#include "ext4_types.h"
#include "ext4_fs.h"
#include "ext4_super.h"
+#include "ext4_journal.h"
#include "ext4_errno.h"
#include "ext4_blockdev.h"
#include "ext4_crc32c.h"
#include <string.h>
#include <stdlib.h>
+/**@brief Revoke entry during journal replay.*/
struct revoke_entry {
+ /**@brief Block number not to be replayed.*/
ext4_fsblk_t block;
+
+ /**@brief For any transaction id smaller
+ * than trans_id, records of @block
+ * in those transactions should not
+ * be replayed.*/
uint32_t trans_id;
+
+ /**@brief Revoke tree node.*/
RB_ENTRY(revoke_entry) revoke_node;
};
+/**@brief Valid journal replay information.*/
struct recover_info {
+ /**@brief Starting transaction id.*/
uint32_t start_trans_id;
+
+ /**@brief Ending transaction id.*/
uint32_t last_trans_id;
+
+ /**@brief Used as internal argument.*/
uint32_t this_trans_id;
+
+ /**@brief RB-Tree storing revoke entries.*/
RB_HEAD(jbd_revoke, revoke_entry) revoke_root;
};
+/**@brief Journal replay internal arguments.*/
struct replay_arg {
+ /**@brief Journal replay information.*/
struct recover_info *info;
+
+ /**@brief Current block we are on.*/
uint32_t *this_block;
+
+ /**@brief Current trans_id we are on.*/
uint32_t this_trans_id;
};
#define jbd_alloc_revoke_entry() calloc(1, sizeof(struct revoke_entry))
#define jbd_free_revoke_entry(addr) free(addr)
-int jbd_inode_bmap(struct jbd_fs *jbd_fs,
- ext4_lblk_t iblock,
- ext4_fsblk_t *fblock);
-
-int jbd_sb_write(struct jbd_fs *jbd_fs, struct jbd_sb *s)
+/**@brief Write jbd superblock to disk.
+ * @param jbd_fs jbd filesystem
+ * @param s jbd superblock
+ * @return standard error code*/
+static int jbd_sb_write(struct jbd_fs *jbd_fs, struct jbd_sb *s)
{
int rc;
struct ext4_fs *fs = jbd_fs->inode_ref.fs;
EXT4_SUPERBLOCK_SIZE);
}
-int jbd_sb_read(struct jbd_fs *jbd_fs, struct jbd_sb *s)
+/**@brief Read jbd superblock from disk.
+ * @param jbd_fs jbd filesystem
+ * @param s jbd superblock
+ * @return standard error code*/
+static int jbd_sb_read(struct jbd_fs *jbd_fs, struct jbd_sb *s)
{
int rc;
struct ext4_fs *fs = jbd_fs->inode_ref.fs;
EXT4_SUPERBLOCK_SIZE);
}
+/**@brief Verify jbd superblock.
+ * @param sb jbd superblock
+ * @return true if jbd superblock is valid */
static bool jbd_verify_sb(struct jbd_sb *sb)
{
struct jbd_bhdr *header = &sb->header;
return true;
}
+/**@brief Write back dirty jbd superblock to disk.
+ * @param jbd_fs jbd filesystem
+ * @return standard error code*/
static int jbd_write_sb(struct jbd_fs *jbd_fs)
{
int rc = EOK;
return rc;
}
+/**@brief Get reference to jbd filesystem.
+ * @param fs Filesystem to load journal of
+ * @param jbd_fs jbd filesystem
+ * @return standard error code*/
int jbd_get_fs(struct ext4_fs *fs,
struct jbd_fs *jbd_fs)
{
uint32_t journal_ino;
memset(jbd_fs, 0, sizeof(struct jbd_fs));
+ /* See if there is journal inode on this filesystem.*/
+ /* FIXME: detection on existance ofbkejournal bdev is
+ * missing.*/
journal_ino = ext4_get32(&fs->sb, journal_inode_number);
rc = ext4_fs_get_inode_ref(fs,
return rc;
}
+/**@brief Put reference of jbd filesystem.
+ * @param jbd_fs jbd filesystem
+ * @return standard error code*/
int jbd_put_fs(struct jbd_fs *jbd_fs)
{
int rc = EOK;
return rc;
}
+/**@brief Data block lookup helper.
+ * @param jbd_fs jbd filesystem
+ * @param iblock block index
+ * @param fblock logical block address
+ * @return standard error code*/
int jbd_inode_bmap(struct jbd_fs *jbd_fs,
ext4_lblk_t iblock,
ext4_fsblk_t *fblock)
return rc;
}
-int jbd_block_get(struct jbd_fs *jbd_fs,
+/**@brief jbd block get function (through cache).
+ * @param jbd_fs jbd filesystem
+ * @param block block descriptor
+ * @param fblock jbd logical block address
+ * @return standard error code*/
+static int jbd_block_get(struct jbd_fs *jbd_fs,
struct ext4_block *block,
ext4_fsblk_t fblock)
{
/* TODO: journal device. */
int rc;
ext4_lblk_t iblock = (ext4_lblk_t)fblock;
+
+ /* Lookup the logical block address of
+ * fblock.*/
rc = jbd_inode_bmap(jbd_fs, iblock,
&fblock);
if (rc != EOK)
struct ext4_blockdev *bdev = jbd_fs->inode_ref.fs->bdev;
rc = ext4_block_get(bdev, block, fblock);
+
+ /* If succeeded, mark buffer as BC_FLUSH to indicate
+ * that data should be written to disk immediately.*/
+ if (rc == EOK)
+ ext4_bcache_set_flag(block->buf, BC_FLUSH);
+
return rc;
}
-int jbd_block_get_noread(struct jbd_fs *jbd_fs,
+/**@brief jbd block get function (through cache, don't read).
+ * @param jbd_fs jbd filesystem
+ * @param block block descriptor
+ * @param fblock jbd logical block address
+ * @return standard error code*/
+static int jbd_block_get_noread(struct jbd_fs *jbd_fs,
struct ext4_block *block,
ext4_fsblk_t fblock)
{
struct ext4_blockdev *bdev = jbd_fs->inode_ref.fs->bdev;
rc = ext4_block_get_noread(bdev, block, fblock);
+ if (rc == EOK)
+ ext4_bcache_set_flag(block->buf, BC_FLUSH);
+
return rc;
}
-int jbd_block_set(struct jbd_fs *jbd_fs,
+/**@brief jbd block set procedure (through cache).
+ * @param jbd_fs jbd filesystem
+ * @param block block descriptor
+ * @return standard error code*/
+static int jbd_block_set(struct jbd_fs *jbd_fs,
struct ext4_block *block)
{
return ext4_block_set(jbd_fs->inode_ref.fs->bdev,
block);
}
-/*
- * helper functions to deal with 32 or 64bit block numbers.
- */
-int jbd_tag_bytes(struct jbd_fs *jbd_fs)
+/**@brief helper functions to calculate
+ * block tag size, not including UUID part.
+ * @param jbd_fs jbd filesystem
+ * @return tag size in bytes*/
+static int jbd_tag_bytes(struct jbd_fs *jbd_fs)
{
int size;
+ /* It is very easy to deal with the case which
+ * JBD_FEATURE_INCOMPAT_CSUM_V3 is enabled.*/
if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
JBD_FEATURE_INCOMPAT_CSUM_V3))
return sizeof(struct jbd_block_tag3);
size = sizeof(struct jbd_block_tag);
+ /* If JBD_FEATURE_INCOMPAT_CSUM_V2 is enabled,
+ * add 2 bytes to size.*/
if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
JBD_FEATURE_INCOMPAT_CSUM_V2))
size += sizeof(uint16_t);
JBD_FEATURE_INCOMPAT_64BIT))
return size;
+ /* If block number is 4 bytes in size,
+ * minus 4 bytes from size */
return size - sizeof(uint32_t);
}
-/**@brief: tag information. */
+/**@brief Tag information. */
struct tag_info {
+ /**@brief Tag size in bytes, including UUID part.*/
int tag_bytes;
+
+ /**@brief block number stored in this tag.*/
ext4_fsblk_t block;
+
+ /**@brief whether UUID part exists or not.*/
bool uuid_exist;
+
+ /**@brief UUID content if UUID part exists.*/
uint8_t uuid[UUID_SIZE];
+
+ /**@brief Is this the last tag? */
bool last_tag;
};
+/**@brief Extract information from a block tag.
+ * @param __tag pointer to the block tag
+ * @param tag_bytes block tag size of this jbd filesystem
+ * @param remaining size in buffer containing the block tag
+ * @param tag_info information of this tag.
+ * @return EOK when succeed, otherwise return EINVAL.*/
static int
jbd_extract_block_tag(struct jbd_fs *jbd_fs,
void *__tag,
tag_info->uuid_exist = false;
tag_info->last_tag = false;
+ /* See whether it is possible to hold a valid block tag.*/
if (remain_buf_size - tag_bytes < 0)
return EINVAL;
tag_info->block = 0;
if (!(jbd_get32(tag, flags) & JBD_FLAG_SAME_UUID)) {
+ /* See whether it is possible to hold UUID part.*/
if (remain_buf_size - tag_bytes < UUID_SIZE)
return EINVAL;
tag_info->block = 0;
if (!(jbd_get16(tag, flags) & JBD_FLAG_SAME_UUID)) {
+ /* See whether it is possible to hold UUID part.*/
if (remain_buf_size - tag_bytes < UUID_SIZE)
return EINVAL;
return EOK;
}
+/**@brief Write information to a block tag.
+ * @param __tag pointer to the block tag
+ * @param remaining size in buffer containing the block tag
+ * @param tag_info information of this tag.
+ * @return EOK when succeed, otherwise return EINVAL.*/
static int
jbd_write_block_tag(struct jbd_fs *jbd_fs,
void *__tag,
tag_info->tag_bytes = tag_bytes;
+ /* See whether it is possible to hold a valid block tag.*/
if (remain_buf_size - tag_bytes < 0)
return EINVAL;
if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
JBD_FEATURE_INCOMPAT_CSUM_V3)) {
struct jbd_block_tag3 *tag = __tag;
+ memset(tag, 0, sizeof(struct jbd_block_tag3));
jbd_set32(tag, blocknr, tag_info->block);
if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
JBD_FEATURE_INCOMPAT_64BIT))
jbd_set32(tag, blocknr_high, tag_info->block >> 32);
- if (!tag_info->uuid_exist) {
+ if (tag_info->uuid_exist) {
+ /* See whether it is possible to hold UUID part.*/
if (remain_buf_size - tag_bytes < UUID_SIZE)
return EINVAL;
uuid_start = (char *)tag + tag_bytes;
tag_info->tag_bytes += UUID_SIZE;
memcpy(uuid_start, tag_info->uuid, UUID_SIZE);
+ } else
jbd_set32(tag, flags,
jbd_get32(tag, flags) | JBD_FLAG_SAME_UUID);
- }
if (tag_info->last_tag)
jbd_set32(tag, flags,
} else {
struct jbd_block_tag *tag = __tag;
+ memset(tag, 0, sizeof(struct jbd_block_tag));
jbd_set32(tag, blocknr, tag_info->block);
if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
JBD_FEATURE_INCOMPAT_64BIT))
jbd_set32(tag, blocknr_high, tag_info->block >> 32);
- if (!tag_info->uuid_exist) {
+ if (tag_info->uuid_exist) {
+ /* See whether it is possible to hold UUID part.*/
if (remain_buf_size - tag_bytes < UUID_SIZE)
return EINVAL;
uuid_start = (char *)tag + tag_bytes;
tag_info->tag_bytes += UUID_SIZE;
memcpy(uuid_start, tag_info->uuid, UUID_SIZE);
+ } else
jbd_set16(tag, flags,
jbd_get16(tag, flags) | JBD_FLAG_SAME_UUID);
- }
if (tag_info->last_tag)
jbd_set16(tag, flags,
return EOK;
}
+/**@brief Iterate all block tags in a block.
+ * @param jbd_fs jbd filesystem
+ * @param __tag_start pointer to the block
+ * @param tag_tbl_size size of the block
+ * @param func callback routine to indicate that
+ * a block tag is found
+ * @param arg additional argument to be passed to func */
static void
jbd_iterate_block_table(struct jbd_fs *jbd_fs,
void *__tag_start,
void *arg),
void *arg)
{
- ext4_fsblk_t block = 0;
char *tag_start, *tag_ptr;
int tag_bytes = jbd_tag_bytes(jbd_fs);
tag_start = __tag_start;
tag_ptr = tag_start;
+ /* Cut off the size of block tail storing checksum. */
if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
JBD_FEATURE_INCOMPAT_CSUM_V2) ||
JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
break;
if (func)
- func(jbd_fs, block, tag_info.uuid, arg);
+ func(jbd_fs, tag_info.block, tag_info.uuid, arg);
+ /* Stop the iteration when we reach the last tag. */
if (tag_info.last_tag)
break;
return RB_FIND(jbd_revoke, &info->revoke_root, &tmp);
}
+/**@brief Replay a block in a transaction.
+ * @param jbd_fs jbd filesystem
+ * @param block block address to be replayed.*/
static void jbd_replay_block_tags(struct jbd_fs *jbd_fs,
ext4_fsblk_t block,
uint8_t *uuid __unused,
(*this_block)++;
+ /* We replay this block only if the current transaction id
+ * is equal or greater than that in revoke entry.*/
revoke_entry = jbd_revoke_entry_lookup(info, block);
if (revoke_entry &&
arg->this_trans_id < revoke_entry->trans_id)
if (r != EOK)
return;
+ /* We need special treatment for ext4 superblock. */
if (block) {
r = ext4_block_get_noread(fs->bdev, &ext4_block, block);
if (r != EOK) {
return;
}
+/**@brief Add block address to revoke tree, along with
+ * its transaction id.
+ * @param info journal replay info
+ * @param block block address to be replayed.*/
static void jbd_add_revoke_block_tags(struct recover_info *info,
ext4_fsblk_t block)
{
struct revoke_entry *revoke_entry;
ext4_dbg(DEBUG_JBD, "Add block %" PRIu64 " to revoke tree\n", block);
+ /* If the revoke entry with respect to the block address
+ * exists already, update its transaction id.*/
revoke_entry = jbd_revoke_entry_lookup(info, block);
if (revoke_entry) {
revoke_entry->trans_id = info->this_trans_id;
#define ACTION_REVOKE 1
#define ACTION_RECOVER 2
-
+/**@brief Add entries in a revoke block to revoke tree.
+ * @param jbd_fs jbd filesystem
+ * @param header revoke block header
+ * @param recover_info journal replay info*/
static void jbd_build_revoke_tree(struct jbd_fs *jbd_fs,
struct jbd_bhdr *header,
struct recover_info *info)
struct jbd_revoke_header *revoke_hdr =
(struct jbd_revoke_header *)header;
uint32_t i, nr_entries, record_len = 4;
+
+ /* If we are working on a 64bit jbd filesystem, */
if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
JBD_FEATURE_INCOMPAT_64BIT))
record_len = 8;
arg);
}
-int jbd_iterate_log(struct jbd_fs *jbd_fs,
- struct recover_info *info,
- int action)
+/**@brief The core routine of journal replay.
+ * @param jbd_fs jbd filesystem
+ * @param recover_info journal replay info
+ * @param action action needed to be taken
+ * @return standard error code*/
+static int jbd_iterate_log(struct jbd_fs *jbd_fs,
+ struct recover_info *info,
+ int action)
{
int r = EOK;
bool log_end = false;
uint32_t start_trans_id, this_trans_id;
uint32_t start_block, this_block;
+ /* We start iterating valid blocks in the whole journal.*/
start_trans_id = this_trans_id = jbd_get32(sb, sequence);
start_block = this_block = jbd_get32(sb, start);
while (!log_end) {
struct ext4_block block;
struct jbd_bhdr *header;
+ /* If we are not scanning for the last
+ * valid transaction in the journal,
+ * we will stop when we reach the end of
+ * the journal.*/
if (action != ACTION_SCAN)
if (this_trans_id > info->last_trans_id) {
log_end = true;
break;
header = (struct jbd_bhdr *)block.data;
+ /* This block does not have a valid magic number,
+ * so we have reached the end of the journal.*/
if (jbd_get32(header, magic) != JBD_MAGIC_NUMBER) {
jbd_block_set(jbd_fs, &block);
log_end = true;
continue;
}
+ /* If the transaction id we found is not expected,
+ * we may have reached the end of the journal.
+ *
+ * If we are not scanning the journal, something
+ * bad might have taken place. :-( */
if (jbd_get32(header, sequence) != this_trans_id) {
if (action != ACTION_SCAN)
r = EIO;
ext4_dbg(DEBUG_JBD, "Commit block: %" PRIu32", "
"trans_id: %" PRIu32"\n",
this_block, this_trans_id);
+ /* This is the end of a transaction,
+ * we may now proceed to the next transaction.
+ */
this_trans_id++;
break;
case JBD_REVOKE_BLOCK:
}
ext4_dbg(DEBUG_JBD, "End of journal.\n");
if (r == EOK && action == ACTION_SCAN) {
+ /* We have finished scanning the journal. */
info->start_trans_id = start_trans_id;
if (this_trans_id > start_trans_id)
info->last_trans_id = this_trans_id - 1;
return r;
}
+/**@brief Replay journal.
+ * @param jbd_fs jbd filesystem
+ * @return standard error code*/
int jbd_recover(struct jbd_fs *jbd_fs)
{
int r;
r = jbd_iterate_log(jbd_fs, &info, ACTION_RECOVER);
if (r == EOK) {
+ /* If we successfully replay the journal,
+ * clear EXT4_FINCOM_RECOVER flag on the
+ * ext4 superblock, and set the start of
+ * journal to 0.*/
+ uint32_t features_incompatible =
+ ext4_get32(&jbd_fs->inode_ref.fs->sb,
+ features_incompatible);
jbd_set32(&jbd_fs->sb, start, 0);
+ features_incompatible &= ~EXT4_FINCOM_RECOVER;
+ ext4_set32(&jbd_fs->inode_ref.fs->sb,
+ features_incompatible,
+ features_incompatible);
jbd_fs->dirty = true;
+ r = ext4_sb_write(jbd_fs->inode_ref.fs->bdev,
+ &jbd_fs->inode_ref.fs->sb);
}
jbd_destroy_revoke_tree(&info);
return r;
}
-void jbd_journal_write_sb(struct jbd_journal *journal)
+static void jbd_journal_write_sb(struct jbd_journal *journal)
{
struct jbd_fs *jbd_fs = journal->jbd_fs;
- jbd_set32(&jbd_fs->sb, start, journal->first);
+ jbd_set32(&jbd_fs->sb, start, journal->start);
jbd_set32(&jbd_fs->sb, sequence, journal->trans_id);
jbd_fs->dirty = true;
}
+/**@brief Start accessing the journal.
+ * @param jbd_fs jbd filesystem
+ * @param journal current journal session
+ * @return standard error code*/
int jbd_journal_start(struct jbd_fs *jbd_fs,
struct jbd_journal *journal)
{
+ int r;
+ uint32_t features_incompatible =
+ ext4_get32(&jbd_fs->inode_ref.fs->sb,
+ features_incompatible);
+ features_incompatible |= EXT4_FINCOM_RECOVER;
+ ext4_set32(&jbd_fs->inode_ref.fs->sb,
+ features_incompatible,
+ features_incompatible);
+ r = ext4_sb_write(jbd_fs->inode_ref.fs->bdev,
+ &jbd_fs->inode_ref.fs->sb);
+ if (r != EOK)
+ return r;
+
journal->first = jbd_get32(&jbd_fs->sb, first);
journal->start = journal->first;
journal->last = journal->first;
journal->trans_id = 1;
+ journal->alloc_trans_id = 1;
journal->block_size = jbd_get32(&jbd_fs->sb, blocksize);
TAILQ_INIT(&journal->trans_queue);
+ TAILQ_INIT(&journal->cp_queue);
journal->jbd_fs = jbd_fs;
jbd_journal_write_sb(journal);
return jbd_write_sb(jbd_fs);
}
+/**@brief Stop accessing the journal.
+ * @param journal current journal session
+ * @return standard error code*/
int jbd_journal_stop(struct jbd_journal *journal)
{
+ int r;
+ struct jbd_fs *jbd_fs = journal->jbd_fs;
+ uint32_t features_incompatible;
+
+ /* Commit all the transactions to the journal.*/
+ jbd_journal_commit_all(journal);
+ /* Make sure that journalled content have reached
+ * the disk.*/
+ ext4_block_cache_flush(jbd_fs->inode_ref.fs->bdev);
+
+ features_incompatible =
+ ext4_get32(&jbd_fs->inode_ref.fs->sb,
+ features_incompatible);
+ features_incompatible &= ~EXT4_FINCOM_RECOVER;
+ ext4_set32(&jbd_fs->inode_ref.fs->sb,
+ features_incompatible,
+ features_incompatible);
+ r = ext4_sb_write(jbd_fs->inode_ref.fs->bdev,
+ &jbd_fs->inode_ref.fs->sb);
+ if (r != EOK)
+ return r;
+
journal->start = 0;
journal->trans_id = 0;
jbd_journal_write_sb(journal);
return jbd_write_sb(journal->jbd_fs);
}
-static inline bool jbd_has_enough_space(struct jbd_journal *journal,
- uint32_t blk_cnt)
+/**@brief Allocate a block in the journal.
+ * @param journal current journal session
+ * @param trans transaction
+ * @return allocated block address*/
+static uint32_t jbd_journal_alloc_block(struct jbd_journal *journal,
+ struct jbd_trans *trans)
{
- uint32_t new_last = journal->last + blk_cnt;
- wrap(&journal->jbd_fs->sb, new_last);
- if (new_last >= journal->start)
- return false;
-
- return true;
-}
+ uint32_t start_block;
-static uint32_t jbd_journal_alloc_block(struct jbd_journal *journal)
-{
- uint32_t start_block = journal->last++;
+ start_block = journal->last++;
+ trans->alloc_blocks++;
wrap(&journal->jbd_fs->sb, journal->last);
+
+ /* If there is no space left, flush all journalled
+ * blocks to disk first.*/
+ if (journal->last == journal->start)
+ ext4_block_cache_flush(journal->jbd_fs->inode_ref.fs->bdev);
+
return start_block;
}
+/**@brief Allocate a new transaction
+ * @param journal current journal session
+ * @return transaction allocated*/
struct jbd_trans *
-jbd_journal_new_trans()
+jbd_journal_new_trans(struct jbd_journal *journal)
{
struct jbd_trans *trans = calloc(1, sizeof(struct jbd_trans));
if (!trans)
/* We will assign a trans_id to this transaction,
* once it has been committed.*/
+ trans->journal = journal;
+ trans->error = EOK;
return trans;
}
+static void jbd_trans_end_write(struct ext4_bcache *bc __unused,
+ struct ext4_buf *buf __unused,
+ int res,
+ void *arg);
+
+/**@brief Add block to a transaction and gain
+ * access to it before making any modications.
+ * @param trans transaction
+ * @param block block descriptor
+ * @return standard error code*/
int jbd_trans_add_block(struct jbd_trans *trans,
struct ext4_block *block)
{
- struct jbd_buf *buf = calloc(1, sizeof(struct jbd_buf));
+ struct jbd_buf *buf;
+ struct ext4_fs *fs =
+ trans->journal->jbd_fs->inode_ref.fs;
+
+ /* If the buffer has already been modified, we should
+ * flush dirty data in this buffer to disk.*/
+ if (ext4_bcache_test_flag(block->buf, BC_DIRTY)) {
+ /* XXX: i don't want to know whether the call
+ * succeeds or not. */
+ ext4_block_flush_buf(fs->bdev, block->buf);
+ }
+
+ buf = calloc(1, sizeof(struct jbd_buf));
if (!buf)
return ENOMEM;
buf->trans = trans;
buf->block = *block;
ext4_bcache_inc_ref(block->buf);
- trans->buf_cnt++;
+
+ /* If the content reach the disk, notify us
+ * so that we may do a checkpoint. */
+ block->buf->end_write = jbd_trans_end_write;
+ block->buf->end_write_arg = buf;
+
+ trans->data_cnt++;
LIST_INSERT_HEAD(&trans->buf_list, buf, buf_node);
return EOK;
}
+/**@brief Add block to be revoked to a transaction
+ * @param trans transaction
+ * @param lba logical block address
+ * @return standard error code*/
int jbd_trans_revoke_block(struct jbd_trans *trans,
ext4_fsblk_t lba)
{
return ENOMEM;
rec->lba = lba;
- trans->revoke_cnt++;
LIST_INSERT_HEAD(&trans->revoke_list, rec, revoke_node);
return EOK;
}
-void jbd_journal_abort_trans(struct jbd_journal *journal,
- struct jbd_trans *trans)
+/**@brief Free a transaction
+ * @param journal current journal session
+ * @param trans transaction
+ * @param abort discard all the modifications on the block?
+ * @return standard error code*/
+void jbd_journal_free_trans(struct jbd_journal *journal,
+ struct jbd_trans *trans,
+ bool abort)
{
- struct jbd_buf *var, *tmp;
+ struct jbd_buf *jbd_buf, *tmp;
struct jbd_revoke_rec *rec, *tmp2;
- LIST_FOREACH_SAFE(var, &trans->buf_list, buf_node,
+ struct ext4_fs *fs = journal->jbd_fs->inode_ref.fs;
+ LIST_FOREACH_SAFE(jbd_buf, &trans->buf_list, buf_node,
tmp) {
- ext4_block_set(journal->jbd_fs->bdev, &var->block);
- LIST_REMOVE(var, buf_node);
- free(var);
+ if (abort) {
+ ext4_bcache_clear_dirty(jbd_buf->block.buf);
+ ext4_block_set(fs->bdev, &jbd_buf->block);
+ }
+
+ LIST_REMOVE(jbd_buf, buf_node);
+ free(jbd_buf);
}
LIST_FOREACH_SAFE(rec, &trans->revoke_list, revoke_node,
tmp2) {
free(trans);
}
-static void jbd_journal_end_write()
+/**@brief Write commit block for a transaction
+ * @param trans transaction
+ * @return standard error code*/
+static int jbd_trans_write_commit_block(struct jbd_trans *trans)
{
+ int rc;
+ struct jbd_commit_header *header;
+ uint32_t commit_iblock = 0;
+ struct ext4_block commit_block;
+ struct jbd_journal *journal = trans->journal;
+
+ commit_iblock = jbd_journal_alloc_block(journal, trans);
+ rc = jbd_block_get_noread(journal->jbd_fs,
+ &commit_block, commit_iblock);
+ if (rc != EOK)
+ return rc;
+
+ header = (struct jbd_commit_header *)commit_block.data;
+ jbd_set32(&header->header, magic, JBD_MAGIC_NUMBER);
+ jbd_set32(&header->header, blocktype, JBD_COMMIT_BLOCK);
+ jbd_set32(&header->header, sequence, trans->trans_id);
+
+ ext4_bcache_set_dirty(commit_block.buf);
+ rc = jbd_block_set(journal->jbd_fs, &commit_block);
+ if (rc != EOK)
+ return rc;
+ return EOK;
}
+/**@brief Write descriptor block for a transaction
+ * @param journal current journal session
+ * @param trans transaction
+ * @return standard error code*/
static int jbd_journal_prepare(struct jbd_journal *journal,
struct jbd_trans *trans)
{
uint32_t desc_iblock = 0;
uint32_t data_iblock = 0;
char *tag_start = NULL, *tag_ptr = NULL;
- struct jbd_buf *jbd_buf;
+ struct jbd_buf *jbd_buf, *tmp;
struct ext4_block desc_block, data_block;
+ struct ext4_fs *fs = journal->jbd_fs->inode_ref.fs;
- LIST_FOREACH(jbd_buf, &trans->buf_list, buf_node) {
+ LIST_FOREACH_SAFE(jbd_buf, &trans->buf_list, buf_node, tmp) {
struct tag_info tag_info;
bool uuid_exist = false;
+ if (!ext4_bcache_test_flag(jbd_buf->block.buf,
+ BC_DIRTY)) {
+ /* The buffer has not been modified, just release
+ * that jbd_buf. */
+ ext4_block_set(fs->bdev, &jbd_buf->block);
+ LIST_REMOVE(jbd_buf, buf_node);
+ free(jbd_buf);
+ continue;
+ }
again:
if (!desc_iblock) {
struct jbd_bhdr *bhdr;
- desc_iblock = jbd_journal_alloc_block(journal);
+ desc_iblock = jbd_journal_alloc_block(journal, trans);
rc = jbd_block_get_noread(journal->jbd_fs,
&desc_block, desc_iblock);
- if (!rc)
+ if (rc != EOK)
break;
ext4_bcache_set_dirty(desc_block.buf);
bhdr = (struct jbd_bhdr *)desc_block.data;
- bhdr->magic = JBD_MAGIC_NUMBER;
- bhdr->blocktype = JBD_DESCRIPTOR_BLOCK;
- bhdr->sequence = trans->trans_id;
+ jbd_set32(bhdr, magic, JBD_MAGIC_NUMBER);
+ jbd_set32(bhdr, blocktype, JBD_DESCRIPTOR_BLOCK);
+ jbd_set32(bhdr, sequence, trans->trans_id);
tag_start = (char *)(bhdr + 1);
tag_ptr = tag_start;
uuid_exist = true;
tag_tbl_size = journal->block_size -
sizeof(struct jbd_bhdr);
+
+ if (!trans->start_iblock)
+ trans->start_iblock = desc_iblock;
+
}
tag_info.block = jbd_buf->block.lb_id;
tag_info.uuid_exist = uuid_exist;
- if (i == trans->buf_cnt - 1)
+ if (i == trans->data_cnt - 1)
tag_info.last_tag = true;
if (uuid_exist)
goto again;
}
- data_iblock = jbd_journal_alloc_block(journal);
+ data_iblock = jbd_journal_alloc_block(journal, trans);
rc = jbd_block_get_noread(journal->jbd_fs,
&data_block, data_iblock);
if (rc != EOK)
i++;
}
- if (rc != EOK)
- jbd_journal_abort_trans(journal, trans);
-
- if (desc_iblock)
+ if (rc == EOK && desc_iblock)
jbd_block_set(journal->jbd_fs, &desc_block);
return rc;
}
+/**@brief Write revoke block for a transaction
+ * @param journal current journal session
+ * @param trans transaction
+ * @return standard error code*/
static int
jbd_journal_prepare_revoke(struct jbd_journal *journal,
struct jbd_trans *trans)
int32_t tag_tbl_size;
uint32_t desc_iblock = 0;
char *blocks_entry = NULL;
- struct jbd_revoke_rec *rec;
+ struct jbd_revoke_rec *rec, *tmp;
struct ext4_block desc_block;
struct jbd_revoke_header *header = NULL;
int32_t record_len = 4;
JBD_FEATURE_INCOMPAT_64BIT))
record_len = 8;
- LIST_FOREACH(rec, &trans->revoke_list, revoke_node) {
+ LIST_FOREACH_SAFE(rec, &trans->revoke_list, revoke_node,
+ tmp) {
again:
if (!desc_iblock) {
struct jbd_bhdr *bhdr;
- desc_iblock = jbd_journal_alloc_block(journal);
+ desc_iblock = jbd_journal_alloc_block(journal, trans);
rc = jbd_block_get_noread(journal->jbd_fs,
&desc_block, desc_iblock);
- if (!rc)
+ if (rc != EOK) {
break;
+ }
ext4_bcache_set_dirty(desc_block.buf);
bhdr = (struct jbd_bhdr *)desc_block.data;
- bhdr->magic = JBD_MAGIC_NUMBER;
- bhdr->blocktype = JBD_REVOKE_BLOCK;
- bhdr->sequence = trans->trans_id;
+ jbd_set32(bhdr, magic, JBD_MAGIC_NUMBER);
+ jbd_set32(bhdr, blocktype, JBD_REVOKE_BLOCK);
+ jbd_set32(bhdr, sequence, trans->trans_id);
header = (struct jbd_revoke_header *)bhdr;
blocks_entry = (char *)(header + 1);
tag_tbl_size = journal->block_size -
- sizeof(struct jbd_bhdr);
+ sizeof(struct jbd_revoke_header);
+
+ if (!trans->start_iblock)
+ trans->start_iblock = desc_iblock;
+
}
if (tag_tbl_size < record_len) {
- header->count = journal->block_size - tag_tbl_size;
+ jbd_set32(header, count,
+ journal->block_size - tag_tbl_size);
jbd_block_set(journal->jbd_fs, &desc_block);
desc_iblock = 0;
header = NULL;
i++;
}
- if (rc != EOK)
- jbd_journal_abort_trans(journal, trans);
-
- if (desc_iblock) {
+ if (rc == EOK && desc_iblock) {
if (header != NULL)
- header->count = journal->block_size - tag_tbl_size;
+ jbd_set32(header, count,
+ journal->block_size - tag_tbl_size);
jbd_block_set(journal->jbd_fs, &desc_block);
}
return rc;
}
+/**@brief Submit the transaction to transaction queue.
+ * @param journal current journal session
+ * @param trans transaction*/
void
-jbd_journal_commit_trans(struct jbd_journal *journal,
+jbd_journal_submit_trans(struct jbd_journal *journal,
struct jbd_trans *trans)
{
- trans->trans_id = journal->trans_id++;
TAILQ_INSERT_TAIL(&journal->trans_queue,
trans,
trans_node);
}
+/**@brief Put references of block descriptors in a transaction.
+ * @param journal current journal session
+ * @param trans transaction*/
+void jbd_journal_cp_trans(struct jbd_journal *journal, struct jbd_trans *trans)
+{
+ struct jbd_buf *jbd_buf, *tmp;
+ struct ext4_fs *fs = journal->jbd_fs->inode_ref.fs;
+ LIST_FOREACH_SAFE(jbd_buf, &trans->buf_list, buf_node,
+ tmp) {
+ struct ext4_block block = jbd_buf->block;
+ ext4_block_set(fs->bdev, &block);
+ }
+}
+
+/**@brief Update the start block of the journal when
+ * all the contents in a transaction reach the disk.*/
+static void jbd_trans_end_write(struct ext4_bcache *bc __unused,
+ struct ext4_buf *buf,
+ int res,
+ void *arg)
+{
+ struct jbd_buf *jbd_buf = arg;
+ struct jbd_trans *trans = jbd_buf->trans;
+ struct jbd_journal *journal = trans->journal;
+ bool first_in_queue =
+ trans == TAILQ_FIRST(&journal->cp_queue);
+ if (res != EOK)
+ trans->error = res;
+
+ /* Clear the end_write and end_write_arg fields. */
+ buf->end_write = NULL;
+ buf->end_write_arg = NULL;
+
+ trans->written_cnt++;
+ if (trans->written_cnt == trans->data_cnt) {
+ TAILQ_REMOVE(&journal->cp_queue, trans, trans_node);
+
+ if (first_in_queue) {
+ journal->start = trans->start_iblock +
+ trans->alloc_blocks;
+ wrap(&journal->jbd_fs->sb, journal->start);
+ journal->trans_id = trans->trans_id + 1;
+ }
+ jbd_journal_free_trans(journal, trans, false);
+
+ if (first_in_queue) {
+ while ((trans = TAILQ_FIRST(&journal->cp_queue))) {
+ if (!trans->data_cnt) {
+ TAILQ_REMOVE(&journal->cp_queue,
+ trans,
+ trans_node);
+ journal->start = trans->start_iblock +
+ trans->alloc_blocks;
+ wrap(&journal->jbd_fs->sb, journal->start);
+ journal->trans_id = trans->trans_id + 1;
+ jbd_journal_free_trans(journal,
+ trans, false);
+ } else {
+ journal->start = trans->start_iblock;
+ wrap(&journal->jbd_fs->sb, journal->start);
+ journal->trans_id = trans->trans_id;
+ break;
+ }
+ }
+ jbd_journal_write_sb(journal);
+ jbd_write_sb(journal->jbd_fs);
+ }
+ }
+}
+
+/**@brief Commit a transaction to the journal immediately.
+ * @param journal current journal session
+ * @param trans transaction
+ * @return standard error code*/
+int jbd_journal_commit_trans(struct jbd_journal *journal,
+ struct jbd_trans *trans)
+{
+ int rc = EOK;
+ uint32_t last = journal->last;
+
+ trans->trans_id = journal->alloc_trans_id;
+ rc = jbd_journal_prepare(journal, trans);
+ if (rc != EOK)
+ goto Finish;
+
+ rc = jbd_journal_prepare_revoke(journal, trans);
+ if (rc != EOK)
+ goto Finish;
+
+ if (LIST_EMPTY(&trans->buf_list) &&
+ LIST_EMPTY(&trans->revoke_list)) {
+ /* Since there are no entries in both buffer list
+ * and revoke entry list, we do not consider trans as
+ * complete transaction and just return EOK.*/
+ jbd_journal_free_trans(journal, trans, false);
+ goto Finish;
+ }
+
+ rc = jbd_trans_write_commit_block(trans);
+ if (rc != EOK)
+ goto Finish;
+
+ journal->alloc_trans_id++;
+ if (TAILQ_EMPTY(&journal->cp_queue)) {
+ if (trans->data_cnt) {
+ journal->start = trans->start_iblock;
+ wrap(&journal->jbd_fs->sb, journal->start);
+ journal->trans_id = trans->trans_id;
+ jbd_journal_write_sb(journal);
+ jbd_write_sb(journal->jbd_fs);
+ TAILQ_INSERT_TAIL(&journal->cp_queue, trans,
+ trans_node);
+ jbd_journal_cp_trans(journal, trans);
+ } else {
+ journal->start = trans->start_iblock +
+ trans->alloc_blocks;
+ wrap(&journal->jbd_fs->sb, journal->start);
+ journal->trans_id = trans->trans_id + 1;
+ jbd_journal_write_sb(journal);
+ jbd_journal_free_trans(journal, trans, false);
+ }
+ } else {
+ TAILQ_INSERT_TAIL(&journal->cp_queue, trans,
+ trans_node);
+ if (trans->data_cnt)
+ jbd_journal_cp_trans(journal, trans);
+
+ }
+Finish:
+ if (rc != EOK) {
+ journal->last = last;
+ jbd_journal_free_trans(journal, trans, true);
+ }
+ return rc;
+}
+
+/**@brief Commit one transaction on transaction queue
+ * to the journal.
+ * @param journal current journal session.*/
+void jbd_journal_commit_one(struct jbd_journal *journal)
+{
+ struct jbd_trans *trans;
+
+ if ((trans = TAILQ_FIRST(&journal->trans_queue))) {
+ TAILQ_REMOVE(&journal->trans_queue, trans, trans_node);
+ jbd_journal_commit_trans(journal, trans);
+ }
+}
+
+/**@brief Commit all the transactions on transaction queue
+ * to the journal.
+ * @param journal current journal session.*/
+void jbd_journal_commit_all(struct jbd_journal *journal)
+{
+ while (!TAILQ_EMPTY(&journal->trans_queue)) {
+ jbd_journal_commit_one(journal);
+ }
+}
+
/**
* @}
*/