ceph: file operations

File open and close operations, and read and write methods that ensure
we have obtained the proper capabilities from the MDS cluster before
performing IO on a file.  We take references on held capabilities for
the duration of the read/write to avoid prematurely releasing them
back to the MDS.

We implement two main paths for read and write: one that is buffered
(and uses generic_aio_{read,write}), and one that is fully synchronous
and blocking (operating either on a __user pointer or, if O_DIRECT,
directly on user pages).

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

fs/ceph/file.c

@@ -0,0 +1,904 @@
+#include "ceph_debug.h"
+
+#include <linux/sched.h>
+#include <linux/file.h>
+#include <linux/namei.h>
+#include <linux/writeback.h>
+
+#include "super.h"
+#include "mds_client.h"
+
+/*
+ * Ceph file operations
+ *
+ * Implement basic open/close functionality, and implement
+ * read/write.
+ *
+ * We implement three modes of file I/O:
+ *  - buffered uses the generic_file_aio_{read,write} helpers
+ *
+ *  - synchronous is used when there is multi-client read/write
+ *    sharing, avoids the page cache, and synchronously waits for an
+ *    ack from the OSD.
+ *
+ *  - direct io takes the variant of the sync path that references
+ *    user pages directly.
+ *
+ * fsync() flushes and waits on dirty pages, but just queues metadata
+ * for writeback: since the MDS can recover size and mtime there is no
+ * need to wait for MDS acknowledgement.
+ */
+
+
+/*
+ * Prepare an open request.  Preallocate ceph_cap to avoid an
+ * inopportune ENOMEM later.
+ */
+static struct ceph_mds_request *
+prepare_open_request(struct super_block *sb, int flags, int create_mode)
+{
+	struct ceph_client *client = ceph_sb_to_client(sb);
+	struct ceph_mds_client *mdsc = &client->mdsc;
+	struct ceph_mds_request *req;
+	int want_auth = USE_ANY_MDS;
+	int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
+
+	if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
+		want_auth = USE_AUTH_MDS;
+
+	req = ceph_mdsc_create_request(mdsc, op, want_auth);
+	if (IS_ERR(req))
+		goto out;
+	req->r_fmode = ceph_flags_to_mode(flags);
+	req->r_args.open.flags = cpu_to_le32(flags);
+	req->r_args.open.mode = cpu_to_le32(create_mode);
+	req->r_args.open.preferred = -1;
+out:
+	return req;
+}
+
+/*
+ * initialize private struct file data.
+ * if we fail, clean up by dropping fmode reference on the ceph_inode
+ */
+static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
+{
+	struct ceph_file_info *cf;
+	int ret = 0;
+
+	switch (inode->i_mode & S_IFMT) {
+	case S_IFREG:
+	case S_IFDIR:
+		dout("init_file %p %p 0%o (regular)\n", inode, file,
+		     inode->i_mode);
+		cf = kmem_cache_alloc(ceph_file_cachep, GFP_NOFS | __GFP_ZERO);
+		if (cf == NULL) {
+			ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
+			return -ENOMEM;
+		}
+		cf->fmode = fmode;
+		cf->next_offset = 2;
+		file->private_data = cf;
+		BUG_ON(inode->i_fop->release != ceph_release);
+		break;
+
+	case S_IFLNK:
+		dout("init_file %p %p 0%o (symlink)\n", inode, file,
+		     inode->i_mode);
+		ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
+		break;
+
+	default:
+		dout("init_file %p %p 0%o (special)\n", inode, file,
+		     inode->i_mode);
+		/*
+		 * we need to drop the open ref now, since we don't
+		 * have .release set to ceph_release.
+		 */
+		ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
+		BUG_ON(inode->i_fop->release == ceph_release);
+
+		/* call the proper open fop */
+		ret = inode->i_fop->open(inode, file);
+	}
+	return ret;
+}
+
+/*
+ * If the filp already has private_data, that means the file was
+ * already opened by intent during lookup, and we do nothing.
+ *
+ * If we already have the requisite capabilities, we can satisfy
+ * the open request locally (no need to request new caps from the
+ * MDS).  We do, however, need to inform the MDS (asynchronously)
+ * if our wanted caps set expands.
+ */
+int ceph_open(struct inode *inode, struct file *file)
+{
+	struct ceph_inode_info *ci = ceph_inode(inode);
+	struct ceph_client *client = ceph_sb_to_client(inode->i_sb);
+	struct ceph_mds_client *mdsc = &client->mdsc;
+	struct ceph_mds_request *req;
+	struct ceph_file_info *cf = file->private_data;
+	struct inode *parent_inode = file->f_dentry->d_parent->d_inode;
+	int err;
+	int flags, fmode, wanted;
+
+	if (cf) {
+		dout("open file %p is already opened\n", file);
+		return 0;
+	}
+
+	/* filter out O_CREAT|O_EXCL; vfs did that already.  yuck. */
+	flags = file->f_flags & ~(O_CREAT|O_EXCL);
+	if (S_ISDIR(inode->i_mode))
+		flags = O_DIRECTORY;  /* mds likes to know */
+
+	dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
+	     ceph_vinop(inode), file, flags, file->f_flags);
+	fmode = ceph_flags_to_mode(flags);
+	wanted = ceph_caps_for_mode(fmode);
+
+	/* snapped files are read-only */
+	if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
+		return -EROFS;
+
+	/* trivially open snapdir */
+	if (ceph_snap(inode) == CEPH_SNAPDIR) {
+		spin_lock(&inode->i_lock);
+		__ceph_get_fmode(ci, fmode);
+		spin_unlock(&inode->i_lock);
+		return ceph_init_file(inode, file, fmode);
+	}
+
+	/*
+	 * No need to block if we have any caps.  Update wanted set
+	 * asynchronously.
+	 */
+	spin_lock(&inode->i_lock);
+	if (__ceph_is_any_real_caps(ci)) {
+		int mds_wanted = __ceph_caps_mds_wanted(ci);
+		int issued = __ceph_caps_issued(ci, NULL);
+
+		dout("open %p fmode %d want %s issued %s using existing\n",
+		     inode, fmode, ceph_cap_string(wanted),
+		     ceph_cap_string(issued));
+		__ceph_get_fmode(ci, fmode);
+		spin_unlock(&inode->i_lock);
+
+		/* adjust wanted? */
+		if ((issued & wanted) != wanted &&
+		    (mds_wanted & wanted) != wanted &&
+		    ceph_snap(inode) != CEPH_SNAPDIR)
+			ceph_check_caps(ci, 0, NULL);
+
+		return ceph_init_file(inode, file, fmode);
+	} else if (ceph_snap(inode) != CEPH_NOSNAP &&
+		   (ci->i_snap_caps & wanted) == wanted) {
+		__ceph_get_fmode(ci, fmode);
+		spin_unlock(&inode->i_lock);
+		return ceph_init_file(inode, file, fmode);
+	}
+	spin_unlock(&inode->i_lock);
+
+	dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
+	req = prepare_open_request(inode->i_sb, flags, 0);
+	if (IS_ERR(req)) {
+		err = PTR_ERR(req);
+		goto out;
+	}
+	req->r_inode = igrab(inode);
+	req->r_num_caps = 1;
+	err = ceph_mdsc_do_request(mdsc, parent_inode, req);
+	if (!err)
+		err = ceph_init_file(inode, file, req->r_fmode);
+	ceph_mdsc_put_request(req);
+	dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
+out:
+	return err;
+}
+
+
+/*
+ * Do a lookup + open with a single request.
+ *
+ * If this succeeds, but some subsequent check in the vfs
+ * may_open() fails, the struct *file gets cleaned up (i.e.
+ * ceph_release gets called).  So fear not!
+ */
+/*
+ * flags
+ *  path_lookup_open   -> LOOKUP_OPEN
+ *  path_lookup_create -> LOOKUP_OPEN|LOOKUP_CREATE
+ */
+struct dentry *ceph_lookup_open(struct inode *dir, struct dentry *dentry,
+				struct nameidata *nd, int mode,
+				int locked_dir)
+{
+	struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
+	struct ceph_mds_client *mdsc = &client->mdsc;
+	struct file *file = nd->intent.open.file;
+	struct inode *parent_inode = get_dentry_parent_inode(file->f_dentry);
+	struct ceph_mds_request *req;
+	int err;
+	int flags = nd->intent.open.flags - 1;  /* silly vfs! */
+
+	dout("ceph_lookup_open dentry %p '%.*s' flags %d mode 0%o\n",
+	     dentry, dentry->d_name.len, dentry->d_name.name, flags, mode);
+
+	/* do the open */
+	req = prepare_open_request(dir->i_sb, flags, mode);
+	if (IS_ERR(req))
+		return ERR_PTR(PTR_ERR(req));
+	req->r_dentry = dget(dentry);
+	req->r_num_caps = 2;
+	if (flags & O_CREAT) {
+		req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
+		req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
+	}
+	req->r_locked_dir = dir;           /* caller holds dir->i_mutex */
+	err = ceph_mdsc_do_request(mdsc, parent_inode, req);
+	dentry = ceph_finish_lookup(req, dentry, err);
+	if (!err && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
+		err = ceph_handle_notrace_create(dir, dentry);
+	if (!err)
+		err = ceph_init_file(req->r_dentry->d_inode, file,
+				     req->r_fmode);
+	ceph_mdsc_put_request(req);
+	dout("ceph_lookup_open result=%p\n", dentry);
+	return dentry;
+}
+
+int ceph_release(struct inode *inode, struct file *file)
+{
+	struct ceph_inode_info *ci = ceph_inode(inode);
+	struct ceph_file_info *cf = file->private_data;
+
+	dout("release inode %p file %p\n", inode, file);
+	ceph_put_fmode(ci, cf->fmode);
+	if (cf->last_readdir)
+		ceph_mdsc_put_request(cf->last_readdir);
+	kfree(cf->last_name);
+	kfree(cf->dir_info);
+	dput(cf->dentry);
+	kmem_cache_free(ceph_file_cachep, cf);
+	return 0;
+}
+
+/*
+ * build a vector of user pages
+ */
+static struct page **get_direct_page_vector(const char __user *data,
+					    int num_pages,
+					    loff_t off, size_t len)
+{
+	struct page **pages;
+	int rc;
+
+	pages = kmalloc(sizeof(*pages) * num_pages, GFP_NOFS);
+	if (!pages)
+		return ERR_PTR(-ENOMEM);
+
+	down_read(&current->mm->mmap_sem);
+	rc = get_user_pages(current, current->mm, (unsigned long)data,
+			    num_pages, 0, 0, pages, NULL);
+	up_read(&current->mm->mmap_sem);
+	if (rc < 0)
+		goto fail;
+	return pages;
+
+fail:
+	kfree(pages);
+	return ERR_PTR(rc);
+}
+
+static void put_page_vector(struct page **pages, int num_pages)
+{
+	int i;
+
+	for (i = 0; i < num_pages; i++)
+		put_page(pages[i]);
+	kfree(pages);
+}
+
+void ceph_release_page_vector(struct page **pages, int num_pages)
+{
+	int i;
+
+	for (i = 0; i < num_pages; i++)
+		__free_pages(pages[i], 0);
+	kfree(pages);
+}
+
+/*
+ * allocate a vector new pages
+ */
+static struct page **alloc_page_vector(int num_pages)
+{
+	struct page **pages;
+	int i;
+
+	pages = kmalloc(sizeof(*pages) * num_pages, GFP_NOFS);
+	if (!pages)
+		return ERR_PTR(-ENOMEM);
+	for (i = 0; i < num_pages; i++) {
+		pages[i] = alloc_page(GFP_NOFS);
+		if (pages[i] == NULL) {
+			ceph_release_page_vector(pages, i);
+			return ERR_PTR(-ENOMEM);
+		}
+	}
+	return pages;
+}
+
+/*
+ * copy user data into a page vector
+ */
+static int copy_user_to_page_vector(struct page **pages,
+				    const char __user *data,
+				    loff_t off, size_t len)
+{
+	int i = 0;
+	int po = off & ~PAGE_CACHE_MASK;
+	int left = len;
+	int l, bad;
+
+	while (left > 0) {
+		l = min_t(int, PAGE_CACHE_SIZE-po, left);
+		bad = copy_from_user(page_address(pages[i]) + po, data, l);
+		if (bad == l)
+			return -EFAULT;
+		data += l - bad;
+		left -= l - bad;
+		if (po) {
+			po += l - bad;
+			if (po == PAGE_CACHE_SIZE)
+				po = 0;
+		}
+	}
+	return len;
+}
+
+/*
+ * copy user data from a page vector into a user pointer
+ */
+static int copy_page_vector_to_user(struct page **pages, char __user *data,
+				    loff_t off, size_t len)
+{
+	int i = 0;
+	int po = off & ~PAGE_CACHE_MASK;
+	int left = len;
+	int l, bad;
+
+	while (left > 0) {
+		l = min_t(int, left, PAGE_CACHE_SIZE-po);
+		bad = copy_to_user(data, page_address(pages[i]) + po, l);
+		if (bad == l)
+			return -EFAULT;
+		data += l - bad;
+		left -= l - bad;
+		if (po) {
+			po += l - bad;
+			if (po == PAGE_CACHE_SIZE)
+				po = 0;
+		}
+		i++;
+	}
+	return len;
+}
+
+/*
+ * Zero an extent within a page vector.  Offset is relative to the
+ * start of the first page.
+ */
+static void zero_page_vector_range(int off, int len, struct page **pages)
+{
+	int i = off >> PAGE_CACHE_SHIFT;
+
+	dout("zero_page_vector_page %u~%u\n", off, len);
+	BUG_ON(len < PAGE_CACHE_SIZE);
+
+	/* leading partial page? */
+	if (off & ~PAGE_CACHE_MASK) {
+		dout("zeroing %d %p head from %d\n", i, pages[i],
+		     (int)(off & ~PAGE_CACHE_MASK));
+		zero_user_segment(pages[i], off & ~PAGE_CACHE_MASK,
+				  PAGE_CACHE_SIZE);
+		off += PAGE_CACHE_SIZE;
+		off &= PAGE_CACHE_MASK;
+		i++;
+	}
+	while (len >= PAGE_CACHE_SIZE) {
+		dout("zeroing %d %p\n", i, pages[i]);
+		zero_user_segment(pages[i], 0, PAGE_CACHE_SIZE);
+		off += PAGE_CACHE_SIZE;
+		len -= PAGE_CACHE_SIZE;
+		i++;
+	}
+	/* trailing partial page? */
+	if (len) {
+		dout("zeroing %d %p tail to %d\n", i, pages[i], (int)len);
+		zero_user_segment(pages[i], 0, len);
+	}
+}
+
+
+/*
+ * Read a range of bytes striped over one or more objects.  Iterate over
+ * objects we stripe over.  (That's not atomic, but good enough for now.)
+ *
+ * If we get a short result from the OSD, check against i_size; we need to
+ * only return a short read to the caller if we hit EOF.
+ */
+static int striped_read(struct inode *inode,
+			u64 off, u64 len,
+			struct page **pages, int num_pages)
+{
+	struct ceph_client *client = ceph_inode_to_client(inode);
+	struct ceph_inode_info *ci = ceph_inode(inode);
+	u64 pos, this_len;
+	int page_off = off & ~PAGE_CACHE_SIZE; /* first byte's offset in page */
+	int left, pages_left;
+	int read;
+	struct page **page_pos;
+	int ret;
+	bool hit_stripe, was_short;
+
+	/*
+	 * we may need to do multiple reads.  not atomic, unfortunately.
+	 */
+	pos = off;
+	left = len;
+	page_pos = pages;
+	pages_left = num_pages;
+	read = 0;
+
+more:
+	this_len = left;
+	ret = ceph_osdc_readpages(&client->osdc, ceph_vino(inode),
+				  &ci->i_layout, pos, &this_len,
+				  ci->i_truncate_seq,
+				  ci->i_truncate_size,
+				  page_pos, pages_left);
+	hit_stripe = this_len < left;
+	was_short = ret >= 0 && ret < this_len;
+	if (ret == -ENOENT)
+		ret = 0;
+	dout("striped_read %llu~%u (read %u) got %d%s%s\n", pos, left, read,
+	     ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
+
+	if (ret > 0) {
+		int didpages =
+			((pos & ~PAGE_CACHE_MASK) + ret) >> PAGE_CACHE_SHIFT;
+
+		if (read < pos - off) {
+			dout(" zero gap %llu to %llu\n", off + read, pos);
+			zero_page_vector_range(page_off + read,
+					       pos - off - read, pages);
+		}
+		pos += ret;
+		read = pos - off;
+		left -= ret;
+		page_pos += didpages;
+		pages_left -= didpages;
+
+		/* hit stripe? */
+		if (left && hit_stripe)
+			goto more;
+	}
+
+	if (was_short) {
+		/* was original extent fully inside i_size? */
+		if (pos + left <= inode->i_size) {
+			dout("zero tail\n");
+			zero_page_vector_range(page_off + read, len - read,
+					       pages);
+			goto out;
+		}
+
+		/* check i_size */
+		ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE);
+		if (ret < 0)
+			goto out;
+
+		/* hit EOF? */
+		if (pos >= inode->i_size)
+			goto out;
+
+		goto more;
+	}
+
+out:
+	if (ret >= 0)
+		ret = read;
+	dout("striped_read returns %d\n", ret);
+	return ret;
+}
+
+/*
+ * Completely synchronous read and write methods.  Direct from __user
+ * buffer to osd, or directly to user pages (if O_DIRECT).
+ *
+ * If the read spans object boundary, just do multiple reads.
+ */
+static ssize_t ceph_sync_read(struct file *file, char __user *data,
+			      unsigned len, loff_t *poff)
+{
+	struct inode *inode = file->f_dentry->d_inode;
+	struct page **pages;
+	u64 off = *poff;
+	int num_pages = calc_pages_for(off, len);
+	int ret;
+
+	dout("sync_read on file %p %llu~%u %s\n", file, off, len,
+	     (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
+
+	if (file->f_flags & O_DIRECT) {
+		pages = get_direct_page_vector(data, num_pages, off, len);
+
+		/*
+		 * flush any page cache pages in this range.  this
+		 * will make concurrent normal and O_DIRECT io slow,
+		 * but it will at least behave sensibly when they are
+		 * in sequence.
+		 */
+		filemap_write_and_wait(inode->i_mapping);
+	} else {
+		pages = alloc_page_vector(num_pages);
+	}
+	if (IS_ERR(pages))
+		return PTR_ERR(pages);
+
+	ret = striped_read(inode, off, len, pages, num_pages);
+
+	if (ret >= 0 && (file->f_flags & O_DIRECT) == 0)
+		ret = copy_page_vector_to_user(pages, data, off, ret);
+	if (ret >= 0)
+		*poff = off + ret;
+
+	if (file->f_flags & O_DIRECT)
+		put_page_vector(pages, num_pages);
+	else
+		ceph_release_page_vector(pages, num_pages);
+	dout("sync_read result %d\n", ret);
+	return ret;
+}
+
+/*
+ * Write commit callback, called if we requested both an ACK and
+ * ONDISK commit reply from the OSD.
+ */
+static void sync_write_commit(struct ceph_osd_request *req,
+			      struct ceph_msg *msg)
+{
+	struct ceph_inode_info *ci = ceph_inode(req->r_inode);
+
+	dout("sync_write_commit %p tid %llu\n", req, req->r_tid);
+	spin_lock(&ci->i_unsafe_lock);
+	list_del_init(&req->r_unsafe_item);
+	spin_unlock(&ci->i_unsafe_lock);
+	ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
+}
+
+/*
+ * Synchronous write, straight from __user pointer or user pages (if
+ * O_DIRECT).
+ *
+ * If write spans object boundary, just do multiple writes.  (For a
+ * correct atomic write, we should e.g. take write locks on all
+ * objects, rollback on failure, etc.)
+ */
+static ssize_t ceph_sync_write(struct file *file, const char __user *data,
+			       size_t left, loff_t *offset)
+{
+	struct inode *inode = file->f_dentry->d_inode;
+	struct ceph_inode_info *ci = ceph_inode(inode);
+	struct ceph_client *client = ceph_inode_to_client(inode);
+	struct ceph_osd_request *req;
+	struct page **pages;
+	int num_pages;
+	long long unsigned pos;
+	u64 len;
+	int written = 0;
+	int flags;
+	int do_sync = 0;
+	int check_caps = 0;
+	int ret;
+	struct timespec mtime = CURRENT_TIME;
+
+	if (ceph_snap(file->f_dentry->d_inode) != CEPH_NOSNAP)
+		return -EROFS;
+
+	dout("sync_write on file %p %lld~%u %s\n", file, *offset,
+	     (unsigned)left, (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
+
+	if (file->f_flags & O_APPEND)
+		pos = i_size_read(inode);
+	else
+		pos = *offset;
+
+	flags = CEPH_OSD_FLAG_ORDERSNAP |
+		CEPH_OSD_FLAG_ONDISK |
+		CEPH_OSD_FLAG_WRITE;
+	if ((file->f_flags & (O_SYNC|O_DIRECT)) == 0)
+		flags |= CEPH_OSD_FLAG_ACK;
+	else
+		do_sync = 1;
+
+	/*
+	 * we may need to do multiple writes here if we span an object
+	 * boundary.  this isn't atomic, unfortunately.  :(
+	 */
+more:
+	len = left;
+	req = ceph_osdc_new_request(&client->osdc, &ci->i_layout,
+				    ceph_vino(inode), pos, &len,
+				    CEPH_OSD_OP_WRITE, flags,
+				    ci->i_snap_realm->cached_context,
+				    do_sync,
+				    ci->i_truncate_seq, ci->i_truncate_size,
+				    &mtime, false, 2);
+	if (IS_ERR(req))
+		return PTR_ERR(req);
+
+	num_pages = calc_pages_for(pos, len);
+
+	if (file->f_flags & O_DIRECT) {
+		pages = get_direct_page_vector(data, num_pages, pos, len);
+		if (IS_ERR(pages)) {
+			ret = PTR_ERR(pages);
+			goto out;
+		}
+
+		/*
+		 * throw out any page cache pages in this range. this
+		 * may block.
+		 */
+		truncate_inode_pages_range(inode->i_mapping, pos, pos+len);
+	} else {
+		pages = alloc_page_vector(num_pages);
+		if (IS_ERR(pages)) {
+			ret = PTR_ERR(pages);
+			goto out;
+		}
+		ret = copy_user_to_page_vector(pages, data, pos, len);
+		if (ret < 0) {
+			ceph_release_page_vector(pages, num_pages);
+			goto out;
+		}
+
+		if ((file->f_flags & O_SYNC) == 0) {
+			/* get a second commit callback */
+			req->r_safe_callback = sync_write_commit;
+			req->r_own_pages = 1;
+		}
+	}
+	req->r_pages = pages;
+	req->r_num_pages = num_pages;
+	req->r_inode = inode;
+
+	ret = ceph_osdc_start_request(&client->osdc, req, false);
+	if (!ret) {
+		if (req->r_safe_callback) {
+			/*
+			 * Add to inode unsafe list only after we
+			 * start_request so that a tid has been assigned.
+			 */
+			spin_lock(&ci->i_unsafe_lock);
+			list_add(&ci->i_unsafe_writes, &req->r_unsafe_item);
+			spin_unlock(&ci->i_unsafe_lock);
+			ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR);
+		}
+		ret = ceph_osdc_wait_request(&client->osdc, req);
+	}
+
+	if (file->f_flags & O_DIRECT)
+		put_page_vector(pages, num_pages);
+	else if (file->f_flags & O_SYNC)
+		ceph_release_page_vector(pages, num_pages);
+
+out:
+	ceph_osdc_put_request(req);
+	if (ret == 0) {
+		pos += len;
+		written += len;
+		left -= len;
+		if (left)
+			goto more;
+
+		ret = written;
+		*offset = pos;
+		if (pos > i_size_read(inode))
+			check_caps = ceph_inode_set_size(inode, pos);
+		if (check_caps)
+			ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY,
+					NULL);
+	}
+	return ret;
+}
+
+/*
+ * Wrap generic_file_aio_read with checks for cap bits on the inode.
+ * Atomically grab references, so that those bits are not released
+ * back to the MDS mid-read.
+ *
+ * Hmm, the sync read case isn't actually async... should it be?
+ */
+static ssize_t ceph_aio_read(struct kiocb *iocb, const struct iovec *iov,
+			     unsigned long nr_segs, loff_t pos)
+{
+	struct file *filp = iocb->ki_filp;
+	loff_t *ppos = &iocb->ki_pos;
+	size_t len = iov->iov_len;
+	struct inode *inode = filp->f_dentry->d_inode;
+	struct ceph_inode_info *ci = ceph_inode(inode);
+	ssize_t ret;
+	int got = 0;
+
+	dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
+	     inode, ceph_vinop(inode), pos, (unsigned)len, inode);
+	__ceph_do_pending_vmtruncate(inode);
+	ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, CEPH_CAP_FILE_CACHE,
+			    &got, -1);
+	if (ret < 0)
+		goto out;
+	dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
+	     inode, ceph_vinop(inode), pos, (unsigned)len,
+	     ceph_cap_string(got));
+
+	if ((got & CEPH_CAP_FILE_CACHE) == 0 ||
+	    (iocb->ki_filp->f_flags & O_DIRECT) ||
+	    (inode->i_sb->s_flags & MS_SYNCHRONOUS))
+		/* hmm, this isn't really async... */
+		ret = ceph_sync_read(filp, iov->iov_base, len, ppos);
+	else
+		ret = generic_file_aio_read(iocb, iov, nr_segs, pos);
+
+out:
+	dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
+	     inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
+	ceph_put_cap_refs(ci, got);
+	return ret;
+}
+
+/*
+ * Take cap references to avoid releasing caps to MDS mid-write.
+ *
+ * If we are synchronous, and write with an old snap context, the OSD
+ * may return EOLDSNAPC.  In that case, retry the write.. _after_
+ * dropping our cap refs and allowing the pending snap to logically
+ * complete _before_ this write occurs.
+ *
+ * If we are near ENOSPC, write synchronously.
+ */
+static ssize_t ceph_aio_write(struct kiocb *iocb, const struct iovec *iov,
+		       unsigned long nr_segs, loff_t pos)
+{
+	struct file *file = iocb->ki_filp;
+	struct inode *inode = file->f_dentry->d_inode;
+	struct ceph_inode_info *ci = ceph_inode(inode);
+	struct ceph_osd_client *osdc = &ceph_client(inode->i_sb)->osdc;
+	loff_t endoff = pos + iov->iov_len;
+	int got = 0;
+	int ret;
+
+	if (ceph_snap(inode) != CEPH_NOSNAP)
+		return -EROFS;
+
+retry_snap:
+	if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
+		return -ENOSPC;
+	__ceph_do_pending_vmtruncate(inode);
+	dout("aio_write %p %llx.%llx %llu~%u getting caps. i_size %llu\n",
+	     inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
+	     inode->i_size);
+	ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, CEPH_CAP_FILE_BUFFER,
+			    &got, endoff);
+	if (ret < 0)
+		goto out;
+
+	dout("aio_write %p %llx.%llx %llu~%u  got cap refs on %s\n",
+	     inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
+	     ceph_cap_string(got));
+
+	if ((got & CEPH_CAP_FILE_BUFFER) == 0 ||
+	    (iocb->ki_filp->f_flags & O_DIRECT) ||
+	    (inode->i_sb->s_flags & MS_SYNCHRONOUS)) {
+		ret = ceph_sync_write(file, iov->iov_base, iov->iov_len,
+			&iocb->ki_pos);
+	} else {
+		ret = generic_file_aio_write(iocb, iov, nr_segs, pos);
+
+		if ((ret >= 0 || ret == -EIOCBQUEUED) &&
+		    ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host)
+		     || ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL)))
+			ret = vfs_fsync_range(file, file->f_path.dentry,
+					      pos, pos + ret - 1, 1);
+	}
+	if (ret >= 0) {
+		spin_lock(&inode->i_lock);
+		__ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
+		spin_unlock(&inode->i_lock);
+	}
+
+out:
+	dout("aio_write %p %llx.%llx %llu~%u  dropping cap refs on %s\n",
+	     inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
+	     ceph_cap_string(got));
+	ceph_put_cap_refs(ci, got);
+
+	if (ret == -EOLDSNAPC) {
+		dout("aio_write %p %llx.%llx %llu~%u got EOLDSNAPC, retrying\n",
+		     inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len);
+		goto retry_snap;
+	}
+
+	return ret;
+}
+
+/*
+ * llseek.  be sure to verify file size on SEEK_END.
+ */
+static loff_t ceph_llseek(struct file *file, loff_t offset, int origin)
+{
+	struct inode *inode = file->f_mapping->host;
+	int ret;
+
+	mutex_lock(&inode->i_mutex);
+	__ceph_do_pending_vmtruncate(inode);
+	switch (origin) {
+	case SEEK_END:
+		ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE);
+		if (ret < 0) {
+			offset = ret;
+			goto out;
+		}
+		offset += inode->i_size;
+		break;
+	case SEEK_CUR:
+		/*
+		 * Here we special-case the lseek(fd, 0, SEEK_CUR)
+		 * position-querying operation.  Avoid rewriting the "same"
+		 * f_pos value back to the file because a concurrent read(),
+		 * write() or lseek() might have altered it
+		 */
+		if (offset == 0) {
+			offset = file->f_pos;
+			goto out;
+		}
+		offset += file->f_pos;
+		break;
+	}
+
+	if (offset < 0 || offset > inode->i_sb->s_maxbytes) {
+		offset = -EINVAL;
+		goto out;
+	}
+
+	/* Special lock needed here? */
+	if (offset != file->f_pos) {
+		file->f_pos = offset;
+		file->f_version = 0;
+	}
+
+out:
+	mutex_unlock(&inode->i_mutex);
+	return offset;
+}
+
+const struct file_operations ceph_file_fops = {
+	.open = ceph_open,
+	.release = ceph_release,
+	.llseek = ceph_llseek,
+	.read = do_sync_read,
+	.write = do_sync_write,
+	.aio_read = ceph_aio_read,
+	.aio_write = ceph_aio_write,
+	.mmap = ceph_mmap,
+	.fsync = ceph_fsync,
+	.splice_read = generic_file_splice_read,
+	.splice_write = generic_file_splice_write,
+	.unlocked_ioctl = ceph_ioctl,
+	.compat_ioctl	= ceph_ioctl,
+};
+