add hlist_bl_lock/unlock helpers
[linux-2.6.git] / fs / gfs2 / ops_fstype.c
index 52fb6c0..d3c69eb 100644 (file)
@@ -17,7 +17,7 @@
 #include <linux/namei.h>
 #include <linux/mount.h>
 #include <linux/gfs2_ondisk.h>
-#include <linux/slow-work.h>
+#include <linux/quotaops.h>
 
 #include "gfs2.h"
 #include "incore.h"
 #define DO 0
 #define UNDO 1
 
-static const u32 gfs2_old_fs_formats[] = {
-        0
-};
-
-static const u32 gfs2_old_multihost_formats[] = {
-        0
-};
-
 /**
  * gfs2_tune_init - Fill a gfs2_tune structure with default values
  * @gt: tune
@@ -56,19 +48,13 @@ static void gfs2_tune_init(struct gfs2_tune *gt)
 {
        spin_lock_init(&gt->gt_spin);
 
-       gt->gt_incore_log_blocks = 1024;
-       gt->gt_logd_secs = 1;
        gt->gt_quota_simul_sync = 64;
        gt->gt_quota_warn_period = 10;
        gt->gt_quota_scale_num = 1;
        gt->gt_quota_scale_den = 1;
-       gt->gt_quota_quantum = 60;
        gt->gt_new_files_jdata = 0;
        gt->gt_max_readahead = 1 << 18;
-       gt->gt_stall_secs = 600;
        gt->gt_complain_secs = 10;
-       gt->gt_statfs_quantum = 30;
-       gt->gt_statfs_slow = 0;
 }
 
 static struct gfs2_sbd *init_sbd(struct super_block *sb)
@@ -81,9 +67,11 @@ static struct gfs2_sbd *init_sbd(struct super_block *sb)
 
        sb->s_fs_info = sdp;
        sdp->sd_vfs = sb;
-
+       set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
        gfs2_tune_init(&sdp->sd_tune);
 
+       init_waitqueue_head(&sdp->sd_glock_wait);
+       atomic_set(&sdp->sd_glock_disposal, 0);
        spin_lock_init(&sdp->sd_statfs_spin);
 
        spin_lock_init(&sdp->sd_rindex_spin);
@@ -102,14 +90,16 @@ static struct gfs2_sbd *init_sbd(struct super_block *sb)
        spin_lock_init(&sdp->sd_trunc_lock);
 
        spin_lock_init(&sdp->sd_log_lock);
-
+       atomic_set(&sdp->sd_log_pinned, 0);
        INIT_LIST_HEAD(&sdp->sd_log_le_buf);
        INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
        INIT_LIST_HEAD(&sdp->sd_log_le_rg);
        INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
        INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
 
-       mutex_init(&sdp->sd_log_reserve_mutex);
+       init_waitqueue_head(&sdp->sd_log_waitq);
+       init_waitqueue_head(&sdp->sd_logd_waitq);
+       spin_lock_init(&sdp->sd_ail_lock);
        INIT_LIST_HEAD(&sdp->sd_ail1_list);
        INIT_LIST_HEAD(&sdp->sd_ail2_list);
 
@@ -138,8 +128,6 @@ static struct gfs2_sbd *init_sbd(struct super_block *sb)
 
 static int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
 {
-       unsigned int x;
-
        if (sb->sb_magic != GFS2_MAGIC ||
            sb->sb_type != GFS2_METATYPE_SB) {
                if (!silent)
@@ -153,55 +141,9 @@ static int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int sile
            sb->sb_multihost_format == GFS2_FORMAT_MULTI)
                return 0;
 
-       if (sb->sb_fs_format != GFS2_FORMAT_FS) {
-               for (x = 0; gfs2_old_fs_formats[x]; x++)
-                       if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
-                               break;
-
-               if (!gfs2_old_fs_formats[x]) {
-                       printk(KERN_WARNING
-                              "GFS2: code version (%u, %u) is incompatible "
-                              "with ondisk format (%u, %u)\n",
-                              GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
-                              sb->sb_fs_format, sb->sb_multihost_format);
-                       printk(KERN_WARNING
-                              "GFS2: I don't know how to upgrade this FS\n");
-                       return -EINVAL;
-               }
-       }
-
-       if (sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
-               for (x = 0; gfs2_old_multihost_formats[x]; x++)
-                       if (gfs2_old_multihost_formats[x] ==
-                           sb->sb_multihost_format)
-                               break;
-
-               if (!gfs2_old_multihost_formats[x]) {
-                       printk(KERN_WARNING
-                              "GFS2: code version (%u, %u) is incompatible "
-                              "with ondisk format (%u, %u)\n",
-                              GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
-                              sb->sb_fs_format, sb->sb_multihost_format);
-                       printk(KERN_WARNING
-                              "GFS2: I don't know how to upgrade this FS\n");
-                       return -EINVAL;
-               }
-       }
-
-       if (!sdp->sd_args.ar_upgrade) {
-               printk(KERN_WARNING
-                      "GFS2: code version (%u, %u) is incompatible "
-                      "with ondisk format (%u, %u)\n",
-                      GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
-                      sb->sb_fs_format, sb->sb_multihost_format);
-               printk(KERN_INFO
-                      "GFS2: Use the \"upgrade\" mount option to upgrade "
-                      "the FS\n");
-               printk(KERN_INFO "GFS2: See the manual for more details\n");
-               return -EINVAL;
-       }
+       fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
 
-       return 0;
+       return -EINVAL;
 }
 
 static void end_bio_io_page(struct bio *bio, int error)
@@ -277,7 +219,7 @@ static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector)
 
        bio->bi_end_io = end_bio_io_page;
        bio->bi_private = page;
-       submit_bio(READ_SYNC | (1 << BIO_RW_META), bio);
+       submit_bio(READ_SYNC | REQ_META, bio);
        wait_on_page_locked(page);
        bio_put(bio);
        if (!PageUptodate(page)) {
@@ -499,7 +441,6 @@ static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
                iput(inode);
                return -ENOMEM;
        }
-       dentry->d_op = &gfs2_dops;
        *dptr = dentry;
        return 0;
 }
@@ -589,7 +530,7 @@ static int map_journal_extents(struct gfs2_sbd *sdp)
 
        prev_db = 0;
 
-       for (lb = 0; lb < ip->i_disksize >> sdp->sd_sb.sb_bsize_shift; lb++) {
+       for (lb = 0; lb < i_size_read(jd->jd_inode) >> sdp->sd_sb.sb_bsize_shift; lb++) {
                bh.b_state = 0;
                bh.b_blocknr = 0;
                bh.b_size = 1 << ip->i_inode.i_blkbits;
@@ -675,7 +616,7 @@ static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
                        break;
 
                INIT_LIST_HEAD(&jd->extent_list);
-               slow_work_init(&jd->jd_work, &gfs2_recover_ops);
+               INIT_WORK(&jd->jd_work, gfs2_recover_func);
                jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
                if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
                        if (!jd->jd_inode)
@@ -725,7 +666,7 @@ static int init_journal(struct gfs2_sbd *sdp, int undo)
                goto fail;
        }
 
-       error = -EINVAL;
+       error = -EUSERS;
        if (!gfs2_jindex_size(sdp)) {
                fs_err(sdp, "no journals!\n");
                goto fail_jindex;
@@ -734,6 +675,8 @@ static int init_journal(struct gfs2_sbd *sdp, int undo)
        if (sdp->sd_args.ar_spectator) {
                sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
                atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
+               atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
+               atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
        } else {
                if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
                        fs_err(sdp, "can't mount journal #%u\n",
@@ -771,6 +714,8 @@ static int init_journal(struct gfs2_sbd *sdp, int undo)
                        goto fail_jinode_gh;
                }
                atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
+               atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
+               atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 
                /* Map the extents for this journal's blocks */
                map_journal_extents(sdp);
@@ -780,7 +725,8 @@ static int init_journal(struct gfs2_sbd *sdp, int undo)
        if (sdp->sd_lockstruct.ls_first) {
                unsigned int x;
                for (x = 0; x < sdp->sd_journals; x++) {
-                       error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x));
+                       error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x),
+                                                    true);
                        if (error) {
                                fs_err(sdp, "error recovering journal %u: %d\n",
                                       x, error);
@@ -790,7 +736,7 @@ static int init_journal(struct gfs2_sbd *sdp, int undo)
 
                gfs2_others_may_mount(sdp);
        } else if (!sdp->sd_args.ar_spectator) {
-               error = gfs2_recover_journal(sdp->sd_jdesc);
+               error = gfs2_recover_journal(sdp->sd_jdesc, true);
                if (error) {
                        fs_err(sdp, "error recovering my journal: %d\n", error);
                        goto fail_jinode_gh;
@@ -952,8 +898,6 @@ static int init_threads(struct gfs2_sbd *sdp, int undo)
        if (undo)
                goto fail_quotad;
 
-       sdp->sd_log_flush_time = jiffies;
-
        p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
        error = IS_ERR(p);
        if (error) {
@@ -987,14 +931,14 @@ static const match_table_t nolock_tokens = {
 
 static const struct lm_lockops nolock_ops = {
        .lm_proto_name = "lock_nolock",
-       .lm_put_lock = kmem_cache_free,
+       .lm_put_lock = gfs2_glock_free,
        .lm_tokens = &nolock_tokens,
 };
 
 /**
  * gfs2_lm_mount - mount a locking protocol
  * @sdp: the filesystem
- * @args: mount arguements
+ * @args: mount arguments
  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
  *
  * Returns: errno
@@ -1014,7 +958,6 @@ static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
        if (!strcmp("lock_nolock", proto)) {
                lm = &nolock_ops;
                sdp->sd_args.ar_localflocks = 1;
-               sdp->sd_args.ar_localcaching = 1;
 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
        } else if (!strcmp("lock_dlm", proto)) {
                lm = &gfs2_dlm_ops;
@@ -1042,7 +985,8 @@ static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
                        ret = match_int(&tmp[0], &option);
                        if (ret || option < 0) 
                                goto hostdata_error;
-                       ls->ls_jid = option;
+                       if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
+                               ls->ls_jid = option;
                        break;
                case Opt_id:
                        /* Obsolete, but left for backward compat purposes */
@@ -1094,6 +1038,22 @@ void gfs2_lm_unmount(struct gfs2_sbd *sdp)
                lm->lm_unmount(sdp);
 }
 
+static int gfs2_journalid_wait(void *word)
+{
+       if (signal_pending(current))
+               return -EINTR;
+       schedule();
+       return 0;
+}
+
+static int wait_on_journal(struct gfs2_sbd *sdp)
+{
+       if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
+               return 0;
+
+       return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
+}
+
 void gfs2_online_uevent(struct gfs2_sbd *sdp)
 {
        struct super_block *sb = sdp->sd_vfs;
@@ -1114,7 +1074,7 @@ void gfs2_online_uevent(struct gfs2_sbd *sdp)
  * Returns: errno
  */
 
-static int fill_super(struct super_block *sb, void *data, int silent)
+static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
 {
        struct gfs2_sbd *sdp;
        struct gfs2_holder mount_gh;
@@ -1125,17 +1085,7 @@ static int fill_super(struct super_block *sb, void *data, int silent)
                printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
                return -ENOMEM;
        }
-
-       sdp->sd_args.ar_quota = GFS2_QUOTA_DEFAULT;
-       sdp->sd_args.ar_data = GFS2_DATA_DEFAULT;
-       sdp->sd_args.ar_commit = 60;
-       sdp->sd_args.ar_errors = GFS2_ERRORS_DEFAULT;
-
-       error = gfs2_mount_args(sdp, &sdp->sd_args, data);
-       if (error) {
-               printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
-               goto fail;
-       }
+       sdp->sd_args = *args;
 
        if (sdp->sd_args.ar_spectator) {
                 sb->s_flags |= MS_RDONLY;
@@ -1143,11 +1093,16 @@ static int fill_super(struct super_block *sb, void *data, int silent)
        }
        if (sdp->sd_args.ar_posix_acl)
                sb->s_flags |= MS_POSIXACL;
+       if (sdp->sd_args.ar_nobarrier)
+               set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
 
        sb->s_magic = GFS2_MAGIC;
        sb->s_op = &gfs2_super_ops;
+       sb->s_d_op = &gfs2_dops;
        sb->s_export_op = &gfs2_export_ops;
        sb->s_xattr = gfs2_xattr_handlers;
+       sb->s_qcop = &gfs2_quotactl_ops;
+       sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
        sb->s_time_gran = 1;
        sb->s_maxbytes = MAX_LFS_FILESIZE;
 
@@ -1159,7 +1114,16 @@ static int fill_super(struct super_block *sb, void *data, int silent)
                                GFS2_BASIC_BLOCK_SHIFT;
        sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 
-       sdp->sd_tune.gt_log_flush_secs = sdp->sd_args.ar_commit;
+       sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
+       sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
+       if (sdp->sd_args.ar_statfs_quantum) {
+               sdp->sd_tune.gt_statfs_slow = 0;
+               sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
+       }
+       else {
+               sdp->sd_tune.gt_statfs_slow = 1;
+               sdp->sd_tune.gt_statfs_quantum = 30;
+       }
 
        error = init_names(sdp, silent);
        if (error)
@@ -1183,6 +1147,24 @@ static int fill_super(struct super_block *sb, void *data, int silent)
        if (error)
                goto fail_locking;
 
+       error = wait_on_journal(sdp);
+       if (error)
+               goto fail_sb;
+
+       /*
+        * If user space has failed to join the cluster or some similar
+        * failure has occurred, then the journal id will contain a
+        * negative (error) number. This will then be returned to the
+        * caller (of the mount syscall). We do this even for spectator
+        * mounts (which just write a jid of 0 to indicate "ok" even though
+        * the jid is unused in the spectator case)
+        */
+       if (sdp->sd_lockstruct.ls_jid < 0) {
+               error = sdp->sd_lockstruct.ls_jid;
+               sdp->sd_lockstruct.ls_jid = 0;
+               goto fail_sb;
+       }
+
        error = init_inodes(sdp, DO);
        if (error)
                goto fail_sb;
@@ -1232,8 +1214,6 @@ fail_locking:
 fail_lm:
        gfs2_gl_hash_clear(sdp);
        gfs2_lm_unmount(sdp);
-       while (invalidate_inodes(sb))
-               yield();
 fail_sys:
        gfs2_sys_fs_del(sdp);
 fail:
@@ -1243,25 +1223,129 @@ fail:
        return error;
 }
 
-static int gfs2_get_sb(struct file_system_type *fs_type, int flags,
-                      const char *dev_name, void *data, struct vfsmount *mnt)
+static int set_gfs2_super(struct super_block *s, void *data)
 {
-       return get_sb_bdev(fs_type, flags, dev_name, data, fill_super, mnt);
+       s->s_bdev = data;
+       s->s_dev = s->s_bdev->bd_dev;
+
+       /*
+        * We set the bdi here to the queue backing, file systems can
+        * overwrite this in ->fill_super()
+        */
+       s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
+       return 0;
 }
 
-static int test_meta_super(struct super_block *s, void *ptr)
+static int test_gfs2_super(struct super_block *s, void *ptr)
 {
        struct block_device *bdev = ptr;
        return (bdev == s->s_bdev);
 }
 
+/**
+ * gfs2_mount - Get the GFS2 superblock
+ * @fs_type: The GFS2 filesystem type
+ * @flags: Mount flags
+ * @dev_name: The name of the device
+ * @data: The mount arguments
+ *
+ * Q. Why not use get_sb_bdev() ?
+ * A. We need to select one of two root directories to mount, independent
+ *    of whether this is the initial, or subsequent, mount of this sb
+ *
+ * Returns: 0 or -ve on error
+ */
+
+static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
+                      const char *dev_name, void *data)
+{
+       struct block_device *bdev;
+       struct super_block *s;
+       fmode_t mode = FMODE_READ | FMODE_EXCL;
+       int error;
+       struct gfs2_args args;
+       struct gfs2_sbd *sdp;
+
+       if (!(flags & MS_RDONLY))
+               mode |= FMODE_WRITE;
+
+       bdev = blkdev_get_by_path(dev_name, mode, fs_type);
+       if (IS_ERR(bdev))
+               return ERR_CAST(bdev);
+
+       /*
+        * once the super is inserted into the list by sget, s_umount
+        * will protect the lockfs code from trying to start a snapshot
+        * while we are mounting
+        */
+       mutex_lock(&bdev->bd_fsfreeze_mutex);
+       if (bdev->bd_fsfreeze_count > 0) {
+               mutex_unlock(&bdev->bd_fsfreeze_mutex);
+               error = -EBUSY;
+               goto error_bdev;
+       }
+       s = sget(fs_type, test_gfs2_super, set_gfs2_super, bdev);
+       mutex_unlock(&bdev->bd_fsfreeze_mutex);
+       error = PTR_ERR(s);
+       if (IS_ERR(s))
+               goto error_bdev;
+
+       if (s->s_root)
+               blkdev_put(bdev, mode);
+
+       memset(&args, 0, sizeof(args));
+       args.ar_quota = GFS2_QUOTA_DEFAULT;
+       args.ar_data = GFS2_DATA_DEFAULT;
+       args.ar_commit = 30;
+       args.ar_statfs_quantum = 30;
+       args.ar_quota_quantum = 60;
+       args.ar_errors = GFS2_ERRORS_DEFAULT;
+
+       error = gfs2_mount_args(&args, data);
+       if (error) {
+               printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
+               goto error_super;
+       }
+
+       if (s->s_root) {
+               error = -EBUSY;
+               if ((flags ^ s->s_flags) & MS_RDONLY)
+                       goto error_super;
+       } else {
+               char b[BDEVNAME_SIZE];
+
+               s->s_flags = flags;
+               s->s_mode = mode;
+               strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
+               sb_set_blocksize(s, block_size(bdev));
+               error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
+               if (error)
+                       goto error_super;
+               s->s_flags |= MS_ACTIVE;
+               bdev->bd_super = s;
+       }
+
+       sdp = s->s_fs_info;
+       if (args.ar_meta)
+               return dget(sdp->sd_master_dir);
+       else
+               return dget(sdp->sd_root_dir);
+
+error_super:
+       deactivate_locked_super(s);
+       return ERR_PTR(error);
+error_bdev:
+       blkdev_put(bdev, mode);
+       return ERR_PTR(error);
+}
+
 static int set_meta_super(struct super_block *s, void *ptr)
 {
        return -EINVAL;
 }
 
-static int gfs2_get_sb_meta(struct file_system_type *fs_type, int flags,
-                           const char *dev_name, void *data, struct vfsmount *mnt)
+static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
+                       int flags, const char *dev_name, void *data)
 {
        struct super_block *s;
        struct gfs2_sbd *sdp;
@@ -1272,19 +1356,21 @@ static int gfs2_get_sb_meta(struct file_system_type *fs_type, int flags,
        if (error) {
                printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
                       dev_name, error);
-               return error;
+               return ERR_PTR(error);
        }
-       s = sget(&gfs2_fs_type, test_meta_super, set_meta_super,
+       s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super,
                 path.dentry->d_inode->i_sb->s_bdev);
        path_put(&path);
        if (IS_ERR(s)) {
                printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
-               return PTR_ERR(s);
+               return ERR_CAST(s);
+       }
+       if ((flags ^ s->s_flags) & MS_RDONLY) {
+               deactivate_locked_super(s);
+               return ERR_PTR(-EBUSY);
        }
        sdp = s->s_fs_info;
-       mnt->mnt_sb = s;
-       mnt->mnt_root = dget(sdp->sd_master_dir);
-       return 0;
+       return dget(sdp->sd_master_dir);
 }
 
 static void gfs2_kill_sb(struct super_block *sb)
@@ -1310,7 +1396,7 @@ static void gfs2_kill_sb(struct super_block *sb)
 struct file_system_type gfs2_fs_type = {
        .name = "gfs2",
        .fs_flags = FS_REQUIRES_DEV,
-       .get_sb = gfs2_get_sb,
+       .mount = gfs2_mount,
        .kill_sb = gfs2_kill_sb,
        .owner = THIS_MODULE,
 };
@@ -1318,7 +1404,7 @@ struct file_system_type gfs2_fs_type = {
 struct file_system_type gfs2meta_fs_type = {
        .name = "gfs2meta",
        .fs_flags = FS_REQUIRES_DEV,
-       .get_sb = gfs2_get_sb_meta,
+       .mount = gfs2_mount_meta,
        .owner = THIS_MODULE,
 };