Change include type policy
[lwext4.git] / src / ext4_journal.c
index a8482604893a0c14cbb4f1d004fc953307720f4b..aa3322b2a9295c0d26fb82437fa0639c410d9e92 100644 (file)
  * @brief Journal handle functions
  */
 
-#include "ext4_config.h"
-#include "ext4_types.h"
-#include "ext4_misc.h"
-#include "ext4_errno.h"
-#include "ext4_debug.h"
-
-#include "ext4_fs.h"
-#include "ext4_super.h"
-#include "ext4_journal.h"
-#include "ext4_blockdev.h"
-#include "ext4_crc32.h"
-#include "ext4_journal.h"
+#include <ext4_config.h>
+#include <ext4_types.h>
+#include <ext4_misc.h>
+#include <ext4_errno.h>
+#include <ext4_debug.h>
+
+#include <ext4_fs.h>
+#include <ext4_super.h>
+#include <ext4_journal.h>
+#include <ext4_blockdev.h>
+#include <ext4_crc32.h>
+#include <ext4_journal.h>
 
 #include <string.h>
 #include <stdlib.h>
@@ -619,6 +619,10 @@ struct tag_info {
        /**@brief  block number stored in this tag.*/
        ext4_fsblk_t block;
 
+       /**@brief  Is the first 4 bytes of block equals to
+        *         JBD_MAGIC_NUMBER? */
+       bool is_escape;
+
        /**@brief  whether UUID part exists or not.*/
        bool uuid_exist;
 
@@ -649,6 +653,7 @@ jbd_extract_block_tag(struct jbd_fs *jbd_fs,
        tag_info->tag_bytes = tag_bytes;
        tag_info->uuid_exist = false;
        tag_info->last_tag = false;
+       tag_info->is_escape = false;
 
        /* See whether it is possible to hold a valid block tag.*/
        if (remain_buf_size - tag_bytes < 0)
@@ -664,7 +669,7 @@ jbd_extract_block_tag(struct jbd_fs *jbd_fs,
                                 (uint64_t)jbd_get32(tag, blocknr_high) << 32;
 
                if (jbd_get32(tag, flags) & JBD_FLAG_ESCAPE)
-                       tag_info->block = 0;
+                       tag_info->is_escape = true;
 
                if (!(jbd_get32(tag, flags) & JBD_FLAG_SAME_UUID)) {
                        /* See whether it is possible to hold UUID part.*/
@@ -689,7 +694,7 @@ jbd_extract_block_tag(struct jbd_fs *jbd_fs,
                                 (uint64_t)jbd_get32(tag, blocknr_high) << 32;
 
                if (jbd_get16(tag, flags) & JBD_FLAG_ESCAPE)
-                       tag_info->block = 0;
+                       tag_info->is_escape = true;
 
                if (!(jbd_get16(tag, flags) & JBD_FLAG_SAME_UUID)) {
                        /* See whether it is possible to hold UUID part.*/
@@ -756,6 +761,10 @@ jbd_write_block_tag(struct jbd_fs *jbd_fs,
                        jbd_set32(tag, flags,
                                  jbd_get32(tag, flags) | JBD_FLAG_LAST_TAG);
 
+               if (tag_info->is_escape)
+                       jbd_set32(tag, flags,
+                                 jbd_get32(tag, flags) | JBD_FLAG_ESCAPE);
+
        } else {
                struct jbd_block_tag *tag = __tag;
                memset(tag, 0, sizeof(struct jbd_block_tag));
@@ -782,6 +791,11 @@ jbd_write_block_tag(struct jbd_fs *jbd_fs,
                        jbd_set16(tag, flags,
                                  jbd_get16(tag, flags) | JBD_FLAG_LAST_TAG);
 
+
+               if (tag_info->is_escape)
+                       jbd_set16(tag, flags,
+                                 jbd_get16(tag, flags) | JBD_FLAG_ESCAPE);
+
        }
        return EOK;
 }
@@ -798,9 +812,8 @@ jbd_iterate_block_table(struct jbd_fs *jbd_fs,
                        void *__tag_start,
                        int32_t tag_tbl_size,
                        void (*func)(struct jbd_fs * jbd_fs,
-                                       ext4_fsblk_t block,
-                                       uint8_t *uuid,
-                                       void *arg),
+                                    struct tag_info *tag_info,
+                                    void *arg),
                        void *arg)
 {
        char *tag_start, *tag_ptr;
@@ -826,7 +839,7 @@ jbd_iterate_block_table(struct jbd_fs *jbd_fs,
                        break;
 
                if (func)
-                       func(jbd_fs, tag_info.block, tag_info.uuid, arg);
+                       func(jbd_fs, &tag_info, arg);
 
                /* Stop the iteration when we reach the last tag. */
                if (tag_info.last_tag)
@@ -838,16 +851,14 @@ jbd_iterate_block_table(struct jbd_fs *jbd_fs,
 }
 
 static void jbd_display_block_tags(struct jbd_fs *jbd_fs,
-                                  ext4_fsblk_t block,
-                                  uint8_t *uuid,
+                                  struct tag_info *tag_info,
                                   void *arg)
 {
        uint32_t *iblock = arg;
-       ext4_dbg(DEBUG_JBD, "Block in block_tag: %" PRIu64 "\n", block);
+       ext4_dbg(DEBUG_JBD, "Block in block_tag: %" PRIu64 "\n", tag_info->block);
        (*iblock)++;
        wrap(&jbd_fs->sb, *iblock);
        (void)jbd_fs;
-       (void)uuid;
        return;
 }
 
@@ -863,10 +874,9 @@ jbd_revoke_entry_lookup(struct recover_info *info, ext4_fsblk_t block)
 
 /**@brief  Replay a block in a transaction.
  * @param  jbd_fs jbd filesystem
- * @param  block  block address to be replayed.*/
+ * @param  tag_info tag_info of the logged block.*/
 static void jbd_replay_block_tags(struct jbd_fs *jbd_fs,
-                                 ext4_fsblk_t block,
-                                 uint8_t *uuid __unused,
+                                 struct tag_info *tag_info,
                                  void *__arg)
 {
        int r;
@@ -882,22 +892,22 @@ static void jbd_replay_block_tags(struct jbd_fs *jbd_fs,
 
        /* 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);
+       revoke_entry = jbd_revoke_entry_lookup(info, tag_info->block);
        if (revoke_entry &&
            trans_id_diff(arg->this_trans_id, revoke_entry->trans_id) <= 0)
                return;
 
        ext4_dbg(DEBUG_JBD,
                 "Replaying block in block_tag: %" PRIu64 "\n",
-                block);
+                tag_info->block);
 
        r = jbd_block_get(jbd_fs, &journal_block, *this_block);
        if (r != EOK)
                return;
 
        /* We need special treatment for ext4 superblock. */
-       if (block) {
-               r = ext4_block_get_noread(fs->bdev, &ext4_block, block);
+       if (tag_info->block) {
+               r = ext4_block_get_noread(fs->bdev, &ext4_block, tag_info->block);
                if (r != EOK) {
                        jbd_block_set(jbd_fs, &journal_block);
                        return;
@@ -907,6 +917,10 @@ static void jbd_replay_block_tags(struct jbd_fs *jbd_fs,
                        journal_block.data,
                        jbd_get32(&jbd_fs->sb, blocksize));
 
+               if (tag_info->is_escape)
+                       ((struct jbd_bhdr *)ext4_block.data)->magic =
+                                       to_be32(JBD_MAGIC_NUMBER);
+
                ext4_bcache_set_dirty(ext4_block.buf);
                ext4_block_set(fs->bdev, &ext4_block);
        } else {
@@ -1143,11 +1157,13 @@ static int jbd_iterate_log(struct jbd_fs *jbd_fs,
                        ext4_dbg(DEBUG_JBD, "Commit block: %" PRIu32", "
                                            "trans_id: %" PRIu32"\n",
                                            this_block, this_trans_id);
-                       /* This is the end of a transaction,
+                       /*
+                        * This is the end of a transaction,
                         * we may now proceed to the next transaction.
                         */
                        this_trans_id++;
-                       info->trans_cnt++;
+                       if (action == ACTION_SCAN)
+                               info->trans_cnt++;
                        break;
                case JBD_REVOKE_BLOCK:
                        if (!jbd_verify_meta_csum(jbd_fs, header)) {
@@ -1222,6 +1238,7 @@ int jbd_recover(struct jbd_fs *jbd_fs)
                        ext4_get32(&jbd_fs->inode_ref.fs->sb,
                                   features_incompatible);
                jbd_set32(&jbd_fs->sb, start, 0);
+               jbd_set32(&jbd_fs->sb, sequence, info.last_trans_id);
                features_incompatible &= ~EXT4_FINCOM_RECOVER;
                ext4_set32(&jbd_fs->inode_ref.fs->sb,
                           features_incompatible,
@@ -1253,7 +1270,6 @@ int jbd_journal_start(struct jbd_fs *jbd_fs,
        uint32_t features_incompatible =
                        ext4_get32(&jbd_fs->inode_ref.fs->sb,
                                   features_incompatible);
-       struct ext4_block block = EXT4_BLOCK_ZERO();
        features_incompatible |= EXT4_FINCOM_RECOVER;
        ext4_set32(&jbd_fs->inode_ref.fs->sb,
                        features_incompatible,
@@ -1266,26 +1282,16 @@ int jbd_journal_start(struct jbd_fs *jbd_fs,
        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;
+       /*
+        * To invalidate any stale records we need to start from
+        * the checkpoint transaction ID of the previous journalling session
+        * plus 1.
+        */
+       journal->trans_id = jbd_get32(&jbd_fs->sb, sequence) + 1;
+       journal->alloc_trans_id = journal->trans_id;
 
        journal->block_size = jbd_get32(&jbd_fs->sb, blocksize);
 
-       r = jbd_block_get_noread(jbd_fs,
-                        &block,
-                        journal->start);
-       if (r != EOK) {
-               memset(journal, 0, sizeof(struct jbd_journal));
-               return r;
-       }
-       memset(block.data, 0, journal->block_size);
-       ext4_bcache_set_dirty(block.buf);
-       r = jbd_block_set(jbd_fs, &block);
-       if (r != EOK) {
-               memset(journal, 0, sizeof(struct jbd_journal));
-               return r;
-       }
-
        TAILQ_INIT(&journal->cp_queue);
        RB_INIT(&journal->block_rec_root);
        journal->jbd_fs = jbd_fs;
@@ -1453,10 +1459,12 @@ static uint32_t jbd_journal_alloc_block(struct jbd_journal *journal,
        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)
-               jbd_journal_purge_cp_trans(journal, true, false);
+       /* If there is no space left, flush just one journalled
+        * transaction.*/
+       if (journal->last == journal->start) {
+               jbd_journal_purge_cp_trans(journal, true, true);
+               ext4_assert(journal->last != journal->start);
+       }
 
        return start_block;
 }
@@ -1759,29 +1767,12 @@ static int jbd_trans_write_commit_block(struct jbd_trans *trans)
        int rc;
        struct ext4_block block;
        struct jbd_commit_header *header;
-       uint32_t commit_iblock, orig_commit_iblock;
+       uint32_t commit_iblock;
        struct jbd_journal *journal = trans->journal;
 
        commit_iblock = jbd_journal_alloc_block(journal, trans);
-       orig_commit_iblock = commit_iblock;
-       commit_iblock++;
-       wrap(&journal->jbd_fs->sb, commit_iblock);
-
-       /* To prevent accidental reference to stale journalling metadata. */
-       if (orig_commit_iblock < commit_iblock) {
-               rc = jbd_block_get_noread(journal->jbd_fs, &block, commit_iblock);
-               if (rc != EOK)
-                       return rc;
 
-               memset(block.data, 0, journal->block_size);
-               ext4_bcache_set_dirty(block.buf);
-               ext4_bcache_set_flag(block.buf, BC_TMP);
-               rc = jbd_block_set(journal->jbd_fs, &block);
-               if (rc != EOK)
-                       return rc;
-       }
-
-       rc = jbd_block_get_noread(journal->jbd_fs, &block, orig_commit_iblock);
+       rc = jbd_block_get_noread(journal->jbd_fs, &block, commit_iblock);
        if (rc != EOK)
                return rc;
 
@@ -1860,6 +1851,7 @@ static int jbd_journal_prepare(struct jbd_journal *journal,
        TAILQ_FOREACH_SAFE(jbd_buf, &trans->buf_queue, buf_node, tmp) {
                struct tag_info tag_info;
                bool uuid_exist = false;
+               bool is_escape = false;
                struct jbd_revoke_rec tmp_rec = {
                        .lba = jbd_buf->block_rec->lba
                };
@@ -1894,6 +1886,10 @@ static int jbd_journal_prepare(struct jbd_journal *journal,
                                          jbd_buf->block.data,
                                          checksum,
                                          trans->trans_id);
+               if (((struct jbd_bhdr *)jbd_buf->block.data)->magic ==
+                               to_be32(JBD_MAGIC_NUMBER))
+                       is_escape = true;
+
 again:
                if (!desc_iblock) {
                        desc_iblock = jbd_journal_alloc_block(journal, trans);
@@ -1923,6 +1919,7 @@ again:
                }
                tag_info.block = jbd_buf->block.lb_id;
                tag_info.uuid_exist = uuid_exist;
+               tag_info.is_escape = is_escape;
                if (i == trans->data_cnt - 1)
                        tag_info.last_tag = true;
                else
@@ -1960,6 +1957,9 @@ again:
                data = data_block.data;
                memcpy(data, jbd_buf->block.data,
                        journal->block_size);
+               if (is_escape)
+                       ((struct jbd_bhdr *)data)->magic = 0;
+
                ext4_bcache_set_dirty(data_block.buf);
                ext4_bcache_set_flag(data_block.buf, BC_TMP);
                rc = jbd_block_set(journal->jbd_fs, &data_block);