ceph: support ceph_pagelist for message payload
authorSage Weil <sage@newdream.net>
Wed, 23 Dec 2009 20:12:31 +0000 (12:12 -0800)
committerSage Weil <sage@newdream.net>
Wed, 23 Dec 2009 20:12:31 +0000 (12:12 -0800)
The ceph_pagelist is a simple list of whole pages, strung together via
their lru list_head.  It facilitates encoding to a "buffer" of unknown
size.  Allow its use in place of the ceph_msg page vector.

This will be used to fix the huge buffer preallocation woes of MDS
reconnection.

Signed-off-by: Sage Weil <sage@newdream.net>
fs/ceph/Makefile
fs/ceph/messenger.c
fs/ceph/messenger.h
fs/ceph/pagelist.c [new file with mode: 0644]
fs/ceph/pagelist.h [new file with mode: 0644]

index 827629c85768b67506ceadf04c68e2e574145ee0..47caf2f1b75af704f22038af52996e6d2e20ff22 100644 (file)
@@ -8,7 +8,7 @@ obj-$(CONFIG_CEPH_FS) += ceph.o
 
 ceph-objs := super.o inode.o dir.o file.o addr.o ioctl.o \
        export.o caps.o snap.o xattr.o \
-       messenger.o msgpool.o buffer.o \
+       messenger.o msgpool.o buffer.o pagelist.o \
        mds_client.o mdsmap.o \
        mon_client.o \
        osd_client.o osdmap.o crush/crush.o crush/mapper.o crush/hash.o \
index 68052f6642808bda98e3a6226860699c186a491e..c1106e8360f0f458442bec7849da6c180372f7db 100644 (file)
@@ -13,6 +13,7 @@
 #include "super.h"
 #include "messenger.h"
 #include "decode.h"
+#include "pagelist.h"
 
 /*
  * Ceph uses the messenger to exchange ceph_msg messages with other
@@ -728,6 +729,11 @@ static int write_partial_msg_pages(struct ceph_connection *con)
                        page = msg->pages[con->out_msg_pos.page];
                        if (crc)
                                kaddr = kmap(page);
+               } else if (msg->pagelist) {
+                       page = list_first_entry(&msg->pagelist->head,
+                                               struct page, lru);
+                       if (crc)
+                               kaddr = kmap(page);
                } else {
                        page = con->msgr->zero_page;
                        if (crc)
@@ -750,7 +756,7 @@ static int write_partial_msg_pages(struct ceph_connection *con)
                                      MSG_DONTWAIT | MSG_NOSIGNAL |
                                      MSG_MORE);
 
-               if (crc && msg->pages)
+               if (crc && (msg->pages || msg->pagelist))
                        kunmap(page);
 
                if (ret <= 0)
@@ -762,6 +768,9 @@ static int write_partial_msg_pages(struct ceph_connection *con)
                        con->out_msg_pos.page_pos = 0;
                        con->out_msg_pos.page++;
                        con->out_msg_pos.did_page_crc = 0;
+                       if (msg->pagelist)
+                               list_move_tail(&page->lru,
+                                              &msg->pagelist->head);
                }
        }
 
@@ -1051,13 +1060,13 @@ static int process_banner(struct ceph_connection *con)
                                       &con->actual_peer_addr) &&
            !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
              con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
-               pr_err("wrong peer, want %s/%d, "
-                      "got %s/%d, wtf\n",
+               pr_warning("wrong peer, want %s/%d, "
+                      "got %s/%d\n",
                       pr_addr(&con->peer_addr.in_addr),
                       con->peer_addr.nonce,
                       pr_addr(&con->actual_peer_addr.in_addr),
                       con->actual_peer_addr.nonce);
-               con->error_msg = "protocol error, wrong peer";
+               con->error_msg = "wrong peer at address";
                return -1;
        }
 
@@ -2096,6 +2105,7 @@ struct ceph_msg *ceph_msg_new(int type, int front_len,
        /* data */
        m->nr_pages = calc_pages_for(page_off, page_len);
        m->pages = pages;
+       m->pagelist = NULL;
 
        dout("ceph_msg_new %p page %d~%d -> %d\n", m, page_off, page_len,
             m->nr_pages);
@@ -2181,6 +2191,12 @@ void ceph_msg_last_put(struct kref *kref)
        m->nr_pages = 0;
        m->pages = NULL;
 
+       if (m->pagelist) {
+               ceph_pagelist_release(m->pagelist);
+               kfree(m->pagelist);
+               m->pagelist = NULL;
+       }
+
        if (m->pool)
                ceph_msgpool_put(m->pool, m);
        else
index 7e2aab1d3ce2bd9ff13025cec401df4d2a5c880a..a7b6841450924557a3ca6396f05b80dcaa5d43bb 100644 (file)
@@ -85,6 +85,7 @@ struct ceph_msg {
        struct ceph_buffer *middle;
        struct page **pages;            /* data payload.  NOT OWNER. */
        unsigned nr_pages;              /* size of page array */
+       struct ceph_pagelist *pagelist; /* instead of pages */
        struct list_head list_head;
        struct kref kref;
        bool front_is_vmalloc;
diff --git a/fs/ceph/pagelist.c b/fs/ceph/pagelist.c
new file mode 100644 (file)
index 0000000..370e936
--- /dev/null
@@ -0,0 +1,54 @@
+
+#include <linux/pagemap.h>
+#include <linux/highmem.h>
+
+#include "pagelist.h"
+
+int ceph_pagelist_release(struct ceph_pagelist *pl)
+{
+       if (pl->mapped_tail)
+               kunmap(pl->mapped_tail);
+       while (!list_empty(&pl->head)) {
+               struct page *page = list_first_entry(&pl->head, struct page,
+                                                    lru);
+               list_del(&page->lru);
+               __free_page(page);
+       }
+       return 0;
+}
+
+static int ceph_pagelist_addpage(struct ceph_pagelist *pl)
+{
+       struct page *page = alloc_page(GFP_NOFS);
+       if (!page)
+               return -ENOMEM;
+       pl->room += PAGE_SIZE;
+       list_add_tail(&page->lru, &pl->head);
+       if (pl->mapped_tail)
+               kunmap(pl->mapped_tail);
+       pl->mapped_tail = kmap(page);
+       return 0;
+}
+
+int ceph_pagelist_append(struct ceph_pagelist *pl, void *buf, size_t len)
+{
+       while (pl->room < len) {
+               size_t bit = pl->room;
+               int ret;
+
+               memcpy(pl->mapped_tail + (pl->length & ~PAGE_CACHE_MASK),
+                      buf, bit);
+               pl->length += bit;
+               pl->room -= bit;
+               buf += bit;
+               len -= bit;
+               ret = ceph_pagelist_addpage(pl);
+               if (ret)
+                       return ret;
+       }
+
+       memcpy(pl->mapped_tail + (pl->length & ~PAGE_CACHE_MASK), buf, len);
+       pl->length += len;
+       pl->room -= len;
+       return 0;
+}
diff --git a/fs/ceph/pagelist.h b/fs/ceph/pagelist.h
new file mode 100644 (file)
index 0000000..e8a4187
--- /dev/null
@@ -0,0 +1,54 @@
+#ifndef __FS_CEPH_PAGELIST_H
+#define __FS_CEPH_PAGELIST_H
+
+#include <linux/list.h>
+
+struct ceph_pagelist {
+       struct list_head head;
+       void *mapped_tail;
+       size_t length;
+       size_t room;
+};
+
+static inline void ceph_pagelist_init(struct ceph_pagelist *pl)
+{
+       INIT_LIST_HEAD(&pl->head);
+       pl->mapped_tail = NULL;
+       pl->length = 0;
+       pl->room = 0;
+}
+extern int ceph_pagelist_release(struct ceph_pagelist *pl);
+
+extern int ceph_pagelist_append(struct ceph_pagelist *pl, void *d, size_t l);
+
+static inline int ceph_pagelist_encode_64(struct ceph_pagelist *pl, u64 v)
+{
+       __le64 ev = cpu_to_le64(v);
+       return ceph_pagelist_append(pl, &ev, sizeof(ev));
+}
+static inline int ceph_pagelist_encode_32(struct ceph_pagelist *pl, u32 v)
+{
+       __le32 ev = cpu_to_le32(v);
+       return ceph_pagelist_append(pl, &ev, sizeof(ev));
+}
+static inline int ceph_pagelist_encode_16(struct ceph_pagelist *pl, u16 v)
+{
+       __le16 ev = cpu_to_le16(v);
+       return ceph_pagelist_append(pl, &ev, sizeof(ev));
+}
+static inline int ceph_pagelist_encode_8(struct ceph_pagelist *pl, u8 v)
+{
+       return ceph_pagelist_append(pl, &v, 1);
+}
+static inline int ceph_pagelist_encode_string(struct ceph_pagelist *pl,
+                                             char *s, size_t len)
+{
+       int ret = ceph_pagelist_encode_32(pl, len);
+       if (ret)
+               return ret;
+       if (len)
+               return ceph_pagelist_append(pl, s, len);
+       return 0;
+}
+
+#endif