied in,
and ensure that dirty data is written out to the OSDs in snapshort
order.

A writepage() on a page that is not currently writeable due to
snapshot writeback ordering constraints is ignored (it was presumably
called from kswapd).

Signed-off-by: Sage Weil <sage@newdream.net>

fs/ceph/addr.c

@@ -0,0 +1,1115 @@
+#include "ceph_debug.h"
+
+#include <linux/backing-dev.h>
+#include <linux/fs.h>
+#include <linux/mm.h>
+#include <linux/pagemap.h>
+#include <linux/writeback.h>	/* generic_writepages */
+#include <linux/pagevec.h>
+#include <linux/task_io_accounting_ops.h>
+
+#include "super.h"
+#include "osd_client.h"
+
+/*
+ * Ceph address space ops.
+ *
+ * There are a few funny things going on here.
+ *
+ * The page->private field is used to reference a struct
+ * ceph_snap_context for _every_ dirty page.  This indicates which
+ * snapshot the page was logically dirtied in, and thus which snap
+ * context needs to be associated with the osd write during writeback.
+ *
+ * Similarly, struct ceph_inode_info maintains a set of counters to
+ * count dirty pages on the inode.  In the absense of snapshots,
+ * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
+ *
+ * When a snapshot is taken (that is, when the client receives
+ * notification that a snapshot was taken), each inode with caps and
+ * with dirty pages (dirty pages implies there is a cap) gets a new
+ * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
+ * order, new snaps go to the tail).  The i_wrbuffer_ref_head count is
+ * moved to capsnap->dirty. (Unless a sync write is currently in
+ * progress.  In that case, the capsnap is said to be "pending", new
+ * writes cannot start, and the capsnap isn't "finalized" until the
+ * write completes (or fails) and a final size/mtime for the inode for
+ * that snap can be settled upon.)  i_wrbuffer_ref_head is reset to 0.
+ *
+ * On writeback, we must submit writes to the osd IN SNAP ORDER.  So,
+ * we look for the first capsnap in i_cap_snaps and write out pages in
+ * that snap context _only_.  Then we move on to the next capsnap,
+ * eventually reaching the "live" or "head" context (i.e., pages that
+ * are not yet snapped) and are writing the most recently dirtied
+ * pages.
+ *
+ * Invalidate and so forth must take care to ensure the dirty page
+ * accounting is preserved.
+ */
+
+
+/*
+ * Dirty a page.  Optimistically adjust accounting, on the assumption
+ * that we won't race with invalidate.  If we do, readjust.
+ */
+static int ceph_set_page_dirty(struct page *page)
+{
+	struct address_space *mapping = page->mapping;
+	struct inode *inode;
+	struct ceph_inode_info *ci;
+	int undo = 0;
+	struct ceph_snap_context *snapc;
+
+	if (unlikely(!mapping))
+		return !TestSetPageDirty(page);
+
+	if (TestSetPageDirty(page)) {
+		dout("%p set_page_dirty %p idx %lu -- already dirty\n",
+		     mapping->host, page, page->index);
+		return 0;
+	}
+
+	inode = mapping->host;
+	ci = ceph_inode(inode);
+
+	/*
+	 * Note that we're grabbing a snapc ref here without holding
+	 * any locks!
+	 */
+	snapc = ceph_get_snap_context(ci->i_snap_realm->cached_context);
+
+	/* dirty the head */
+	spin_lock(&inode->i_lock);
+	if (ci->i_wrbuffer_ref_head == 0)
+		ci->i_head_snapc = ceph_get_snap_context(snapc);
+	++ci->i_wrbuffer_ref_head;
+	if (ci->i_wrbuffer_ref == 0)
+		igrab(inode);
+	++ci->i_wrbuffer_ref;
+	dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
+	     "snapc %p seq %lld (%d snaps)\n",
+	     mapping->host, page, page->index,
+	     ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
+	     ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
+	     snapc, snapc->seq, snapc->num_snaps);
+	spin_unlock(&inode->i_lock);
+
+	/* now adjust page */
+	spin_lock_irq(&mapping->tree_lock);
+	if (page->mapping) {	/* Race with truncate? */
+		WARN_ON_ONCE(!PageUptodate(page));
+
+		if (mapping_cap_account_dirty(mapping)) {
+			__inc_zone_page_state(page, NR_FILE_DIRTY);
+			__inc_bdi_stat(mapping->backing_dev_info,
+					BDI_RECLAIMABLE);
+			task_io_account_write(PAGE_CACHE_SIZE);
+		}
+		radix_tree_tag_set(&mapping->page_tree,
+				page_index(page), PAGECACHE_TAG_DIRTY);
+
+		/*
+		 * Reference snap context in page->private.  Also set
+		 * PagePrivate so that we get invalidatepage callback.
+		 */
+		page->private = (unsigned long)snapc;
+		SetPagePrivate(page);
+	} else {
+		dout("ANON set_page_dirty %p (raced truncate?)\n", page);
+		undo = 1;
+	}
+
+	spin_unlock_irq(&mapping->tree_lock);
+
+	if (undo)
+		/* whoops, we failed to dirty the page */
+		ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
+
+	__mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
+
+	BUG_ON(!PageDirty(page));
+	return 1;
+}
+
+/*
+ * If we are truncating the full page (i.e. offset == 0), adjust the
+ * dirty page counters appropriately.  Only called if there is private
+ * data on the page.
+ */
+static void ceph_invalidatepage(struct page *page, unsigned long offset)
+{
+	struct inode *inode = page->mapping->host;
+	struct ceph_inode_info *ci;
+	struct ceph_snap_context *snapc = (void *)page->private;
+
+	BUG_ON(!PageLocked(page));
+	BUG_ON(!page->private);
+	BUG_ON(!PagePrivate(page));
+	BUG_ON(!page->mapping);
+
+	/*
+	 * We can get non-dirty pages here due to races between
+	 * set_page_dirty and truncate_complete_page; just spit out a
+	 * warning, in case we end up with accounting problems later.
+	 */
+	if (!PageDirty(page))
+		pr_err("%p invalidatepage %p page not dirty\n", inode, page);
+
+	if (offset == 0)
+		ClearPageChecked(page);
+
+	ci = ceph_inode(inode);
+	if (offset == 0) {
+		dout("%p invalidatepage %p idx %lu full dirty page %lu\n",
+		     inode, page, page->index, offset);
+		ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
+		ceph_put_snap_context(snapc);
+		page->private = 0;
+		ClearPagePrivate(page);
+	} else {
+		dout("%p invalidatepage %p idx %lu partial dirty page\n",
+		     inode, page, page->index);
+	}
+}
+
+/* just a sanity check */
+static int ceph_releasepage(struct page *page, gfp_t g)
+{
+	struct inode *inode = page->mapping ? page->mapping->host : NULL;
+	dout("%p releasepage %p idx %lu\n", inode, page, page->index);
+	WARN_ON(PageDirty(page));
+	WARN_ON(page->private);
+	WARN_ON(PagePrivate(page));
+	return 0;
+}
+
+/*
+ * read a single page, without unlocking it.
+ */
+static int readpage_nounlock(struct file *filp, struct page *page)
+{
+	struct inode *inode = filp->f_dentry->d_inode;
+	struct ceph_inode_info *ci = ceph_inode(inode);
+	struct ceph_osd_client *osdc = &ceph_inode_to_client(inode)->osdc;
+	int err = 0;
+	u64 len = PAGE_CACHE_SIZE;
+
+	dout("readpage inode %p file %p page %p index %lu\n",
+	     inode, filp, page, page->index);
+	err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
+				  page->index << PAGE_CACHE_SHIFT, &len,
+				  ci->i_truncate_seq, ci->i_truncate_size,
+				  &page, 1);
+	if (err == -ENOENT)
+		err = 0;
+	if (err < 0) {
+		SetPageError(page);
+		goto out;
+	} else if (err < PAGE_CACHE_SIZE) {
+		/* zero fill remainder of page */
+		zero_user_segment(page, err, PAGE_CACHE_SIZE);
+	}
+	SetPageUptodate(page);
+
+out:
+	return err < 0 ? err : 0;
+}
+
+static int ceph_readpage(struct file *filp, struct page *page)
+{
+	int r = readpage_nounlock(filp, page);
+	unlock_page(page);
+	return r;
+}
+
+/*
+ * Build a vector of contiguous pages from the provided page list.
+ */
+static struct page **page_vector_from_list(struct list_head *page_list,
+					   unsigned *nr_pages)
+{
+	struct page **pages;
+	struct page *page;
+	int next_index, contig_pages = 0;
+
+	/* build page vector */
+	pages = kmalloc(sizeof(*pages) * *nr_pages, GFP_NOFS);
+	if (!pages)
+		return ERR_PTR(-ENOMEM);
+
+	BUG_ON(list_empty(page_list));
+	next_index = list_entry(page_list->prev, struct page, lru)->index;
+	list_for_each_entry_reverse(page, page_list, lru) {
+		if (page->index == next_index) {
+			dout("readpages page %d %p\n", contig_pages, page);
+			pages[contig_pages] = page;
+			contig_pages++;
+			next_index++;
+		} else {
+			break;
+		}
+	}
+	*nr_pages = contig_pages;
+	return pages;
+}
+
+/*
+ * Read multiple pages.  Leave pages we don't read + unlock in page_list;
+ * the caller (VM) cleans them up.
+ */
+static int ceph_readpages(struct file *file, struct address_space *mapping,
+			  struct list_head *page_list, unsigned nr_pages)
+{
+	struct inode *inode = file->f_dentry->d_inode;
+	struct ceph_inode_info *ci = ceph_inode(inode);
+	struct ceph_osd_client *osdc = &ceph_inode_to_client(inode)->osdc;
+	int rc = 0;
+	struct page **pages;
+	struct pagevec pvec;
+	loff_t offset;
+	u64 len;
+
+	dout("readpages %p file %p nr_pages %d\n",
+	     inode, file, nr_pages);
+
+	pages = page_vector_from_list(page_list, &nr_pages);
+	if (IS_ERR(pages))
+		return PTR_ERR(pages);
+
+	/* guess read extent */
+	offset = pages[0]->index << PAGE_CACHE_SHIFT;
+	len = nr_pages << PAGE_CACHE_SHIFT;
+	rc = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
+				 offset, &len,
+				 ci->i_truncate_seq, ci->i_truncate_size,
+				 pages, nr_pages);
+	if (rc == -ENOENT)
+		rc = 0;
+	if (rc < 0)
+		goto out;
+
+	/* set uptodate and add to lru in pagevec-sized chunks */
+	pagevec_init(&pvec, 0);
+	for (; !list_empty(page_list) && len > 0;
+	     rc -= PAGE_CACHE_SIZE, len -= PAGE_CACHE_SIZE) {
+		struct page *page =
+			list_entry(page_list->prev, struct page, lru);
+
+		list_del(&page->lru);
+
+		if (rc < (int)PAGE_CACHE_SIZE) {
+			/* zero (remainder of) page */
+			int s = rc < 0 ? 0 : rc;
+			zero_user_segment(page, s, PAGE_CACHE_SIZE);
+		}
+
+		if (add_to_page_cache(page, mapping, page->index, GFP_NOFS)) {
+			page_cache_release(page);
+			dout("readpages %p add_to_page_cache failed %p\n",
+			     inode, page);
+			continue;
+		}
+		dout("readpages %p adding %p idx %lu\n", inode, page,
+		     page->index);
+		flush_dcache_page(page);
+		SetPageUptodate(page);
+		unlock_page(page);
+		if (pagevec_add(&pvec, page) == 0)
+			pagevec_lru_add_file(&pvec);   /* add to lru */
+	}
+	pagevec_lru_add_file(&pvec);
+	rc = 0;
+
+out:
+	kfree(pages);
+	return rc;
+}
+
+/*
+ * Get ref for the oldest snapc for an inode with dirty data... that is, the
+ * only snap context we are allowed to write back.
+ *
+ * Caller holds i_lock.
+ */
+static struct ceph_snap_context *__get_oldest_context(struct inode *inode,
+						      u64 *snap_size)
+{
+	struct ceph_inode_info *ci = ceph_inode(inode);
+	struct ceph_snap_context *snapc = NULL;
+	struct ceph_cap_snap *capsnap = NULL;
+
+	list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
+		dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
+		     capsnap->context, capsnap->dirty_pages);
+		if (capsnap->dirty_pages) {
+			snapc = ceph_get_snap_context(capsnap->context);
+			if (snap_size)
+				*snap_size = capsnap->size;
+			break;
+		}
+	}
+	if (!snapc && ci->i_snap_realm) {
+		snapc = ceph_get_snap_context(ci->i_snap_realm->cached_context);
+		dout(" head snapc %p has %d dirty pages\n",
+		     snapc, ci->i_wrbuffer_ref_head);
+	}
+	return snapc;
+}
+
+static struct ceph_snap_context *get_oldest_context(struct inode *inode,
+						    u64 *snap_size)
+{
+	struct ceph_snap_context *snapc = NULL;
+
+	spin_lock(&inode->i_lock);
+	snapc = __get_oldest_context(inode, snap_size);
+	spin_unlock(&inode->i_lock);
+	return snapc;
+}
+
+/*
+ * Write a single page, but leave the page locked.
+ *
+ * If we get a write error, set the page error bit, but still adjust the
+ * dirty page accounting (i.e., page is no longer dirty).
+ */
+static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
+{
+	struct inode *inode;
+	struct ceph_inode_info *ci;
+	struct ceph_osd_client *osdc;
+	loff_t page_off = page->index << PAGE_CACHE_SHIFT;
+	int len = PAGE_CACHE_SIZE;
+	loff_t i_size;
+	int err = 0;
+	struct ceph_snap_context *snapc;
+	u64 snap_size = 0;
+
+	dout("writepage %p idx %lu\n", page, page->index);
+
+	if (!page->mapping || !page->mapping->host) {
+		dout("writepage %p - no mapping\n", page);
+		return -EFAULT;
+	}
+	inode = page->mapping->host;
+	ci = ceph_inode(inode);
+	osdc = &ceph_inode_to_client(inode)->osdc;
+
+	/* verify this is a writeable snap context */
+	snapc = (void *)page->private;
+	if (snapc == NULL) {
+		dout("writepage %p page %p not dirty?\n", inode, page);
+		goto out;
+	}
+	if (snapc != get_oldest_context(inode, &snap_size)) {
+		dout("writepage %p page %p snapc %p not writeable - noop\n",
+		     inode, page, (void *)page->private);
+		/* we should only noop if called by kswapd */
+		WARN_ON((current->flags & PF_MEMALLOC) == 0);
+		goto out;
+	}
+
+	/* is this a partial page at end of file? */
+	if (snap_size)
+		i_size = snap_size;
+	else
+		i_size = i_size_read(inode);
+	if (i_size < page_off + len)
+		len = i_size - page_off;
+
+	dout("writepage %p page %p index %lu on %llu~%u\n",
+	     inode, page, page->index, page_off, len);
+
+	set_page_writeback(page);
+	err = ceph_osdc_writepages(osdc, ceph_vino(inode),
+				   &ci->i_layout, snapc,
+				   page_off, len,
+				   ci->i_truncate_seq, ci->i_truncate_size,
+				   &inode->i_mtime,
+				   &page, 1, 0, 0, true);
+	if (err < 0) {
+		dout("writepage setting page/mapping error %d %p\n", err, page);
+		SetPageError(page);
+		mapping_set_error(&inode->i_data, err);
+		if (wbc)
+			wbc->pages_skipped++;
+	} else {
+		dout("writepage cleaned page %p\n", page);
+		err = 0;  /* vfs expects us to return 0 */
+	}
+	page->private = 0;
+	ClearPagePrivate(page);
+	end_page_writeback(page);
+	ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
+	ceph_put_snap_context(snapc);
+out:
+	return err;
+}
+
+static int ceph_writepage(struct page *page, struct writeback_control *wbc)
+{
+	int err = writepage_nounlock(page, wbc);
+	unlock_page(page);
+	return err;
+}
+
+
+/*
+ * lame release_pages helper.  release_pages() isn't exported to
+ * modules.
+ */
+static void ceph_release_pages(struct page **pages, int num)
+{
+	struct pagevec pvec;
+	int i;
+
+	pagevec_init(&pvec, 0);
+	for (i = 0; i < num; i++) {
+		if (pagevec_add(&pvec, pages[i]) == 0)
+			pagevec_release(&pvec);
+	}
+	pagevec_release(&pvec);
+}
+
+
+/*
+ * async writeback completion handler.
+ *
+ * If we get an error, set the mapping error bit, but not the individual
+ * page error bits.
+ */
+static void writepages_finish(struct ceph_osd_request *req,
+			      struct ceph_msg *msg)
+{
+	struct inode *inode = req->r_inode;
+	struct ceph_osd_reply_head *replyhead;
+	struct ceph_osd_op *op;
+	struct ceph_inode_info *ci = ceph_inode(inode);
+	unsigned wrote;
+	loff_t offset = req->r_pages[0]->index << PAGE_CACHE_SHIFT;
+	struct page *page;
+	int i;
+	struct ceph_snap_context *snapc = req->r_snapc;
+	struct address_space *mapping = inode->i_mapping;
+	struct writeback_control *wbc = req->r_wbc;
+	__s32 rc = -EIO;
+	u64 bytes = 0;
+
+	/* parse reply */
+	replyhead = msg->front.iov_base;
+	WARN_ON(le32_to_cpu(replyhead->num_ops) == 0);
+	op = (void *)(replyhead + 1);
+	rc = le32_to_cpu(replyhead->result);
+	bytes = le64_to_cpu(op->extent.length);
+
+	if (rc >= 0) {
+		wrote = (bytes + (offset & ~PAGE_CACHE_MASK) + ~PAGE_CACHE_MASK)
+			>> PAGE_CACHE_SHIFT;
+		WARN_ON(wrote != req->r_num_pages);
+	} else {
+		wrote = 0;
+		mapping_set_error(mapping, rc);
+	}
+	dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n",
+	     inode, rc, bytes, wrote);
+
+	/* clean all pages */
+	for (i = 0; i < req->r_num_pages; i++) {
+		page = req->r_pages[i];
+		BUG_ON(!page);
+		WARN_ON(!PageUptodate(page));
+
+		if (i >= wrote) {
+			dout("inode %p skipping page %p\n", inode, page);
+			wbc->pages_skipped++;
+		}
+		page->private = 0;
+		ClearPagePrivate(page);
+		ceph_put_snap_context(snapc);
+		dout("unlocking %d %p\n", i, page);
+		end_page_writeback(page);
+		unlock_page(page);
+	}
+	dout("%p wrote+cleaned %d pages\n", inode, wrote);
+	ceph_put_wrbuffer_cap_refs(ci, req->r_num_pages, snapc);
+
+	ceph_release_pages(req->r_pages, req->r_num_pages);
+	if (req->r_pages_from_pool)
+		mempool_free(req->r_pages,
+			     ceph_client(inode->i_sb)->wb_pagevec_pool);
+	else
+		kfree(req->r_pages);
+	ceph_osdc_put_request(req);
+}
+
+/*
+ * allocate a page vec, either directly, or if necessary, via a the
+ * mempool.  we avoid the mempool if we can because req->r_num_pages
+ * may be less than the maximum write size.
+ */
+static void alloc_page_vec(struct ceph_client *client,
+			   struct ceph_osd_request *req)
+{
+	req->r_pages = kmalloc(sizeof(struct page *) * req->r_num_pages,
+			       GFP_NOFS);
+	if (!req->r_pages) {
+		req->r_pages = mempool_alloc(client->wb_pagevec_pool, GFP_NOFS);
+		req->r_pages_from_pool = 1;
+		WARN_ON(!req->r_pages);
+	}
+}
+
+/*
+ * initiate async writeback
+ */
+static int ceph_writepages_start(struct address_space *mapping,
+				 struct writeback_control *wbc)
+{
+	struct inode *inode = mapping->host;
+	struct backing_dev_info *bdi = mapping->backing_dev_info;
+	struct ceph_inode_info *ci = ceph_inode(inode);
+	struct ceph_client *client = ceph_inode_to_client(inode);
+	pgoff_t index, start, end;
+	int range_whole = 0;
+	int should_loop = 1;
+	pgoff_t max_pages = 0, max_pages_ever = 0;
+	struct ceph_snap_context *snapc = NULL, *last_snapc = NULL;
+	struct pagevec pvec;
+	int done = 0;
+	int rc = 0;
+	unsigned wsize = 1 << inode->i_blkbits;
+	struct ceph_osd_request *req = NULL;
+	int do_sync;
+	u64 snap_size = 0;
+
+	/*
+	 * Include a 'sync' in the OSD request if this is a data
+	 * integrity write (e.g., O_SYNC write or fsync()), or if our
+	 * cap is being revoked.
+	 */
+	do_sync = wbc->sync_mode == WB_SYNC_ALL;
+	if (ceph_caps_revoking(ci, CEPH_CAP_FILE_BUFFER))
+		do_sync = 1;
+	dout("writepages_start %p dosync=%d (mode=%s)\n",
+	     inode, do_sync,
+	     wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
+	     (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
+
+	client = ceph_inode_to_client(inode);
+	if (client->mount_state == CEPH_MOUNT_SHUTDOWN) {
+		pr_warning("writepage_start %p on forced umount\n", inode);
+		return -EIO; /* we're in a forced umount, don't write! */
+	}
+	if (client->mount_args.wsize && client->mount_args.wsize < wsize)
+		wsize = client->mount_args.wsize;
+	if (wsize < PAGE_CACHE_SIZE)
+		wsize = PAGE_CACHE_SIZE;
+	max_pages_ever = wsize >> PAGE_CACHE_SHIFT;
+
+	pagevec_init(&pvec, 0);
+
+	/* ?? */
+	if (wbc->nonblocking && bdi_write_congested(bdi)) {
+		dout(" writepages congested\n");
+		wbc->encountered_congestion = 1;
+		goto out_final;
+	}
+
+	/* where to start/end? */
+	if (wbc->range_cyclic) {
+		start = mapping->writeback_index; /* Start from prev offset */
+		end = -1;
+		dout(" cyclic, start at %lu\n", start);
+	} else {
+		start = wbc->range_start >> PAGE_CACHE_SHIFT;
+		end = wbc->range_end >> PAGE_CACHE_SHIFT;
+		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
+			range_whole = 1;
+		should_loop = 0;
+		dout(" not cyclic, %lu to %lu\n", start, end);
+	}
+	index = start;
+
+retry:
+	/* find oldest snap context with dirty data */
+	ceph_put_snap_context(snapc);
+	snapc = get_oldest_context(inode, &snap_size);
+	if (!snapc) {
+		/* hmm, why does writepages get called when there
+		   is no dirty data? */
+		dout(" no snap context with dirty data?\n");
+		goto out;
+	}
+	dout(" oldest snapc is %p seq %lld (%d snaps)\n",
+	     snapc, snapc->seq, snapc->num_snaps);
+	if (last_snapc && snapc != last_snapc) {
+		/* if we switched to a newer snapc, restart our scan at the
+		 * start of the original file range. */
+		dout("  snapc differs from last pass, restarting at %lu\n",
+		     index);
+		index = start;
+	}
+	last_snapc = snapc;
+
+	while (!done && index <= end) {
+		unsigned i;
+		int first;
+		pgoff_t next;
+		int pvec_pages, locked_pages;
+		struct page *page;
+		int want;
+		u64 offset, len;
+		struct ceph_osd_request_head *reqhead;
+		struct ceph_osd_op *op;
+
+		next = 0;
+		locked_pages = 0;
+		max_pages = max_pages_ever;
+
+get_more_pages:
+		first = -1;
+		want = min(end - index,
+			   min((pgoff_t)PAGEVEC_SIZE,
+			       max_pages - (pgoff_t)locked_pages) - 1)
+			+ 1;
+		pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index,
+						PAGECACHE_TAG_DIRTY,
+						want);
+		dout("pagevec_lookup_tag got %d\n", pvec_pages);
+		if (!pvec_pages && !locked_pages)
+			break;
+		for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
+			page = pvec.pages[i];
+			dout("? %p idx %lu\n", page, page->index);
+			if (locked_pages == 0)
+				lock_page(page);  /* first page */
+			else if (!trylock_page(page))
+				break;
+
+			/* only dirty pages, or our accounting breaks */
+			if (unlikely(!PageDirty(page)) ||
+			    unlikely(page->mapping != mapping)) {
+				dout("!dirty or !mapping %p\n", page);
+				unlock_page(page);
+				break;
+			}
+			if (!wbc->range_cyclic && page->index > end) {
+				dout("end of range %p\n", page);
+				done = 1;
+				unlock_page(page);
+				break;
+			}
+			if (next && (page->index != next)) {
+				dout("not consecutive %p\n", page);
+				unlock_page(page);
+				break;
+			}
+			if (wbc->sync_mode != WB_SYNC_NONE) {
+				dout("waiting on writeback %p\n", page);
+				wait_on_page_writeback(page);
+			}
+			if ((snap_size && page_offset(page) > snap_size) ||
+			    (!snap_size &&
+			     page_offset(page) > i_size_read(inode))) {
+				dout("%p page eof %llu\n", page, snap_size ?
+				     snap_size : i_size_read(inode));
+				done = 1;
+				unlock_page(page);
+				break;
+			}
+			if (PageWriteback(page)) {
+				dout("%p under writeback\n", page);
+				unlock_page(page);
+				break;
+			}
+
+			/* only if matching snap context */
+			if (snapc != (void *)page->private) {
+				dout("page snapc %p != oldest %p\n",
+				     (void *)page->private, snapc);
+				unlock_page(page);
+				if (!locked_pages)
+					continue; /* keep looking for snap */
+				break;
+			}
+
+			if (!clear_page_dirty_for_io(page)) {
+				dout("%p !clear_page_dirty_for_io\n", page);
+				unlock_page(page);
+				break;
+			}
+
+			/* ok */
+			if (locked_pages == 0) {
+				/* prepare async write request */
+				offset = page->index << PAGE_CACHE_SHIFT;
+				len = wsize;
+				req = ceph_osdc_new_request(&client->osdc,
+					    &ci->i_layout,
+					    ceph_vino(inode),
+					    offset, &len,
+					    CEPH_OSD_OP_WRITE,
+					    CEPH_OSD_FLAG_WRITE |
+						    CEPH_OSD_FLAG_ONDISK,
+					    snapc, do_sync,
+					    ci->i_truncate_seq,
+					    ci->i_truncate_size,
+					    &inode->i_mtime, true, 1);
+				max_pages = req->r_num_pages;
+
+				alloc_page_vec(client, req);
+				req->r_callback = writepages_finish;
+				req->r_inode = inode;
+				req->r_wbc = wbc;
+			}
+
+			/* note position of first page in pvec */
+			if (first < 0)
+				first = i;
+			dout("%p will write page %p idx %lu\n",
+			     inode, page, page->index);
+			set_page_writeback(page);
+			req->r_pages[locked_pages] = page;
+			locked_pages++;
+			next = page->index + 1;
+		}
+
+		/* did we get anything? */
+		if (!locked_pages)
+			goto release_pvec_pages;
+		if (i) {
+			int j;
+			BUG_ON(!locked_pages || first < 0);
+
+			if (pvec_pages && i == pvec_pages &&
+			    locked_pages < max_pages) {
+				dout("reached end pvec, trying for more\n");
+				pagevec_reinit(&pvec);
+				goto get_more_pages;
+			}
+
+			/* shift unused pages over in the pvec...  we
+			 * will need to release them below. */
+			for (j = i; j < pvec_pages; j++) {
+				dout(" pvec leftover page %p\n",
+				     pvec.pages[j]);
+				pvec.pages[j-i+first] = pvec.pages[j];
+			}
+			pvec.nr -= i-first;
+		}
+
+		/* submit the write */
+		offset = req->r_pages[0]->index << PAGE_CACHE_SHIFT;
+		len = min((snap_size ? snap_size : i_size_read(inode)) - offset,
+			  (u64)locked_pages << PAGE_CACHE_SHIFT);
+		dout("writepages got %d pages at %llu~%llu\n",
+		     locked_pages, offset, len);
+
+		/* revise final length, page count */
+		req->r_num_pages = locked_pages;
+		reqhead = req->r_request->front.iov_base;
+		op = (void *)(reqhead + 1);
+		op->extent.length = cpu_to_le64(len);
+		op->payload_len = cpu_to_le32(len);
+		req->r_request->hdr.data_len = cpu_to_le32(len);
+
+		ceph_osdc_start_request(&client->osdc, req, true);
+		req = NULL;
+
+		/* continue? */
+		index = next;
+		wbc->nr_to_write -= locked_pages;
+		if (wbc->nr_to_write <= 0)
+			done = 1;
+
+release_pvec_pages:
+		dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
+		     pvec.nr ? pvec.pages[0] : NULL);
+		pagevec_release(&pvec);
+
+		if (locked_pages && !done)
+			goto retry;
+	}
+
+	if (should_loop && !done) {
+		/* more to do; loop back to beginning of file */
+		dout("writepages looping back to beginning of file\n");
+		should_loop = 0;
+		index = 0;
+		goto retry;
+	}
+
+	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
+		mapping->writeback_index = index;
+
+out:
+	if (req)
+		ceph_osdc_put_request(req);
+	if (rc > 0)
+		rc = 0;  /* vfs expects us to return 0 */
+	ceph_put_snap_context(snapc);
+	dout("writepages done, rc = %d\n", rc);
+out_final:
+	return rc;
+}
+
+
+
+/*
+ * See if a given @snapc is either writeable, or already written.
+ */
+static int context_is_writeable_or_written(struct inode *inode,
+					   struct ceph_snap_context *snapc)
+{
+	struct ceph_snap_context *oldest = get_oldest_context(inode, NULL);
+	return !oldest || snapc->seq <= oldest->seq;
+}
+
+/*
+ * We are only allowed to write into/dirty the page if the page is
+ * clean, or already dirty within the same snap context.
+ */
+static int ceph_write_begin(struct file *file, struct address_space *mapping,
+			    loff_t pos, unsigned len, unsigned flags,
+			    struct page **pagep, void **fsdata)
+{
+	struct inode *inode = file->f_dentry->d_inode;
+	struct ceph_inode_info *ci = ceph_inode(inode);
+	struct ceph_mds_client *mdsc = &ceph_inode_to_client(inode)->mdsc;
+	struct page *page;
+	pgoff_t index = pos >> PAGE_CACHE_SHIFT;
+	loff_t page_off = pos & PAGE_CACHE_MASK;
+	int pos_in_page = pos & ~PAGE_CACHE_MASK;
+	int end_in_page = pos_in_page + len;
+	loff_t i_size;
+	struct ceph_snap_context *snapc;
+	int r;
+
+	/* get a page*/
+retry:
+	page = grab_cache_page_write_begin(mapping, index, 0);
+	if (!page)
+		return -ENOMEM;
+	*pagep = page;
+
+	dout("write_begin file %p inode %p page %p %d~%d\n", file,
+	     inode, page, (int)pos, (int)len);
+
+retry_locked:
+	/* writepages currently holds page lock, but if we change that later, */
+	wait_on_page_writeback(page);
+
+	/* check snap context */
+	BUG_ON(!ci->i_snap_realm);
+	down_read(&mdsc->snap_rwsem);
+	BUG_ON(!ci->i_snap_realm->cached_context);
+	if (page->private &&
+	    (void *)page->private != ci->i_snap_realm->cached_context) {
+		/*
+		 * this page is already dirty in another (older) snap
+		 * context!  is it writeable now?
+		 */
+		snapc = get_oldest_context(inode, NULL);
+		up_read(&mdsc->snap_rwsem);
+
+		if (snapc != (void *)page->private) {
+			dout(" page %p snapc %p not current or oldest\n",
+			     page, (void *)page->private);
+			/*
+			 * queue for writeback, and wait for snapc to
+			 * be writeable or written
+			 */
+			snapc = ceph_get_snap_context((void *)page->private);
+			unlock_page(page);
+			if (ceph_queue_writeback(inode))
+				igrab(inode);
+			wait_event_interruptible(ci->i_cap_wq,
+			       context_is_writeable_or_written(inode, snapc));
+			ceph_put_snap_context(snapc);
+			goto retry;
+		}
+
+		/* yay, writeable, do it now (without dropping page lock) */
+		dout(" page %p snapc %p not current, but oldest\n",
+		     page, snapc);
+		if (!clear_page_dirty_for_io(page))
+			goto retry_locked;
+		r = writepage_nounlock(page, NULL);
+		if (r < 0)
+			goto fail_nosnap;
+		goto retry_locked;
+	}
+
+	if (PageUptodate(page)) {
+		dout(" page %p already uptodate\n", page);
+		return 0;
+	}
+
+	/* full page? */
+	if (pos_in_page == 0 && len == PAGE_CACHE_SIZE)
+		return 0;
+
+	/* past end of file? */
+	i_size = inode->i_size;   /* caller holds i_mutex */
+
+	if (i_size + len > inode->i_sb->s_maxbytes) {
+		/* file is too big */
+		r = -EINVAL;
+		goto fail;
+	}
+
+	if (page_off >= i_size ||
+	    (pos_in_page == 0 && (pos+len) >= i_size &&
+	     end_in_page - pos_in_page != PAGE_CACHE_SIZE)) {
+		dout(" zeroing %p 0 - %d and %d - %d\n",
+		     page, pos_in_page, end_in_page, (int)PAGE_CACHE_SIZE);
+		zero_user_segments(page,
+				   0, pos_in_page,
+				   end_in_page, PAGE_CACHE_SIZE);
+		return 0;
+	}
+
+	/* we need to read it. */
+	up_read(&mdsc->snap_rwsem);
+	r = readpage_nounlock(file, page);
+	if (r < 0)
+		goto fail_nosnap;
+	goto retry_locked;
+
+fail:
+	up_read(&mdsc->snap_rwsem);
+fail_nosnap:
+	unlock_page(page);
+	return r;
+}
+
+/*
+ * we don't do anything in here that simple_write_end doesn't do
+ * except adjust dirty page accounting and drop read lock on
+ * mdsc->snap_rwsem.
+ */
+static int ceph_write_end(struct file *file, struct address_space *mapping,
+			  loff_t pos, unsigned len, unsigned copied,
+			  struct page *page, void *fsdata)
+{
+	struct inode *inode = file->f_dentry->d_inode;
+	struct ceph_mds_client *mdsc = &ceph_inode_to_client(inode)->mdsc;
+	unsigned from = pos & (PAGE_CACHE_SIZE - 1);
+	int check_cap = 0;
+
+	dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
+	     inode, page, (int)pos, (int)copied, (int)len);
+
+	/* zero the stale part of the page if we did a short copy */
+	if (copied < len)
+		zero_user_segment(page, from+copied, len);
+
+	/* did file size increase? */
+	/* (no need for i_size_read(); we caller holds i_mutex */
+	if (pos+copied > inode->i_size)
+		check_cap = ceph_inode_set_size(inode, pos+copied);
+
+	if (!PageUptodate(page))
+		SetPageUptodate(page);
+
+	set_page_dirty(page);
+
+	unlock_page(page);
+	up_read(&mdsc->snap_rwsem);
+	page_cache_release(page);
+
+	if (check_cap)
+		ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
+
+	return copied;
+}
+
+/*
+ * we set .direct_IO to indicate direct io is supported, but since we
+ * intercept O_DIRECT reads and writes early, this function should
+ * never get called.
+ */
+static ssize_t ceph_direct_io(int rw, struct kiocb *iocb,
+			      const struct iovec *iov,
+			      loff_t pos, unsigned long nr_segs)
+{
+	WARN_ON(1);
+	return -EINVAL;
+}
+
+const struct address_space_operations ceph_aops = {
+	.readpage = ceph_readpage,
+	.readpages = ceph_readpages,
+	.writepage = ceph_writepage,
+	.writepages = ceph_writepages_start,
+	.write_begin = ceph_write_begin,
+	.write_end = ceph_write_end,
+	.set_page_dirty = ceph_set_page_dirty,
+	.invalidatepage = ceph_invalidatepage,
+	.releasepage = ceph_releasepage,
+	.direct_IO = ceph_direct_io,
+};
+
+
+/*
+ * vm ops
+ */
+
+/*
+ * Reuse write_begin here for simplicity.
+ */
+static int ceph_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
+{
+	struct inode *inode = vma->vm_file->f_dentry->d_inode;
+	struct page *page = vmf->page;
+	struct ceph_mds_client *mdsc = &ceph_inode_to_client(inode)->mdsc;
+	loff_t off = page->index << PAGE_CACHE_SHIFT;
+	loff_t size, len;
+	struct page *locked_page = NULL;
+	void *fsdata = NULL;
+	int ret;
+
+	size = i_size_read(inode);
+	if (off + PAGE_CACHE_SIZE <= size)
+		len = PAGE_CACHE_SIZE;
+	else
+		len = size & ~PAGE_CACHE_MASK;
+
+	dout("page_mkwrite %p %llu~%llu page %p idx %lu\n", inode,
+	     off, len, page, page->index);
+	ret = ceph_write_begin(vma->vm_file, inode->i_mapping, off, len, 0,
+			       &locked_page, &fsdata);
+	WARN_ON(page != locked_page);
+	if (!ret) {
+		/*
+		 * doing the following, instead of calling
+		 * ceph_write_end. Note that we keep the
+		 * page locked
+		 */
+		set_page_dirty(page);
+		up_read(&mdsc->snap_rwsem);
+		page_cache_release(page);
+		ret = VM_FAULT_LOCKED;
+	} else {
+		ret = VM_FAULT_SIGBUS;
+	}
+	dout("page_mkwrite %p %llu~%llu = %d\n", inode, off, len, ret);
+	return ret;
+}
+
+static struct vm_operations_struct ceph_vmops = {
+	.fault		= filemap_fault,
+	.page_mkwrite	= ceph_page_mkwrite,
+};
+
+int ceph_mmap(struct file *file, struct vm_area_struct *vma)
+{
+	struct address_space *mapping = file->f_mapping;
+
+	if (!mapping->a_ops->readpage)
+		return -ENOEXEC;
+	file_accessed(file);
+	vma->vm_ops = &ceph_vmops;
+	vma->vm_flags |= VM_CAN_NONLINEAR;
+	return 0;
+}