Quelle  resize.c   Sprache: C

// SPDX-License-Identifier: GPL-2.0-or-later
/*
 * resize.c
 *
 * volume resize.
 * Inspired by ext3/resize.c.
 *
 * Copyright (C) 2007 Oracle.  All rights reserved.
 */

#include <linux/fs.h>
#include <linux/types.h>

#include <cluster/masklog.h>

#include "ocfs2.h"

#include "alloc.h"
#include "dlmglue.h"
#include "inode.h"
#include "journal.h"
#include "super.h"
#include "sysfile.h"
#include "uptodate.h"
#include "ocfs2_trace.h"

#include "buffer_head_io.h"
#include "suballoc.h"
#include "resize.h"

/*
 * Check whether there are new backup superblocks exist
 * in the last group. If there are some, mark them or clear
 * them in the bitmap.
 *
 * Return how many backups we find in the last group.
 */
static u16 ocfs2_calc_new_backup_super(struct inode *inode,
           struct ocfs2_group_desc *gd,
           u16 cl_cpg,
           u16 old_bg_clusters,
           int set)
{
 int i;
 u16 backups = 0;
 u32 cluster, lgd_cluster;
 u64 blkno, gd_blkno, lgd_blkno = le64_to_cpu(gd->bg_blkno);

 for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
  blkno = ocfs2_backup_super_blkno(inode->i_sb, i);
  cluster = ocfs2_blocks_to_clusters(inode->i_sb, blkno);

  gd_blkno = ocfs2_which_cluster_group(inode, cluster);
  if (gd_blkno < lgd_blkno)
   continue;
  else if (gd_blkno > lgd_blkno)
   break;

  /* check if already done backup super */
  lgd_cluster = ocfs2_blocks_to_clusters(inode->i_sb, lgd_blkno);
  lgd_cluster += old_bg_clusters;
  if (lgd_cluster >= cluster)
   continue;

  if (set)
   ocfs2_set_bit(cluster % cl_cpg,
          (unsigned long *)gd->bg_bitmap);
  else
   ocfs2_clear_bit(cluster % cl_cpg,
     (unsigned long *)gd->bg_bitmap);
  backups++;
 }

 return backups;
}

static int ocfs2_update_last_group_and_inode(handle_t *handle,
          struct inode *bm_inode,
          struct buffer_head *bm_bh,
          struct buffer_head *group_bh,
          u32 first_new_cluster,
          int new_clusters)
{
 int ret = 0;
 struct ocfs2_super *osb = OCFS2_SB(bm_inode->i_sb);
 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bm_bh->b_data;
 struct ocfs2_chain_list *cl = &fe->id2.i_chain;
 struct ocfs2_chain_rec *cr;
 struct ocfs2_group_desc *group;
 u16 chain, num_bits, backups = 0;
 u16 cl_bpc = le16_to_cpu(cl->cl_bpc);
 u16 cl_cpg = le16_to_cpu(cl->cl_cpg);
 u16 old_bg_clusters;
 u16 contig_bits;
 __le16 old_bg_contig_free_bits;

 trace_ocfs2_update_last_group_and_inode(new_clusters,
      first_new_cluster);

 ret = ocfs2_journal_access_gd(handle, INODE_CACHE(bm_inode),
          group_bh, OCFS2_JOURNAL_ACCESS_WRITE);
 if (ret < 0) {
  mlog_errno(ret);
  goto out;
 }

 group = (struct ocfs2_group_desc *)group_bh->b_data;

 old_bg_clusters = le16_to_cpu(group->bg_bits) / cl_bpc;
 /* update the group first. */
 num_bits = new_clusters * cl_bpc;
 le16_add_cpu(&group->bg_bits, num_bits);
 le16_add_cpu(&group->bg_free_bits_count, num_bits);

 /*
 * check whether there are some new backup superblocks exist in
 * this group and update the group bitmap accordingly.
 */
 if (OCFS2_HAS_COMPAT_FEATURE(osb->sb,
         OCFS2_FEATURE_COMPAT_BACKUP_SB)) {
  backups = ocfs2_calc_new_backup_super(bm_inode,
           group,
           cl_cpg, old_bg_clusters, 1);
  le16_add_cpu(&group->bg_free_bits_count, -1 * backups);
 }

 contig_bits = ocfs2_find_max_contig_free_bits(group->bg_bitmap,
     le16_to_cpu(group->bg_bits), 0);
 old_bg_contig_free_bits = group->bg_contig_free_bits;
 group->bg_contig_free_bits = cpu_to_le16(contig_bits);

 ocfs2_journal_dirty(handle, group_bh);

 /* update the inode accordingly. */
 ret = ocfs2_journal_access_di(handle, INODE_CACHE(bm_inode), bm_bh,
          OCFS2_JOURNAL_ACCESS_WRITE);
 if (ret < 0) {
  mlog_errno(ret);
  goto out_rollback;
 }

 chain = le16_to_cpu(group->bg_chain);
 cr = (&cl->cl_recs[chain]);
 le32_add_cpu(&cr->c_total, num_bits);
 le32_add_cpu(&cr->c_free, num_bits);
 le32_add_cpu(&fe->id1.bitmap1.i_total, num_bits);
 le32_add_cpu(&fe->i_clusters, new_clusters);

 if (backups) {
  le32_add_cpu(&cr->c_free, -1 * backups);
  le32_add_cpu(&fe->id1.bitmap1.i_used, backups);
 }

 spin_lock(&OCFS2_I(bm_inode)->ip_lock);
 OCFS2_I(bm_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
 le64_add_cpu(&fe->i_size, (u64)new_clusters << osb->s_clustersize_bits);
 spin_unlock(&OCFS2_I(bm_inode)->ip_lock);
 i_size_write(bm_inode, le64_to_cpu(fe->i_size));

 ocfs2_journal_dirty(handle, bm_bh);

out_rollback:
 if (ret < 0) {
  ocfs2_calc_new_backup_super(bm_inode,
         group,
         cl_cpg, old_bg_clusters, 0);
  le16_add_cpu(&group->bg_free_bits_count, backups);
  le16_add_cpu(&group->bg_bits, -1 * num_bits);
  le16_add_cpu(&group->bg_free_bits_count, -1 * num_bits);
  group->bg_contig_free_bits = old_bg_contig_free_bits;
 }
out:
 if (ret)
  mlog_errno(ret);
 return ret;
}

static int update_backups(struct inode * inode, u32 clusters, char *data)
{
 int i, ret = 0;
 u32 cluster;
 u64 blkno;
 struct buffer_head *backup = NULL;
 struct ocfs2_dinode *backup_di = NULL;
 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);

 /* calculate the real backups we need to update. */
 for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
  blkno = ocfs2_backup_super_blkno(inode->i_sb, i);
  cluster = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
  if (cluster >= clusters)
   break;

  ret = ocfs2_read_blocks_sync(osb, blkno, 1, &backup);
  if (ret < 0) {
   mlog_errno(ret);
   break;
  }

  memcpy(backup->b_data, data, inode->i_sb->s_blocksize);

  backup_di = (struct ocfs2_dinode *)backup->b_data;
  backup_di->i_blkno = cpu_to_le64(blkno);

  ret = ocfs2_write_super_or_backup(osb, backup);
  brelse(backup);
  backup = NULL;
  if (ret < 0) {
   mlog_errno(ret);
   break;
  }
 }

 return ret;
}

static void ocfs2_update_super_and_backups(struct inode *inode,
        int new_clusters)
{
 int ret;
 u32 clusters = 0;
 struct buffer_head *super_bh = NULL;
 struct ocfs2_dinode *super_di = NULL;
 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);

 /*
 * update the superblock last.
 * It doesn't matter if the write failed.
 */
 ret = ocfs2_read_blocks_sync(osb, OCFS2_SUPER_BLOCK_BLKNO, 1,
         &super_bh);
 if (ret < 0) {
  mlog_errno(ret);
  goto out;
 }

 super_di = (struct ocfs2_dinode *)super_bh->b_data;
 le32_add_cpu(&super_di->i_clusters, new_clusters);
 clusters = le32_to_cpu(super_di->i_clusters);

 ret = ocfs2_write_super_or_backup(osb, super_bh);
 if (ret < 0) {
  mlog_errno(ret);
  goto out;
 }

 if (OCFS2_HAS_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_COMPAT_BACKUP_SB))
  ret = update_backups(inode, clusters, super_bh->b_data);

out:
 brelse(super_bh);
 if (ret)
  printk(KERN_WARNING "ocfs2: Failed to update super blocks on %s"
   " during fs resize. This condition is not fatal,"
   " but fsck.ocfs2 should be run to fix it\n",
   osb->dev_str);
 return;
}

/*
 * Extend the filesystem to the new number of clusters specified.  This entry
 * point is only used to extend the current filesystem to the end of the last
 * existing group.
 */
int ocfs2_group_extend(struct inode * inode, int new_clusters)
{
 int ret;
 handle_t *handle;
 struct buffer_head *main_bm_bh = NULL;
 struct buffer_head *group_bh = NULL;
 struct inode *main_bm_inode = NULL;
 struct ocfs2_dinode *fe = NULL;
 struct ocfs2_group_desc *group = NULL;
 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 u16 cl_bpc;
 u32 first_new_cluster;
 u64 lgd_blkno;

 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
  return -EROFS;

 if (new_clusters < 0)
  return -EINVAL;
 else if (new_clusters == 0)
  return 0;

 main_bm_inode = ocfs2_get_system_file_inode(osb,
          GLOBAL_BITMAP_SYSTEM_INODE,
          OCFS2_INVALID_SLOT);
 if (!main_bm_inode) {
  ret = -EINVAL;
  mlog_errno(ret);
  goto out;
 }

 inode_lock(main_bm_inode);

 ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
 if (ret < 0) {
  mlog_errno(ret);
  goto out_mutex;
 }

 fe = (struct ocfs2_dinode *)main_bm_bh->b_data;

 /* main_bm_bh is validated by inode read inside ocfs2_inode_lock(),
 * so any corruption is a code bug. */
 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));

 if (le16_to_cpu(fe->id2.i_chain.cl_cpg) !=
  ocfs2_group_bitmap_size(osb->sb, 0,
     osb->s_feature_incompat) * 8) {
  mlog(ML_ERROR, "The disk is too old and small. "
       "Force to do offline resize.");
  ret = -EINVAL;
  goto out_unlock;
 }

 first_new_cluster = le32_to_cpu(fe->i_clusters);
 lgd_blkno = ocfs2_which_cluster_group(main_bm_inode,
           first_new_cluster - 1);

 ret = ocfs2_read_group_descriptor(main_bm_inode, fe, lgd_blkno,
       &group_bh);
 if (ret < 0) {
  mlog_errno(ret);
  goto out_unlock;
 }
 group = (struct ocfs2_group_desc *)group_bh->b_data;

 cl_bpc = le16_to_cpu(fe->id2.i_chain.cl_bpc);
 if (le16_to_cpu(group->bg_bits) / cl_bpc + new_clusters >
  le16_to_cpu(fe->id2.i_chain.cl_cpg)) {
  ret = -EINVAL;
  goto out_unlock;
 }


 trace_ocfs2_group_extend(
      (unsigned long long)le64_to_cpu(group->bg_blkno), new_clusters);

 handle = ocfs2_start_trans(osb, OCFS2_GROUP_EXTEND_CREDITS);
 if (IS_ERR(handle)) {
  mlog_errno(PTR_ERR(handle));
  ret = -EINVAL;
  goto out_unlock;
 }

 /* update the last group descriptor and inode. */
 ret = ocfs2_update_last_group_and_inode(handle, main_bm_inode,
      main_bm_bh, group_bh,
      first_new_cluster,
      new_clusters);
 if (ret) {
  mlog_errno(ret);
  goto out_commit;
 }

 ocfs2_update_super_and_backups(main_bm_inode, new_clusters);

out_commit:
 ocfs2_commit_trans(osb, handle);
out_unlock:
 brelse(group_bh);
 brelse(main_bm_bh);

 ocfs2_inode_unlock(main_bm_inode, 1);

out_mutex:
 inode_unlock(main_bm_inode);
 iput(main_bm_inode);

out:
 return ret;
}

static int ocfs2_check_new_group(struct inode *inode,
     struct ocfs2_dinode *di,
     struct ocfs2_new_group_input *input,
     struct buffer_head *group_bh)
{
 int ret;
 struct ocfs2_group_desc *gd =
  (struct ocfs2_group_desc *)group_bh->b_data;
 u16 cl_bpc = le16_to_cpu(di->id2.i_chain.cl_bpc);

 ret = ocfs2_check_group_descriptor(inode->i_sb, di, group_bh);
 if (ret)
  goto out;

 ret = -EINVAL;
 if (le16_to_cpu(gd->bg_chain) != input->chain)
  mlog(ML_ERROR, "Group descriptor # %llu has bad chain %u "
       "while input has %u set.\n",
       (unsigned long long)le64_to_cpu(gd->bg_blkno),
       le16_to_cpu(gd->bg_chain), input->chain);
 else if (le16_to_cpu(gd->bg_bits) != input->clusters * cl_bpc)
  mlog(ML_ERROR, "Group descriptor # %llu has bit count %u but "
       "input has %u clusters set\n",
       (unsigned long long)le64_to_cpu(gd->bg_blkno),
       le16_to_cpu(gd->bg_bits), input->clusters);
 else if (le16_to_cpu(gd->bg_free_bits_count) != input->frees * cl_bpc)
  mlog(ML_ERROR, "Group descriptor # %llu has free bit count %u "
       "but it should have %u set\n",
       (unsigned long long)le64_to_cpu(gd->bg_blkno),
       le16_to_cpu(gd->bg_bits),
       input->frees * cl_bpc);
 else
  ret = 0;

out:
 return ret;
}

static int ocfs2_verify_group_and_input(struct inode *inode,
     struct ocfs2_dinode *di,
     struct ocfs2_new_group_input *input,
     struct buffer_head *group_bh)
{
 u16 cl_count = le16_to_cpu(di->id2.i_chain.cl_count);
 u16 cl_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
 u16 next_free = le16_to_cpu(di->id2.i_chain.cl_next_free_rec);
 u32 cluster = ocfs2_blocks_to_clusters(inode->i_sb, input->group);
 u32 total_clusters = le32_to_cpu(di->i_clusters);
 int ret = -EINVAL;

 if (cluster < total_clusters)
  mlog(ML_ERROR, "add a group which is in the current volume.\n");
 else if (input->chain >= cl_count)
  mlog(ML_ERROR, "input chain exceeds the limit.\n");
 else if (next_free != cl_count && next_free != input->chain)
  mlog(ML_ERROR,
       "the add group should be in chain %u\n", next_free);
 else if (total_clusters + input->clusters < total_clusters)
  mlog(ML_ERROR, "add group's clusters overflow.\n");
 else if (input->clusters > cl_cpg)
  mlog(ML_ERROR, "the cluster exceeds the maximum of a group\n");
 else if (input->frees > input->clusters)
  mlog(ML_ERROR, "the free cluster exceeds the total clusters\n");
 else if (total_clusters % cl_cpg != 0)
  mlog(ML_ERROR,
       "the last group isn't full. Use group extend first.\n");
 else if (input->group != ocfs2_which_cluster_group(inode, cluster))
  mlog(ML_ERROR, "group blkno is invalid\n");
 else if ((ret = ocfs2_check_new_group(inode, di, input, group_bh)))
  mlog(ML_ERROR, "group descriptor check failed.\n");
 else
  ret = 0;

 return ret;
}

/* Add a new group descriptor to global_bitmap. */
int ocfs2_group_add(struct inode *inode, struct ocfs2_new_group_input *input)
{
 int ret;
 handle_t *handle;
 struct buffer_head *main_bm_bh = NULL;
 struct inode *main_bm_inode = NULL;
 struct ocfs2_dinode *fe = NULL;
 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 struct buffer_head *group_bh = NULL;
 struct ocfs2_group_desc *group = NULL;
 struct ocfs2_chain_list *cl;
 struct ocfs2_chain_rec *cr;
 u16 cl_bpc;
 u64 bg_ptr;

 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
  return -EROFS;

 main_bm_inode = ocfs2_get_system_file_inode(osb,
          GLOBAL_BITMAP_SYSTEM_INODE,
          OCFS2_INVALID_SLOT);
 if (!main_bm_inode) {
  ret = -EINVAL;
  mlog_errno(ret);
  goto out;
 }

 inode_lock(main_bm_inode);

 ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
 if (ret < 0) {
  mlog_errno(ret);
  goto out_mutex;
 }

 fe = (struct ocfs2_dinode *)main_bm_bh->b_data;

 if (le16_to_cpu(fe->id2.i_chain.cl_cpg) !=
  ocfs2_group_bitmap_size(osb->sb, 0,
     osb->s_feature_incompat) * 8) {
  mlog(ML_ERROR, "The disk is too old and small."
       " Force to do offline resize.");
  ret = -EINVAL;
  goto out_unlock;
 }

 ret = ocfs2_read_blocks_sync(osb, input->group, 1, &group_bh);
 if (ret < 0) {
  mlog(ML_ERROR, "Can't read the group descriptor # %llu "
       "from the device.", (unsigned long long)input->group);
  goto out_unlock;
 }

 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), group_bh);

 ret = ocfs2_verify_group_and_input(main_bm_inode, fe, input, group_bh);
 if (ret) {
  mlog_errno(ret);
  goto out_free_group_bh;
 }

 trace_ocfs2_group_add((unsigned long long)input->group,
          input->chain, input->clusters, input->frees);

 handle = ocfs2_start_trans(osb, OCFS2_GROUP_ADD_CREDITS);
 if (IS_ERR(handle)) {
  mlog_errno(PTR_ERR(handle));
  ret = -EINVAL;
  goto out_free_group_bh;
 }

 cl_bpc = le16_to_cpu(fe->id2.i_chain.cl_bpc);
 cl = &fe->id2.i_chain;
 cr = &cl->cl_recs[input->chain];

 ret = ocfs2_journal_access_gd(handle, INODE_CACHE(main_bm_inode),
          group_bh, OCFS2_JOURNAL_ACCESS_WRITE);
 if (ret < 0) {
  mlog_errno(ret);
  goto out_commit;
 }

 group = (struct ocfs2_group_desc *)group_bh->b_data;
 bg_ptr = le64_to_cpu(group->bg_next_group);
 group->bg_next_group = cr->c_blkno;
 ocfs2_journal_dirty(handle, group_bh);

 ret = ocfs2_journal_access_di(handle, INODE_CACHE(main_bm_inode),
          main_bm_bh, OCFS2_JOURNAL_ACCESS_WRITE);
 if (ret < 0) {
  group->bg_next_group = cpu_to_le64(bg_ptr);
  mlog_errno(ret);
  goto out_commit;
 }

 if (input->chain == le16_to_cpu(cl->cl_next_free_rec)) {
  le16_add_cpu(&cl->cl_next_free_rec, 1);
  memset(cr, 0, sizeof(struct ocfs2_chain_rec));
 }

 cr->c_blkno = cpu_to_le64(input->group);
 le32_add_cpu(&cr->c_total, input->clusters * cl_bpc);
 le32_add_cpu(&cr->c_free, input->frees * cl_bpc);

 le32_add_cpu(&fe->id1.bitmap1.i_total, input->clusters *cl_bpc);
 le32_add_cpu(&fe->id1.bitmap1.i_used,
       (input->clusters - input->frees) * cl_bpc);
 le32_add_cpu(&fe->i_clusters, input->clusters);

 ocfs2_journal_dirty(handle, main_bm_bh);

 spin_lock(&OCFS2_I(main_bm_inode)->ip_lock);
 OCFS2_I(main_bm_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
 le64_add_cpu(&fe->i_size, (u64)input->clusters << osb->s_clustersize_bits);
 spin_unlock(&OCFS2_I(main_bm_inode)->ip_lock);
 i_size_write(main_bm_inode, le64_to_cpu(fe->i_size));

 ocfs2_update_super_and_backups(main_bm_inode, input->clusters);

out_commit:
 ocfs2_commit_trans(osb, handle);

out_free_group_bh:
 if (ret < 0)
  ocfs2_remove_from_cache(INODE_CACHE(inode), group_bh);
 brelse(group_bh);

out_unlock:
 brelse(main_bm_bh);

 ocfs2_inode_unlock(main_bm_inode, 1);

out_mutex:
 inode_unlock(main_bm_inode);
 iput(main_bm_inode);

out:
 return ret;
}