ceph: super.c
authorSage Weil <sage@newdream.net>
Tue, 6 Oct 2009 18:31:07 +0000 (11:31 -0700)
committerSage Weil <sage@newdream.net>
Tue, 6 Oct 2009 18:31:07 +0000 (11:31 -0700)
Mount option parsing, client setup and teardown, and a few odds and
ends (e.g., statfs).

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

diff --git a/fs/ceph/super.c b/fs/ceph/super.c
new file mode 100644 (file)
index 0000000..0723fb6
--- /dev/null
@@ -0,0 +1,936 @@
+
+#include "ceph_debug.h"
+
+#include <linux/backing-dev.h>
+#include <linux/fs.h>
+#include <linux/inet.h>
+#include <linux/in6.h>
+#include <linux/module.h>
+#include <linux/mount.h>
+#include <linux/parser.h>
+#include <linux/rwsem.h>
+#include <linux/sched.h>
+#include <linux/seq_file.h>
+#include <linux/statfs.h>
+#include <linux/string.h>
+#include <linux/version.h>
+#include <linux/vmalloc.h>
+
+#include "ceph_ver.h"
+#include "decode.h"
+#include "super.h"
+#include "mon_client.h"
+
+/*
+ * Ceph superblock operations
+ *
+ * Handle the basics of mounting, unmounting.
+ */
+
+
+/*
+ * find filename portion of a path (/foo/bar/baz -> baz)
+ */
+const char *ceph_file_part(const char *s, int len)
+{
+       const char *e = s + len;
+
+       while (e != s && *(e-1) != '/')
+               e--;
+       return e;
+}
+
+
+/*
+ * super ops
+ */
+static void ceph_put_super(struct super_block *s)
+{
+       struct ceph_client *cl = ceph_client(s);
+
+       dout("put_super\n");
+       ceph_mdsc_close_sessions(&cl->mdsc);
+       return;
+}
+
+static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
+{
+       struct ceph_client *client = ceph_inode_to_client(dentry->d_inode);
+       struct ceph_monmap *monmap = client->monc.monmap;
+       struct ceph_statfs st;
+       u64 fsid;
+       int err;
+
+       dout("statfs\n");
+       err = ceph_monc_do_statfs(&client->monc, &st);
+       if (err < 0)
+               return err;
+
+       /* fill in kstatfs */
+       buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
+
+       /*
+        * express utilization in terms of large blocks to avoid
+        * overflow on 32-bit machines.
+        */
+       buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
+       buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
+       buf->f_bfree = (le64_to_cpu(st.kb) - le64_to_cpu(st.kb_used)) >>
+               (CEPH_BLOCK_SHIFT-10);
+       buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
+
+       buf->f_files = le64_to_cpu(st.num_objects);
+       buf->f_ffree = -1;
+       buf->f_namelen = PATH_MAX;
+       buf->f_frsize = PAGE_CACHE_SIZE;
+
+       /* leave fsid little-endian, regardless of host endianness */
+       fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
+       buf->f_fsid.val[0] = fsid & 0xffffffff;
+       buf->f_fsid.val[1] = fsid >> 32;
+
+       return 0;
+}
+
+
+static int ceph_syncfs(struct super_block *sb, int wait)
+{
+       dout("sync_fs %d\n", wait);
+       ceph_osdc_sync(&ceph_client(sb)->osdc);
+       ceph_mdsc_sync(&ceph_client(sb)->mdsc);
+       return 0;
+}
+
+
+/**
+ * ceph_show_options - Show mount options in /proc/mounts
+ * @m: seq_file to write to
+ * @mnt: mount descriptor
+ */
+static int ceph_show_options(struct seq_file *m, struct vfsmount *mnt)
+{
+       struct ceph_client *client = ceph_sb_to_client(mnt->mnt_sb);
+       struct ceph_mount_args *args = &client->mount_args;
+
+       if (args->flags & CEPH_OPT_FSID)
+               seq_printf(m, ",fsidmajor=%llu,fsidminor%llu",
+                          le64_to_cpu(*(__le64 *)&args->fsid.fsid[0]),
+                          le64_to_cpu(*(__le64 *)&args->fsid.fsid[8]));
+       if (args->flags & CEPH_OPT_NOSHARE)
+               seq_puts(m, ",noshare");
+       if (args->flags & CEPH_OPT_DIRSTAT)
+               seq_puts(m, ",dirstat");
+       if ((args->flags & CEPH_OPT_RBYTES) == 0)
+               seq_puts(m, ",norbytes");
+       if (args->flags & CEPH_OPT_NOCRC)
+               seq_puts(m, ",nocrc");
+       if (args->flags & CEPH_OPT_NOASYNCREADDIR)
+               seq_puts(m, ",noasyncreaddir");
+       if (strcmp(args->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
+               seq_printf(m, ",snapdirname=%s", args->snapdir_name);
+       if (args->secret)
+               seq_puts(m, ",secret=<hidden>");
+       return 0;
+}
+
+/*
+ * caches
+ */
+struct kmem_cache *ceph_inode_cachep;
+struct kmem_cache *ceph_cap_cachep;
+struct kmem_cache *ceph_dentry_cachep;
+struct kmem_cache *ceph_file_cachep;
+
+static void ceph_inode_init_once(void *foo)
+{
+       struct ceph_inode_info *ci = foo;
+       inode_init_once(&ci->vfs_inode);
+}
+
+static int __init init_caches(void)
+{
+       ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
+                                     sizeof(struct ceph_inode_info),
+                                     __alignof__(struct ceph_inode_info),
+                                     (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
+                                     ceph_inode_init_once);
+       if (ceph_inode_cachep == NULL)
+               return -ENOMEM;
+
+       ceph_cap_cachep = KMEM_CACHE(ceph_cap,
+                                    SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
+       if (ceph_cap_cachep == NULL)
+               goto bad_cap;
+
+       ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
+                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
+       if (ceph_dentry_cachep == NULL)
+               goto bad_dentry;
+
+       ceph_file_cachep = KMEM_CACHE(ceph_file_info,
+                                     SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
+       if (ceph_file_cachep == NULL)
+               goto bad_file;
+
+       return 0;
+
+bad_file:
+       kmem_cache_destroy(ceph_dentry_cachep);
+bad_dentry:
+       kmem_cache_destroy(ceph_cap_cachep);
+bad_cap:
+       kmem_cache_destroy(ceph_inode_cachep);
+       return -ENOMEM;
+}
+
+static void destroy_caches(void)
+{
+       kmem_cache_destroy(ceph_inode_cachep);
+       kmem_cache_destroy(ceph_cap_cachep);
+       kmem_cache_destroy(ceph_dentry_cachep);
+       kmem_cache_destroy(ceph_file_cachep);
+}
+
+
+/*
+ * ceph_umount_begin - initiate forced umount.  Tear down down the
+ * mount, skipping steps that may hang while waiting for server(s).
+ */
+static void ceph_umount_begin(struct super_block *sb)
+{
+       struct ceph_client *client = ceph_sb_to_client(sb);
+
+       dout("ceph_umount_begin - starting forced umount\n");
+       if (!client)
+               return;
+       client->mount_state = CEPH_MOUNT_SHUTDOWN;
+       return;
+}
+
+static const struct super_operations ceph_super_ops = {
+       .alloc_inode    = ceph_alloc_inode,
+       .destroy_inode  = ceph_destroy_inode,
+       .write_inode    = ceph_write_inode,
+       .sync_fs        = ceph_syncfs,
+       .put_super      = ceph_put_super,
+       .show_options   = ceph_show_options,
+       .statfs         = ceph_statfs,
+       .umount_begin   = ceph_umount_begin,
+};
+
+
+const char *ceph_msg_type_name(int type)
+{
+       switch (type) {
+       case CEPH_MSG_SHUTDOWN: return "shutdown";
+       case CEPH_MSG_PING: return "ping";
+       case CEPH_MSG_MON_MAP: return "mon_map";
+       case CEPH_MSG_MON_GET_MAP: return "mon_get_map";
+       case CEPH_MSG_MON_SUBSCRIBE: return "mon_subscribe";
+       case CEPH_MSG_MON_SUBSCRIBE_ACK: return "mon_subscribe_ack";
+       case CEPH_MSG_CLIENT_MOUNT: return "client_mount";
+       case CEPH_MSG_CLIENT_MOUNT_ACK: return "client_mount_ack";
+       case CEPH_MSG_STATFS: return "statfs";
+       case CEPH_MSG_STATFS_REPLY: return "statfs_reply";
+       case CEPH_MSG_MDS_GETMAP: return "mds_getmap";
+       case CEPH_MSG_MDS_MAP: return "mds_map";
+       case CEPH_MSG_CLIENT_SESSION: return "client_session";
+       case CEPH_MSG_CLIENT_RECONNECT: return "client_reconnect";
+       case CEPH_MSG_CLIENT_REQUEST: return "client_request";
+       case CEPH_MSG_CLIENT_REQUEST_FORWARD: return "client_request_forward";
+       case CEPH_MSG_CLIENT_REPLY: return "client_reply";
+       case CEPH_MSG_CLIENT_CAPS: return "client_caps";
+       case CEPH_MSG_CLIENT_CAPRELEASE: return "client_cap_release";
+       case CEPH_MSG_CLIENT_SNAP: return "client_snap";
+       case CEPH_MSG_CLIENT_LEASE: return "client_lease";
+       case CEPH_MSG_OSD_GETMAP: return "osd_getmap";
+       case CEPH_MSG_OSD_MAP: return "osd_map";
+       case CEPH_MSG_OSD_OP: return "osd_op";
+       case CEPH_MSG_OSD_OPREPLY: return "osd_opreply";
+       default: return "unknown";
+       }
+}
+
+
+/*
+ * mount options
+ */
+enum {
+       Opt_fsidmajor,
+       Opt_fsidminor,
+       Opt_monport,
+       Opt_wsize,
+       Opt_rsize,
+       Opt_osdtimeout,
+       Opt_mount_timeout,
+       Opt_caps_wanted_delay_min,
+       Opt_caps_wanted_delay_max,
+       Opt_readdir_max_entries,
+       /* int args above */
+       Opt_snapdirname,
+       Opt_secret,
+       /* string args above */
+       Opt_ip,
+       Opt_noshare,
+       Opt_dirstat,
+       Opt_nodirstat,
+       Opt_rbytes,
+       Opt_norbytes,
+       Opt_nocrc,
+       Opt_noasyncreaddir,
+};
+
+static match_table_t arg_tokens = {
+       {Opt_fsidmajor, "fsidmajor=%ld"},
+       {Opt_fsidminor, "fsidminor=%ld"},
+       {Opt_monport, "monport=%d"},
+       {Opt_wsize, "wsize=%d"},
+       {Opt_rsize, "rsize=%d"},
+       {Opt_osdtimeout, "osdtimeout=%d"},
+       {Opt_mount_timeout, "mount_timeout=%d"},
+       {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
+       {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
+       {Opt_readdir_max_entries, "readdir_max_entries=%d"},
+       /* int args above */
+       {Opt_snapdirname, "snapdirname=%s"},
+       {Opt_secret, "secret=%s"},
+       /* string args above */
+       {Opt_ip, "ip=%s"},
+       {Opt_noshare, "noshare"},
+       {Opt_dirstat, "dirstat"},
+       {Opt_nodirstat, "nodirstat"},
+       {Opt_rbytes, "rbytes"},
+       {Opt_norbytes, "norbytes"},
+       {Opt_nocrc, "nocrc"},
+       {Opt_noasyncreaddir, "noasyncreaddir"},
+       {-1, NULL}
+};
+
+
+static int parse_mount_args(struct ceph_client *client,
+                           int flags, char *options, const char *dev_name,
+                           const char **path)
+{
+       struct ceph_mount_args *args = &client->mount_args;
+       const char *c;
+       int err;
+       substring_t argstr[MAX_OPT_ARGS];
+       int num_mon;
+       struct ceph_entity_addr mon_addr[CEPH_MAX_MON_MOUNT_ADDR];
+       int i;
+
+       dout("parse_mount_args dev_name '%s'\n", dev_name);
+       memset(args, 0, sizeof(*args));
+
+       /* start with defaults */
+       args->sb_flags = flags;
+       args->flags = CEPH_OPT_DEFAULT;
+       args->osd_timeout = 5;    /* seconds */
+       args->mount_timeout = CEPH_MOUNT_TIMEOUT_DEFAULT; /* seconds */
+       args->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
+       args->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
+       args->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
+       args->cap_release_safety = CEPH_CAPS_PER_RELEASE * 4;
+       args->max_readdir = 1024;
+
+       /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
+       if (!dev_name)
+               return -EINVAL;
+       *path = strstr(dev_name, ":/");
+       if (*path == NULL) {
+               pr_err("device name is missing path (no :/ in %s)\n",
+                      dev_name);
+               return -EINVAL;
+       }
+
+       /* get mon ip(s) */
+       err = ceph_parse_ips(dev_name, *path, mon_addr,
+                            CEPH_MAX_MON_MOUNT_ADDR, &num_mon);
+       if (err < 0)
+               return err;
+
+       /* build initial monmap */
+       client->monc.monmap = kzalloc(sizeof(*client->monc.monmap) +
+                              num_mon*sizeof(client->monc.monmap->mon_inst[0]),
+                              GFP_KERNEL);
+       if (!client->monc.monmap)
+               return -ENOMEM;
+       for (i = 0; i < num_mon; i++) {
+               client->monc.monmap->mon_inst[i].addr = mon_addr[i];
+               client->monc.monmap->mon_inst[i].addr.erank = 0;
+               client->monc.monmap->mon_inst[i].addr.nonce = 0;
+               client->monc.monmap->mon_inst[i].name.type =
+                       CEPH_ENTITY_TYPE_MON;
+               client->monc.monmap->mon_inst[i].name.num = cpu_to_le64(i);
+       }
+       client->monc.monmap->num_mon = num_mon;
+       memset(&args->my_addr.in_addr, 0, sizeof(args->my_addr.in_addr));
+
+       /* path on server */
+       *path += 2;
+       dout("server path '%s'\n", *path);
+
+       /* parse mount options */
+       while ((c = strsep(&options, ",")) != NULL) {
+               int token, intval, ret;
+               if (!*c)
+                       continue;
+               token = match_token((char *)c, arg_tokens, argstr);
+               if (token < 0) {
+                       pr_err("bad mount option at '%s'\n", c);
+                       return -EINVAL;
+
+               }
+               if (token < Opt_ip) {
+                       ret = match_int(&argstr[0], &intval);
+                       if (ret < 0) {
+                               pr_err("bad mount option arg (not int) "
+                                      "at '%s'\n", c);
+                               continue;
+                       }
+                       dout("got token %d intval %d\n", token, intval);
+               }
+               switch (token) {
+               case Opt_fsidmajor:
+                       *(__le64 *)&args->fsid.fsid[0] = cpu_to_le64(intval);
+                       break;
+               case Opt_fsidminor:
+                       *(__le64 *)&args->fsid.fsid[8] = cpu_to_le64(intval);
+                       break;
+               case Opt_ip:
+                       err = ceph_parse_ips(argstr[0].from,
+                                            argstr[0].to,
+                                            &args->my_addr,
+                                            1, NULL);
+                       if (err < 0)
+                               return err;
+                       args->flags |= CEPH_OPT_MYIP;
+                       break;
+
+               case Opt_snapdirname:
+                       kfree(args->snapdir_name);
+                       args->snapdir_name = kstrndup(argstr[0].from,
+                                             argstr[0].to-argstr[0].from,
+                                             GFP_KERNEL);
+                       break;
+               case Opt_secret:
+                       args->secret = kstrndup(argstr[0].from,
+                                               argstr[0].to-argstr[0].from,
+                                               GFP_KERNEL);
+                       break;
+
+                       /* misc */
+               case Opt_wsize:
+                       args->wsize = intval;
+                       break;
+               case Opt_rsize:
+                       args->rsize = intval;
+                       break;
+               case Opt_osdtimeout:
+                       args->osd_timeout = intval;
+                       break;
+               case Opt_mount_timeout:
+                       args->mount_timeout = intval;
+                       break;
+               case Opt_caps_wanted_delay_min:
+                       args->caps_wanted_delay_min = intval;
+                       break;
+               case Opt_caps_wanted_delay_max:
+                       args->caps_wanted_delay_max = intval;
+                       break;
+               case Opt_readdir_max_entries:
+                       args->max_readdir = intval;
+                       break;
+
+               case Opt_noshare:
+                       args->flags |= CEPH_OPT_NOSHARE;
+                       break;
+
+               case Opt_dirstat:
+                       args->flags |= CEPH_OPT_DIRSTAT;
+                       break;
+               case Opt_nodirstat:
+                       args->flags &= ~CEPH_OPT_DIRSTAT;
+                       break;
+               case Opt_rbytes:
+                       args->flags |= CEPH_OPT_RBYTES;
+                       break;
+               case Opt_norbytes:
+                       args->flags &= ~CEPH_OPT_RBYTES;
+                       break;
+               case Opt_nocrc:
+                       args->flags |= CEPH_OPT_NOCRC;
+                       break;
+               case Opt_noasyncreaddir:
+                       args->flags |= CEPH_OPT_NOASYNCREADDIR;
+                       break;
+
+               default:
+                       BUG_ON(token);
+               }
+       }
+
+       return 0;
+}
+
+static void release_mount_args(struct ceph_mount_args *args)
+{
+       kfree(args->snapdir_name);
+       args->snapdir_name = NULL;
+       kfree(args->secret);
+       args->secret = NULL;
+}
+
+/*
+ * create a fresh client instance
+ */
+static struct ceph_client *ceph_create_client(void)
+{
+       struct ceph_client *client;
+       int err = -ENOMEM;
+
+       client = kzalloc(sizeof(*client), GFP_KERNEL);
+       if (client == NULL)
+               return ERR_PTR(-ENOMEM);
+
+       mutex_init(&client->mount_mutex);
+
+       init_waitqueue_head(&client->mount_wq);
+
+       client->sb = NULL;
+       client->mount_state = CEPH_MOUNT_MOUNTING;
+       client->whoami = -1;
+
+       client->msgr = NULL;
+
+       client->mount_err = 0;
+       client->signed_ticket = NULL;
+       client->signed_ticket_len = 0;
+
+       err = -ENOMEM;
+       client->wb_wq = create_workqueue("ceph-writeback");
+       if (client->wb_wq == NULL)
+               goto fail;
+       client->pg_inv_wq = create_singlethread_workqueue("ceph-pg-invalid");
+       if (client->pg_inv_wq == NULL)
+               goto fail_wb_wq;
+       client->trunc_wq = create_singlethread_workqueue("ceph-trunc");
+       if (client->trunc_wq == NULL)
+               goto fail_pg_inv_wq;
+
+       /* subsystems */
+       err = ceph_monc_init(&client->monc, client);
+       if (err < 0)
+               goto fail_trunc_wq;
+       err = ceph_osdc_init(&client->osdc, client);
+       if (err < 0)
+               goto fail_monc;
+       ceph_mdsc_init(&client->mdsc, client);
+       return client;
+
+fail_monc:
+       ceph_monc_stop(&client->monc);
+fail_trunc_wq:
+       destroy_workqueue(client->trunc_wq);
+fail_pg_inv_wq:
+       destroy_workqueue(client->pg_inv_wq);
+fail_wb_wq:
+       destroy_workqueue(client->wb_wq);
+fail:
+       kfree(client);
+       return ERR_PTR(err);
+}
+
+static void ceph_destroy_client(struct ceph_client *client)
+{
+       dout("destroy_client %p\n", client);
+
+       /* unmount */
+       ceph_mdsc_stop(&client->mdsc);
+       ceph_monc_stop(&client->monc);
+       ceph_osdc_stop(&client->osdc);
+
+       kfree(client->signed_ticket);
+
+       ceph_debugfs_client_cleanup(client);
+       destroy_workqueue(client->wb_wq);
+       destroy_workqueue(client->pg_inv_wq);
+       destroy_workqueue(client->trunc_wq);
+
+       if (client->msgr)
+               ceph_messenger_destroy(client->msgr);
+       if (client->wb_pagevec_pool)
+               mempool_destroy(client->wb_pagevec_pool);
+
+       release_mount_args(&client->mount_args);
+
+       kfree(client);
+       dout("destroy_client %p done\n", client);
+}
+
+/*
+ * true if we have the mon map (and have thus joined the cluster)
+ */
+static int have_mon_map(struct ceph_client *client)
+{
+       return client->monc.monmap && client->monc.monmap->epoch;
+}
+
+/*
+ * Bootstrap mount by opening the root directory.  Note the mount
+ * @started time from caller, and time out if this takes too long.
+ */
+static struct dentry *open_root_dentry(struct ceph_client *client,
+                                      const char *path,
+                                      unsigned long started)
+{
+       struct ceph_mds_client *mdsc = &client->mdsc;
+       struct ceph_mds_request *req = NULL;
+       int err;
+       struct dentry *root;
+
+       /* open dir */
+       dout("open_root_inode opening '%s'\n", path);
+       req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
+       if (IS_ERR(req))
+               return ERR_PTR(PTR_ERR(req));
+       req->r_path1 = kstrdup(path, GFP_NOFS);
+       req->r_ino1.ino = CEPH_INO_ROOT;
+       req->r_ino1.snap = CEPH_NOSNAP;
+       req->r_started = started;
+       req->r_timeout = client->mount_args.mount_timeout * HZ;
+       req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
+       req->r_num_caps = 2;
+       err = ceph_mdsc_do_request(mdsc, NULL, req);
+       if (err == 0) {
+               dout("open_root_inode success\n");
+               if (ceph_ino(req->r_target_inode) == CEPH_INO_ROOT &&
+                   client->sb->s_root == NULL)
+                       root = d_alloc_root(req->r_target_inode);
+               else
+                       root = d_obtain_alias(req->r_target_inode);
+               req->r_target_inode = NULL;
+               dout("open_root_inode success, root dentry is %p\n", root);
+       } else {
+               root = ERR_PTR(err);
+       }
+       ceph_mdsc_put_request(req);
+       return root;
+}
+
+/*
+ * mount: join the ceph cluster, and open root directory.
+ */
+static int ceph_mount(struct ceph_client *client, struct vfsmount *mnt,
+                     const char *path)
+{
+       struct ceph_entity_addr *myaddr = NULL;
+       int err;
+       unsigned long timeout = client->mount_args.mount_timeout * HZ;
+       unsigned long started = jiffies;  /* note the start time */
+       struct dentry *root;
+
+       dout("mount start\n");
+       mutex_lock(&client->mount_mutex);
+
+       /* initialize the messenger */
+       if (client->msgr == NULL) {
+               if (ceph_test_opt(client, MYIP))
+                       myaddr = &client->mount_args.my_addr;
+               client->msgr = ceph_messenger_create(myaddr);
+               if (IS_ERR(client->msgr)) {
+                       err = PTR_ERR(client->msgr);
+                       client->msgr = NULL;
+                       goto out;
+               }
+               client->msgr->nocrc = ceph_test_opt(client, NOCRC);
+       }
+
+       /* send mount request, and wait for mon, mds, and osd maps */
+       err = ceph_monc_request_mount(&client->monc);
+       if (err < 0)
+               goto out;
+
+       while (!have_mon_map(client) && !client->mount_err) {
+               err = -EIO;
+               if (timeout && time_after_eq(jiffies, started + timeout))
+                       goto out;
+
+               /* wait */
+               dout("mount waiting for mount\n");
+               err = wait_event_interruptible_timeout(client->mount_wq,
+                              client->mount_err || have_mon_map(client),
+                              timeout);
+               if (err == -EINTR || err == -ERESTARTSYS)
+                       goto out;
+               if (client->mount_err) {
+                       err = client->mount_err;
+                       goto out;
+               }
+       }
+
+       dout("mount opening root\n");
+       root = open_root_dentry(client, "", started);
+       if (IS_ERR(root)) {
+               err = PTR_ERR(root);
+               goto out;
+       }
+       if (client->sb->s_root)
+               dput(root);
+       else
+               client->sb->s_root = root;
+
+       if (path[0] == 0) {
+               dget(root);
+       } else {
+               dout("mount opening base mountpoint\n");
+               root = open_root_dentry(client, path, started);
+               if (IS_ERR(root)) {
+                       err = PTR_ERR(root);
+                       dput(client->sb->s_root);
+                       client->sb->s_root = NULL;
+                       goto out;
+               }
+       }
+
+       mnt->mnt_root = root;
+       mnt->mnt_sb = client->sb;
+
+       client->mount_state = CEPH_MOUNT_MOUNTED;
+       dout("mount success\n");
+       err = 0;
+
+out:
+       mutex_unlock(&client->mount_mutex);
+       return err;
+}
+
+static int ceph_set_super(struct super_block *s, void *data)
+{
+       struct ceph_client *client = data;
+       int ret;
+
+       dout("set_super %p data %p\n", s, data);
+
+       s->s_flags = client->mount_args.sb_flags;
+       s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
+
+       s->s_fs_info = client;
+       client->sb = s;
+
+       s->s_op = &ceph_super_ops;
+       s->s_export_op = &ceph_export_ops;
+
+       s->s_time_gran = 1000;  /* 1000 ns == 1 us */
+
+       ret = set_anon_super(s, NULL);  /* what is that second arg for? */
+       if (ret != 0)
+               goto fail;
+
+       return ret;
+
+fail:
+       s->s_fs_info = NULL;
+       client->sb = NULL;
+       return ret;
+}
+
+/*
+ * share superblock if same fs AND options
+ */
+static int ceph_compare_super(struct super_block *sb, void *data)
+{
+       struct ceph_client *new = data;
+       struct ceph_mount_args *args = &new->mount_args;
+       struct ceph_client *other = ceph_sb_to_client(sb);
+       int i;
+
+       dout("ceph_compare_super %p\n", sb);
+       if (args->flags & CEPH_OPT_FSID) {
+               if (ceph_fsid_compare(&args->fsid, &other->fsid)) {
+                       dout("fsid doesn't match\n");
+                       return 0;
+               }
+       } else {
+               /* do we share (a) monitor? */
+               for (i = 0; i < new->monc.monmap->num_mon; i++)
+                       if (ceph_monmap_contains(other->monc.monmap,
+                                        &new->monc.monmap->mon_inst[i].addr))
+                               break;
+               if (i == new->monc.monmap->num_mon) {
+                       dout("mon ip not part of monmap\n");
+                       return 0;
+               }
+               dout("mon ip matches existing sb %p\n", sb);
+       }
+       if (args->sb_flags != other->mount_args.sb_flags) {
+               dout("flags differ\n");
+               return 0;
+       }
+       return 1;
+}
+
+/*
+ * construct our own bdi so we can control readahead, etc.
+ */
+static int ceph_init_bdi(struct super_block *sb, struct ceph_client *client)
+{
+       int err;
+
+       err = bdi_init(&client->backing_dev_info);
+       if (err < 0)
+               return err;
+
+       /* set ra_pages based on rsize mount option? */
+       if (client->mount_args.rsize >= PAGE_CACHE_SIZE)
+               client->backing_dev_info.ra_pages =
+                       (client->mount_args.rsize + PAGE_CACHE_SIZE - 1)
+                       >> PAGE_SHIFT;
+
+       err = bdi_register_dev(&client->backing_dev_info, sb->s_dev);
+       return err;
+}
+
+static int ceph_get_sb(struct file_system_type *fs_type,
+                      int flags, const char *dev_name, void *data,
+                      struct vfsmount *mnt)
+{
+       struct super_block *sb;
+       struct ceph_client *client;
+       int err;
+       int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
+       const char *path;
+
+       dout("ceph_get_sb\n");
+
+       /* create client (which we may/may not use) */
+       client = ceph_create_client();
+       if (IS_ERR(client))
+               return PTR_ERR(client);
+
+       err = parse_mount_args(client, flags, data, dev_name, &path);
+       if (err < 0)
+               goto out;
+
+       if (client->mount_args.flags & CEPH_OPT_NOSHARE)
+               compare_super = NULL;
+       sb = sget(fs_type, compare_super, ceph_set_super, client);
+       if (IS_ERR(sb)) {
+               err = PTR_ERR(sb);
+               goto out;
+       }
+
+       if (ceph_client(sb) != client) {
+               ceph_destroy_client(client);
+               client = ceph_client(sb);
+               dout("get_sb got existing client %p\n", client);
+       } else {
+               dout("get_sb using new client %p\n", client);
+
+               /* set up mempools */
+               err = -ENOMEM;
+               client->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
+                             client->mount_args.wsize >> PAGE_CACHE_SHIFT);
+               if (!client->wb_pagevec_pool)
+                       goto out_splat;
+
+               err = ceph_init_bdi(sb, client);
+               if (err < 0)
+                       goto out_splat;
+       }
+
+       err = ceph_mount(client, mnt, path);
+       if (err < 0)
+               goto out_splat;
+       dout("root %p inode %p ino %llx.%llx\n", mnt->mnt_root,
+            mnt->mnt_root->d_inode, ceph_vinop(mnt->mnt_root->d_inode));
+       return 0;
+
+out_splat:
+       ceph_mdsc_close_sessions(&client->mdsc);
+       up_write(&sb->s_umount);
+       deactivate_super(sb);
+       goto out_final;
+
+out:
+       ceph_destroy_client(client);
+out_final:
+       dout("ceph_get_sb fail %d\n", err);
+       return err;
+}
+
+static void ceph_kill_sb(struct super_block *s)
+{
+       struct ceph_client *client = ceph_sb_to_client(s);
+       dout("kill_sb %p\n", s);
+       ceph_mdsc_pre_umount(&client->mdsc);
+       bdi_unregister(&client->backing_dev_info);
+       kill_anon_super(s);    /* will call put_super after sb is r/o */
+       bdi_destroy(&client->backing_dev_info);
+       ceph_destroy_client(client);
+}
+
+static struct file_system_type ceph_fs_type = {
+       .owner          = THIS_MODULE,
+       .name           = "ceph",
+       .get_sb         = ceph_get_sb,
+       .kill_sb        = ceph_kill_sb,
+       .fs_flags       = FS_RENAME_DOES_D_MOVE,
+};
+
+#define _STRINGIFY(x) #x
+#define STRINGIFY(x) _STRINGIFY(x)
+
+static int __init init_ceph(void)
+{
+       int ret = 0;
+
+       ret = ceph_debugfs_init();
+       if (ret < 0)
+               goto out;
+
+       ret = ceph_msgr_init();
+       if (ret < 0)
+               goto out_debugfs;
+
+       ret = init_caches();
+       if (ret)
+               goto out_msgr;
+
+       ceph_caps_init();
+
+       ret = register_filesystem(&ceph_fs_type);
+       if (ret)
+               goto out_icache;
+
+       pr_info("loaded (%s)\n", STRINGIFY(CEPH_GIT_VER));
+       return 0;
+
+out_icache:
+       destroy_caches();
+out_msgr:
+       ceph_msgr_exit();
+out_debugfs:
+       ceph_debugfs_cleanup();
+out:
+       return ret;
+}
+
+static void __exit exit_ceph(void)
+{
+       dout("exit_ceph\n");
+       unregister_filesystem(&ceph_fs_type);
+       ceph_caps_finalize();
+       destroy_caches();
+       ceph_msgr_exit();
+       ceph_debugfs_cleanup();
+}
+
+module_init(init_ceph);
+module_exit(exit_ceph);
+
+MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
+MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
+MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
+MODULE_DESCRIPTION("Ceph filesystem for Linux");
+MODULE_LICENSE("GPL");