nilfs2: do not return io error for bio allocation failure
[safe/jmp/linux-2.6] / fs / nilfs2 / the_nilfs.c
index 661ab76..6241e17 100644 (file)
 #include <linux/slab.h>
 #include <linux/blkdev.h>
 #include <linux/backing-dev.h>
+#include <linux/crc32.h>
 #include "nilfs.h"
 #include "segment.h"
 #include "alloc.h"
 #include "cpfile.h"
 #include "sufile.h"
 #include "dat.h"
-#include "seglist.h"
 #include "segbuf.h"
 
+
+static LIST_HEAD(nilfs_objects);
+static DEFINE_SPINLOCK(nilfs_lock);
+
 void nilfs_set_last_segment(struct the_nilfs *nilfs,
                            sector_t start_blocknr, u64 seq, __u64 cno)
 {
@@ -54,7 +58,7 @@ void nilfs_set_last_segment(struct the_nilfs *nilfs,
  * Return Value: On success, pointer to the_nilfs is returned.
  * On error, NULL is returned.
  */
-struct the_nilfs *alloc_nilfs(struct block_device *bdev)
+static struct the_nilfs *alloc_nilfs(struct block_device *bdev)
 {
        struct the_nilfs *nilfs;
 
@@ -64,10 +68,12 @@ struct the_nilfs *alloc_nilfs(struct block_device *bdev)
 
        nilfs->ns_bdev = bdev;
        atomic_set(&nilfs->ns_count, 1);
-       atomic_set(&nilfs->ns_writer_refcount, -1);
        atomic_set(&nilfs->ns_ndirtyblks, 0);
        init_rwsem(&nilfs->ns_sem);
-       mutex_init(&nilfs->ns_writer_mutex);
+       init_rwsem(&nilfs->ns_super_sem);
+       mutex_init(&nilfs->ns_mount_mutex);
+       init_rwsem(&nilfs->ns_writer_sem);
+       INIT_LIST_HEAD(&nilfs->ns_list);
        INIT_LIST_HEAD(&nilfs->ns_supers);
        spin_lock_init(&nilfs->ns_last_segment_lock);
        nilfs->ns_gc_inodes_h = NULL;
@@ -77,6 +83,45 @@ struct the_nilfs *alloc_nilfs(struct block_device *bdev)
 }
 
 /**
+ * find_or_create_nilfs - find or create nilfs object
+ * @bdev: block device to which the_nilfs is related
+ *
+ * find_nilfs() looks up an existent nilfs object created on the
+ * device and gets the reference count of the object.  If no nilfs object
+ * is found on the device, a new nilfs object is allocated.
+ *
+ * Return Value: On success, pointer to the nilfs object is returned.
+ * On error, NULL is returned.
+ */
+struct the_nilfs *find_or_create_nilfs(struct block_device *bdev)
+{
+       struct the_nilfs *nilfs, *new = NULL;
+
+ retry:
+       spin_lock(&nilfs_lock);
+       list_for_each_entry(nilfs, &nilfs_objects, ns_list) {
+               if (nilfs->ns_bdev == bdev) {
+                       get_nilfs(nilfs);
+                       spin_unlock(&nilfs_lock);
+                       if (new)
+                               put_nilfs(new);
+                       return nilfs; /* existing object */
+               }
+       }
+       if (new) {
+               list_add_tail(&new->ns_list, &nilfs_objects);
+               spin_unlock(&nilfs_lock);
+               return new; /* new object */
+       }
+       spin_unlock(&nilfs_lock);
+
+       new = alloc_nilfs(bdev);
+       if (new)
+               goto retry;
+       return NULL; /* insufficient memory */
+}
+
+/**
  * put_nilfs - release a reference to the_nilfs
  * @nilfs: the_nilfs structure to be released
  *
@@ -85,27 +130,31 @@ struct the_nilfs *alloc_nilfs(struct block_device *bdev)
  */
 void put_nilfs(struct the_nilfs *nilfs)
 {
-       if (!atomic_dec_and_test(&nilfs->ns_count))
+       spin_lock(&nilfs_lock);
+       if (!atomic_dec_and_test(&nilfs->ns_count)) {
+               spin_unlock(&nilfs_lock);
                return;
+       }
+       list_del_init(&nilfs->ns_list);
+       spin_unlock(&nilfs_lock);
+
        /*
-        * Increment of ns_count never occur below because the caller
+        * Increment of ns_count never occurs below because the caller
         * of get_nilfs() holds at least one reference to the_nilfs.
         * Thus its exclusion control is not required here.
         */
+
        might_sleep();
        if (nilfs_loaded(nilfs)) {
-               nilfs_mdt_clear(nilfs->ns_sufile);
                nilfs_mdt_destroy(nilfs->ns_sufile);
-               nilfs_mdt_clear(nilfs->ns_cpfile);
                nilfs_mdt_destroy(nilfs->ns_cpfile);
-               nilfs_mdt_clear(nilfs->ns_dat);
                nilfs_mdt_destroy(nilfs->ns_dat);
-               /* XXX: how and when to clear nilfs->ns_gc_dat? */
                nilfs_mdt_destroy(nilfs->ns_gc_dat);
        }
        if (nilfs_init(nilfs)) {
                nilfs_destroy_gccache(nilfs);
-               brelse(nilfs->ns_sbh);
+               brelse(nilfs->ns_sbh[0]);
+               brelse(nilfs->ns_sbh[1]);
        }
        kfree(nilfs);
 }
@@ -115,6 +164,7 @@ static int nilfs_load_super_root(struct the_nilfs *nilfs,
 {
        struct buffer_head *bh_sr;
        struct nilfs_super_root *raw_sr;
+       struct nilfs_super_block **sbp = nilfs->ns_sbp;
        unsigned dat_entry_size, segment_usage_size, checkpoint_size;
        unsigned inode_size;
        int err;
@@ -124,60 +174,44 @@ static int nilfs_load_super_root(struct the_nilfs *nilfs,
                return err;
 
        down_read(&nilfs->ns_sem);
-       dat_entry_size = le16_to_cpu(nilfs->ns_sbp->s_dat_entry_size);
-       checkpoint_size = le16_to_cpu(nilfs->ns_sbp->s_checkpoint_size);
-       segment_usage_size = le16_to_cpu(nilfs->ns_sbp->s_segment_usage_size);
+       dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
+       checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
+       segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
        up_read(&nilfs->ns_sem);
 
        inode_size = nilfs->ns_inode_size;
 
        err = -ENOMEM;
-       nilfs->ns_dat = nilfs_mdt_new(
-               nilfs, NULL, NILFS_DAT_INO, NILFS_DAT_GFP);
+       nilfs->ns_dat = nilfs_dat_new(nilfs, dat_entry_size);
        if (unlikely(!nilfs->ns_dat))
                goto failed;
 
-       nilfs->ns_gc_dat = nilfs_mdt_new(
-               nilfs, NULL, NILFS_DAT_INO, NILFS_DAT_GFP);
+       nilfs->ns_gc_dat = nilfs_dat_new(nilfs, dat_entry_size);
        if (unlikely(!nilfs->ns_gc_dat))
                goto failed_dat;
 
-       nilfs->ns_cpfile = nilfs_mdt_new(
-               nilfs, NULL, NILFS_CPFILE_INO, NILFS_CPFILE_GFP);
+       nilfs->ns_cpfile = nilfs_cpfile_new(nilfs, checkpoint_size);
        if (unlikely(!nilfs->ns_cpfile))
                goto failed_gc_dat;
 
-       nilfs->ns_sufile = nilfs_mdt_new(
-               nilfs, NULL, NILFS_SUFILE_INO, NILFS_SUFILE_GFP);
+       nilfs->ns_sufile = nilfs_sufile_new(nilfs, segment_usage_size);
        if (unlikely(!nilfs->ns_sufile))
                goto failed_cpfile;
 
-       err = nilfs_palloc_init_blockgroup(nilfs->ns_dat, dat_entry_size);
-       if (unlikely(err))
-               goto failed_sufile;
-
-       err = nilfs_palloc_init_blockgroup(nilfs->ns_gc_dat, dat_entry_size);
-       if (unlikely(err))
-               goto failed_sufile;
-
        nilfs_mdt_set_shadow(nilfs->ns_dat, nilfs->ns_gc_dat);
-       nilfs_mdt_set_entry_size(nilfs->ns_cpfile, checkpoint_size,
-                                sizeof(struct nilfs_cpfile_header));
-       nilfs_mdt_set_entry_size(nilfs->ns_sufile, segment_usage_size,
-                                sizeof(struct nilfs_sufile_header));
 
-       err = nilfs_mdt_read_inode_direct(
-               nilfs->ns_dat, bh_sr, NILFS_SR_DAT_OFFSET(inode_size));
+       err = nilfs_dat_read(nilfs->ns_dat, (void *)bh_sr->b_data +
+                            NILFS_SR_DAT_OFFSET(inode_size));
        if (unlikely(err))
                goto failed_sufile;
 
-       err = nilfs_mdt_read_inode_direct(
-               nilfs->ns_cpfile, bh_sr, NILFS_SR_CPFILE_OFFSET(inode_size));
+       err = nilfs_cpfile_read(nilfs->ns_cpfile, (void *)bh_sr->b_data +
+                               NILFS_SR_CPFILE_OFFSET(inode_size));
        if (unlikely(err))
                goto failed_sufile;
 
-       err = nilfs_mdt_read_inode_direct(
-               nilfs->ns_sufile, bh_sr, NILFS_SR_SUFILE_OFFSET(inode_size));
+       err = nilfs_sufile_read(nilfs->ns_sufile, (void *)bh_sr->b_data +
+                               NILFS_SR_SUFILE_OFFSET(inode_size));
        if (unlikely(err))
                goto failed_sufile;
 
@@ -227,29 +261,30 @@ int load_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi)
        struct nilfs_recovery_info ri;
        unsigned int s_flags = sbi->s_super->s_flags;
        int really_read_only = bdev_read_only(nilfs->ns_bdev);
-       unsigned valid_fs;
-       int err = 0;
-
-       nilfs_init_recovery_info(&ri);
+       int valid_fs = nilfs_valid_fs(nilfs);
+       int err;
 
-       down_write(&nilfs->ns_sem);
-       valid_fs = (nilfs->ns_mount_state & NILFS_VALID_FS);
-       up_write(&nilfs->ns_sem);
+       if (nilfs_loaded(nilfs)) {
+               if (valid_fs ||
+                   ((s_flags & MS_RDONLY) && nilfs_test_opt(sbi, NORECOVERY)))
+                       return 0;
+               printk(KERN_ERR "NILFS: the filesystem is in an incomplete "
+                      "recovery state.\n");
+               return -EINVAL;
+       }
 
-       if (!valid_fs && (s_flags & MS_RDONLY)) {
-               printk(KERN_INFO "NILFS: INFO: recovery "
-                      "required for readonly filesystem.\n");
-               if (really_read_only) {
-                       printk(KERN_ERR "NILFS: write access "
-                              "unavailable, cannot proceed.\n");
-                       err = -EROFS;
-                       goto failed;
+       if (!valid_fs) {
+               printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
+               if (s_flags & MS_RDONLY) {
+                       printk(KERN_INFO "NILFS: INFO: recovery "
+                              "required for readonly filesystem.\n");
+                       printk(KERN_INFO "NILFS: write access will "
+                              "be enabled during recovery.\n");
                }
-               printk(KERN_INFO "NILFS: write access will "
-                      "be enabled during recovery.\n");
-               sbi->s_super->s_flags &= ~MS_RDONLY;
        }
 
+       nilfs_init_recovery_info(&ri);
+
        err = nilfs_search_super_root(nilfs, sbi, &ri);
        if (unlikely(err)) {
                printk(KERN_ERR "NILFS: error searching super root.\n");
@@ -262,22 +297,56 @@ int load_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi)
                goto failed;
        }
 
-       if (!valid_fs) {
-               err = nilfs_recover_logical_segments(nilfs, sbi, &ri);
-               if (unlikely(err)) {
-                       nilfs_mdt_destroy(nilfs->ns_cpfile);
-                       nilfs_mdt_destroy(nilfs->ns_sufile);
-                       nilfs_mdt_destroy(nilfs->ns_dat);
-                       goto failed;
+       if (valid_fs)
+               goto skip_recovery;
+
+       if (s_flags & MS_RDONLY) {
+               if (nilfs_test_opt(sbi, NORECOVERY)) {
+                       printk(KERN_INFO "NILFS: norecovery option specified. "
+                              "skipping roll-forward recovery\n");
+                       goto skip_recovery;
                }
-               if (ri.ri_need_recovery == NILFS_RECOVERY_SR_UPDATED) {
-                       down_write(&nilfs->ns_sem);
-                       nilfs_update_last_segment(sbi, 0);
-                       up_write(&nilfs->ns_sem);
+               if (really_read_only) {
+                       printk(KERN_ERR "NILFS: write access "
+                              "unavailable, cannot proceed.\n");
+                       err = -EROFS;
+                       goto failed_unload;
                }
+               sbi->s_super->s_flags &= ~MS_RDONLY;
+       } else if (nilfs_test_opt(sbi, NORECOVERY)) {
+               printk(KERN_ERR "NILFS: recovery cancelled because norecovery "
+                      "option was specified for a read/write mount\n");
+               err = -EINVAL;
+               goto failed_unload;
        }
 
+       err = nilfs_recover_logical_segments(nilfs, sbi, &ri);
+       if (err)
+               goto failed_unload;
+
+       down_write(&nilfs->ns_sem);
+       nilfs->ns_mount_state |= NILFS_VALID_FS;
+       nilfs->ns_sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
+       err = nilfs_commit_super(sbi, 1);
+       up_write(&nilfs->ns_sem);
+
+       if (err) {
+               printk(KERN_ERR "NILFS: failed to update super block. "
+                      "recovery unfinished.\n");
+               goto failed_unload;
+       }
+       printk(KERN_INFO "NILFS: recovery complete.\n");
+
+ skip_recovery:
        set_nilfs_loaded(nilfs);
+       nilfs_clear_recovery_info(&ri);
+       sbi->s_super->s_flags = s_flags;
+       return 0;
+
+ failed_unload:
+       nilfs_mdt_destroy(nilfs->ns_cpfile);
+       nilfs_mdt_destroy(nilfs->ns_sufile);
+       nilfs_mdt_destroy(nilfs->ns_dat);
 
  failed:
        nilfs_clear_recovery_info(&ri);
@@ -296,9 +365,8 @@ static unsigned long long nilfs_max_size(unsigned int blkbits)
        return res;
 }
 
-static int
-nilfs_store_disk_layout(struct the_nilfs *nilfs, struct super_block *sb,
-                       struct nilfs_super_block *sbp)
+static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
+                                  struct nilfs_super_block *sbp)
 {
        if (le32_to_cpu(sbp->s_rev_level) != NILFS_CURRENT_REV) {
                printk(KERN_ERR "NILFS: revision mismatch "
@@ -309,6 +377,10 @@ nilfs_store_disk_layout(struct the_nilfs *nilfs, struct super_block *sb,
                       NILFS_CURRENT_REV, NILFS_MINOR_REV);
                return -EINVAL;
        }
+       nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
+       if (nilfs->ns_sbsize > BLOCK_SIZE)
+               return -EINVAL;
+
        nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
        nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
 
@@ -330,6 +402,122 @@ nilfs_store_disk_layout(struct the_nilfs *nilfs, struct super_block *sb,
        return 0;
 }
 
+static int nilfs_valid_sb(struct nilfs_super_block *sbp)
+{
+       static unsigned char sum[4];
+       const int sumoff = offsetof(struct nilfs_super_block, s_sum);
+       size_t bytes;
+       u32 crc;
+
+       if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
+               return 0;
+       bytes = le16_to_cpu(sbp->s_bytes);
+       if (bytes > BLOCK_SIZE)
+               return 0;
+       crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
+                      sumoff);
+       crc = crc32_le(crc, sum, 4);
+       crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
+                      bytes - sumoff - 4);
+       return crc == le32_to_cpu(sbp->s_sum);
+}
+
+static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
+{
+       return offset < ((le64_to_cpu(sbp->s_nsegments) *
+                         le32_to_cpu(sbp->s_blocks_per_segment)) <<
+                        (le32_to_cpu(sbp->s_log_block_size) + 10));
+}
+
+static void nilfs_release_super_block(struct the_nilfs *nilfs)
+{
+       int i;
+
+       for (i = 0; i < 2; i++) {
+               if (nilfs->ns_sbp[i]) {
+                       brelse(nilfs->ns_sbh[i]);
+                       nilfs->ns_sbh[i] = NULL;
+                       nilfs->ns_sbp[i] = NULL;
+               }
+       }
+}
+
+void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
+{
+       brelse(nilfs->ns_sbh[0]);
+       nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
+       nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
+       nilfs->ns_sbh[1] = NULL;
+       nilfs->ns_sbp[1] = NULL;
+}
+
+void nilfs_swap_super_block(struct the_nilfs *nilfs)
+{
+       struct buffer_head *tsbh = nilfs->ns_sbh[0];
+       struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
+
+       nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
+       nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
+       nilfs->ns_sbh[1] = tsbh;
+       nilfs->ns_sbp[1] = tsbp;
+}
+
+static int nilfs_load_super_block(struct the_nilfs *nilfs,
+                                 struct super_block *sb, int blocksize,
+                                 struct nilfs_super_block **sbpp)
+{
+       struct nilfs_super_block **sbp = nilfs->ns_sbp;
+       struct buffer_head **sbh = nilfs->ns_sbh;
+       u64 sb2off = NILFS_SB2_OFFSET_BYTES(nilfs->ns_bdev->bd_inode->i_size);
+       int valid[2], swp = 0;
+
+       sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
+                                       &sbh[0]);
+       sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
+
+       if (!sbp[0]) {
+               if (!sbp[1]) {
+                       printk(KERN_ERR "NILFS: unable to read superblock\n");
+                       return -EIO;
+               }
+               printk(KERN_WARNING
+                      "NILFS warning: unable to read primary superblock\n");
+       } else if (!sbp[1])
+               printk(KERN_WARNING
+                      "NILFS warning: unable to read secondary superblock\n");
+
+       valid[0] = nilfs_valid_sb(sbp[0]);
+       valid[1] = nilfs_valid_sb(sbp[1]);
+       swp = valid[1] &&
+               (!valid[0] ||
+                le64_to_cpu(sbp[1]->s_wtime) > le64_to_cpu(sbp[0]->s_wtime));
+
+       if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
+               brelse(sbh[1]);
+               sbh[1] = NULL;
+               sbp[1] = NULL;
+               swp = 0;
+       }
+       if (!valid[swp]) {
+               nilfs_release_super_block(nilfs);
+               printk(KERN_ERR "NILFS: Can't find nilfs on dev %s.\n",
+                      sb->s_id);
+               return -EINVAL;
+       }
+
+       if (swp) {
+               printk(KERN_WARNING "NILFS warning: broken superblock. "
+                      "using spare superblock.\n");
+               nilfs_swap_super_block(nilfs);
+       }
+
+       nilfs->ns_sbwtime[0] = le64_to_cpu(sbp[0]->s_wtime);
+       nilfs->ns_sbwtime[1] = valid[!swp] ? le64_to_cpu(sbp[1]->s_wtime) : 0;
+       nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
+       *sbpp = sbp[0];
+       return 0;
+}
+
 /**
  * init_nilfs - initialize a NILFS instance.
  * @nilfs: the_nilfs structure
@@ -352,16 +540,15 @@ nilfs_store_disk_layout(struct the_nilfs *nilfs, struct super_block *sb,
 int init_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi, char *data)
 {
        struct super_block *sb = sbi->s_super;
-       struct buffer_head *sbh;
        struct nilfs_super_block *sbp;
        struct backing_dev_info *bdi;
        int blocksize;
-       int err = 0;
+       int err;
 
        down_write(&nilfs->ns_sem);
        if (nilfs_init(nilfs)) {
                /* Load values from existing the_nilfs */
-               sbp = nilfs->ns_sbp;
+               sbp = nilfs->ns_sbp[0];
                err = nilfs_store_magic_and_option(sb, sbp, data);
                if (err)
                        goto out;
@@ -377,40 +564,52 @@ int init_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi, char *data)
                goto out;
        }
 
-       sbp = nilfs_load_super_block(sb, &sbh);
-       if (!sbp) {
+       blocksize = sb_min_blocksize(sb, BLOCK_SIZE);
+       if (!blocksize) {
+               printk(KERN_ERR "NILFS: unable to set blocksize\n");
                err = -EINVAL;
                goto out;
        }
+       err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
+       if (err)
+               goto out;
+
        err = nilfs_store_magic_and_option(sb, sbp, data);
        if (err)
                goto failed_sbh;
 
        blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
        if (sb->s_blocksize != blocksize) {
-               sbp = nilfs_reload_super_block(sb, &sbh, blocksize);
-               if (!sbp) {
+               int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
+
+               if (blocksize < hw_blocksize) {
+                       printk(KERN_ERR
+                              "NILFS: blocksize %d too small for device "
+                              "(sector-size = %d).\n",
+                              blocksize, hw_blocksize);
                        err = -EINVAL;
+                       goto failed_sbh;
+               }
+               nilfs_release_super_block(nilfs);
+               sb_set_blocksize(sb, blocksize);
+
+               err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
+               if (err)
                        goto out;
                        /* not failed_sbh; sbh is released automatically
                           when reloading fails. */
-               }
        }
        nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
 
-       err = nilfs_store_disk_layout(nilfs, sb, sbp);
+       err = nilfs_store_disk_layout(nilfs, sbp);
        if (err)
                goto failed_sbh;
 
        sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
 
        nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
-       nilfs->ns_sbh = sbh;
-       nilfs->ns_sbp = sbp;
 
-       bdi = nilfs->ns_bdev->bd_inode_backing_dev_info;
-       if (!bdi)
-               bdi = nilfs->ns_bdev->bd_inode->i_mapping->backing_dev_info;
+       bdi = nilfs->ns_bdev->bd_inode->i_mapping->backing_dev_info;
        nilfs->ns_bdi = bdi ? : &default_backing_dev_info;
 
        /* Finding last segment */
@@ -443,7 +642,7 @@ int init_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi, char *data)
        return err;
 
  failed_sbh:
-       brelse(sbh);
+       nilfs_release_super_block(nilfs);
        goto out;
 }
 
@@ -451,30 +650,73 @@ int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
 {
        struct inode *dat = nilfs_dat_inode(nilfs);
        unsigned long ncleansegs;
-       int err;
 
        down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
-       err = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile, &ncleansegs);
+       ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
        up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
-       if (likely(!err))
-               *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
-       return err;
+       *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
+       return 0;
 }
 
 int nilfs_near_disk_full(struct the_nilfs *nilfs)
 {
-       struct inode *sufile = nilfs->ns_sufile;
        unsigned long ncleansegs, nincsegs;
-       int ret;
 
-       ret = nilfs_sufile_get_ncleansegs(sufile, &ncleansegs);
-       if (likely(!ret)) {
-               nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
-                       nilfs->ns_blocks_per_segment + 1;
-               if (ncleansegs <= nilfs->ns_nrsvsegs + nincsegs)
-                       ret++;
+       ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
+       nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
+               nilfs->ns_blocks_per_segment + 1;
+
+       return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
+}
+
+/**
+ * nilfs_find_sbinfo - find existing nilfs_sb_info structure
+ * @nilfs: nilfs object
+ * @rw_mount: mount type (non-zero value for read/write mount)
+ * @cno: checkpoint number (zero for read-only mount)
+ *
+ * nilfs_find_sbinfo() returns the nilfs_sb_info structure which
+ * @rw_mount and @cno (in case of snapshots) matched.  If no instance
+ * was found, NULL is returned.  Although the super block instance can
+ * be unmounted after this function returns, the nilfs_sb_info struct
+ * is kept on memory until nilfs_put_sbinfo() is called.
+ */
+struct nilfs_sb_info *nilfs_find_sbinfo(struct the_nilfs *nilfs,
+                                       int rw_mount, __u64 cno)
+{
+       struct nilfs_sb_info *sbi;
+
+       down_read(&nilfs->ns_super_sem);
+       /*
+        * The SNAPSHOT flag and sb->s_flags are supposed to be
+        * protected with nilfs->ns_super_sem.
+        */
+       sbi = nilfs->ns_current;
+       if (rw_mount) {
+               if (sbi && !(sbi->s_super->s_flags & MS_RDONLY))
+                       goto found; /* read/write mount */
+               else
+                       goto out;
+       } else if (cno == 0) {
+               if (sbi && (sbi->s_super->s_flags & MS_RDONLY))
+                       goto found; /* read-only mount */
+               else
+                       goto out;
        }
-       return ret;
+
+       list_for_each_entry(sbi, &nilfs->ns_supers, s_list) {
+               if (nilfs_test_opt(sbi, SNAPSHOT) &&
+                   sbi->s_snapshot_cno == cno)
+                       goto found; /* snapshot mount */
+       }
+ out:
+       up_read(&nilfs->ns_super_sem);
+       return NULL;
+
+ found:
+       atomic_inc(&sbi->s_count);
+       up_read(&nilfs->ns_super_sem);
+       return sbi;
 }
 
 int nilfs_checkpoint_is_mounted(struct the_nilfs *nilfs, __u64 cno,
@@ -483,7 +725,7 @@ int nilfs_checkpoint_is_mounted(struct the_nilfs *nilfs, __u64 cno,
        struct nilfs_sb_info *sbi;
        int ret = 0;
 
-       down_read(&nilfs->ns_sem);
+       down_read(&nilfs->ns_super_sem);
        if (cno == 0 || cno > nilfs->ns_cno)
                goto out_unlock;
 
@@ -500,6 +742,6 @@ int nilfs_checkpoint_is_mounted(struct the_nilfs *nilfs, __u64 cno,
                ret++;
 
  out_unlock:
-       up_read(&nilfs->ns_sem);
+       up_read(&nilfs->ns_super_sem);
        return ret;
 }