|
@@ -1,38 +1,38 @@
|
|
|
/*
|
|
/*
|
|
|
-** Write ahead logging implementation copyright Chris Mason 2000
|
|
|
|
|
-**
|
|
|
|
|
-** The background commits make this code very interrelated, and
|
|
|
|
|
-** overly complex. I need to rethink things a bit....The major players:
|
|
|
|
|
-**
|
|
|
|
|
-** journal_begin -- call with the number of blocks you expect to log.
|
|
|
|
|
-** If the current transaction is too
|
|
|
|
|
-** old, it will block until the current transaction is
|
|
|
|
|
-** finished, and then start a new one.
|
|
|
|
|
-** Usually, your transaction will get joined in with
|
|
|
|
|
-** previous ones for speed.
|
|
|
|
|
-**
|
|
|
|
|
-** journal_join -- same as journal_begin, but won't block on the current
|
|
|
|
|
-** transaction regardless of age. Don't ever call
|
|
|
|
|
-** this. Ever. There are only two places it should be
|
|
|
|
|
-** called from, and they are both inside this file.
|
|
|
|
|
-**
|
|
|
|
|
-** journal_mark_dirty -- adds blocks into this transaction. clears any flags
|
|
|
|
|
-** that might make them get sent to disk
|
|
|
|
|
-** and then marks them BH_JDirty. Puts the buffer head
|
|
|
|
|
-** into the current transaction hash.
|
|
|
|
|
-**
|
|
|
|
|
-** journal_end -- if the current transaction is batchable, it does nothing
|
|
|
|
|
-** otherwise, it could do an async/synchronous commit, or
|
|
|
|
|
-** a full flush of all log and real blocks in the
|
|
|
|
|
-** transaction.
|
|
|
|
|
-**
|
|
|
|
|
-** flush_old_commits -- if the current transaction is too old, it is ended and
|
|
|
|
|
-** commit blocks are sent to disk. Forces commit blocks
|
|
|
|
|
-** to disk for all backgrounded commits that have been
|
|
|
|
|
-** around too long.
|
|
|
|
|
-** -- Note, if you call this as an immediate flush from
|
|
|
|
|
-** from within kupdate, it will ignore the immediate flag
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * Write ahead logging implementation copyright Chris Mason 2000
|
|
|
|
|
+ *
|
|
|
|
|
+ * The background commits make this code very interrelated, and
|
|
|
|
|
+ * overly complex. I need to rethink things a bit....The major players:
|
|
|
|
|
+ *
|
|
|
|
|
+ * journal_begin -- call with the number of blocks you expect to log.
|
|
|
|
|
+ * If the current transaction is too
|
|
|
|
|
+ * old, it will block until the current transaction is
|
|
|
|
|
+ * finished, and then start a new one.
|
|
|
|
|
+ * Usually, your transaction will get joined in with
|
|
|
|
|
+ * previous ones for speed.
|
|
|
|
|
+ *
|
|
|
|
|
+ * journal_join -- same as journal_begin, but won't block on the current
|
|
|
|
|
+ * transaction regardless of age. Don't ever call
|
|
|
|
|
+ * this. Ever. There are only two places it should be
|
|
|
|
|
+ * called from, and they are both inside this file.
|
|
|
|
|
+ *
|
|
|
|
|
+ * journal_mark_dirty -- adds blocks into this transaction. clears any flags
|
|
|
|
|
+ * that might make them get sent to disk
|
|
|
|
|
+ * and then marks them BH_JDirty. Puts the buffer head
|
|
|
|
|
+ * into the current transaction hash.
|
|
|
|
|
+ *
|
|
|
|
|
+ * journal_end -- if the current transaction is batchable, it does nothing
|
|
|
|
|
+ * otherwise, it could do an async/synchronous commit, or
|
|
|
|
|
+ * a full flush of all log and real blocks in the
|
|
|
|
|
+ * transaction.
|
|
|
|
|
+ *
|
|
|
|
|
+ * flush_old_commits -- if the current transaction is too old, it is ended and
|
|
|
|
|
+ * commit blocks are sent to disk. Forces commit blocks
|
|
|
|
|
+ * to disk for all backgrounded commits that have been
|
|
|
|
|
+ * around too long.
|
|
|
|
|
+ * -- Note, if you call this as an immediate flush from
|
|
|
|
|
+ * from within kupdate, it will ignore the immediate flag
|
|
|
|
|
+ */
|
|
|
|
|
|
|
|
#include <linux/time.h>
|
|
#include <linux/time.h>
|
|
|
#include <linux/semaphore.h>
|
|
#include <linux/semaphore.h>
|
|
@@ -58,23 +58,19 @@
|
|
|
#define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
|
|
#define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
|
|
|
j_working_list))
|
|
j_working_list))
|
|
|
|
|
|
|
|
-/* the number of mounted filesystems. This is used to decide when to
|
|
|
|
|
-** start and kill the commit workqueue
|
|
|
|
|
-*/
|
|
|
|
|
-static int reiserfs_mounted_fs_count;
|
|
|
|
|
-
|
|
|
|
|
-static struct workqueue_struct *commit_wq;
|
|
|
|
|
-
|
|
|
|
|
-#define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
|
|
|
|
|
- structs at 4k */
|
|
|
|
|
|
|
+/* must be correct to keep the desc and commit structs at 4k */
|
|
|
|
|
+#define JOURNAL_TRANS_HALF 1018
|
|
|
#define BUFNR 64 /*read ahead */
|
|
#define BUFNR 64 /*read ahead */
|
|
|
|
|
|
|
|
/* cnode stat bits. Move these into reiserfs_fs.h */
|
|
/* cnode stat bits. Move these into reiserfs_fs.h */
|
|
|
|
|
|
|
|
-#define BLOCK_FREED 2 /* this block was freed, and can't be written. */
|
|
|
|
|
-#define BLOCK_FREED_HOLDER 3 /* this block was freed during this transaction, and can't be written */
|
|
|
|
|
|
|
+/* this block was freed, and can't be written. */
|
|
|
|
|
+#define BLOCK_FREED 2
|
|
|
|
|
+/* this block was freed during this transaction, and can't be written */
|
|
|
|
|
+#define BLOCK_FREED_HOLDER 3
|
|
|
|
|
|
|
|
-#define BLOCK_NEEDS_FLUSH 4 /* used in flush_journal_list */
|
|
|
|
|
|
|
+/* used in flush_journal_list */
|
|
|
|
|
+#define BLOCK_NEEDS_FLUSH 4
|
|
|
#define BLOCK_DIRTIED 5
|
|
#define BLOCK_DIRTIED 5
|
|
|
|
|
|
|
|
/* journal list state bits */
|
|
/* journal list state bits */
|
|
@@ -87,16 +83,14 @@ static struct workqueue_struct *commit_wq;
|
|
|
#define COMMIT_NOW 2 /* end and commit this transaction */
|
|
#define COMMIT_NOW 2 /* end and commit this transaction */
|
|
|
#define WAIT 4 /* wait for the log blocks to hit the disk */
|
|
#define WAIT 4 /* wait for the log blocks to hit the disk */
|
|
|
|
|
|
|
|
-static int do_journal_end(struct reiserfs_transaction_handle *,
|
|
|
|
|
- struct super_block *, unsigned long nblocks,
|
|
|
|
|
- int flags);
|
|
|
|
|
|
|
+static int do_journal_end(struct reiserfs_transaction_handle *, int flags);
|
|
|
static int flush_journal_list(struct super_block *s,
|
|
static int flush_journal_list(struct super_block *s,
|
|
|
struct reiserfs_journal_list *jl, int flushall);
|
|
struct reiserfs_journal_list *jl, int flushall);
|
|
|
static int flush_commit_list(struct super_block *s,
|
|
static int flush_commit_list(struct super_block *s,
|
|
|
struct reiserfs_journal_list *jl, int flushall);
|
|
struct reiserfs_journal_list *jl, int flushall);
|
|
|
static int can_dirty(struct reiserfs_journal_cnode *cn);
|
|
static int can_dirty(struct reiserfs_journal_cnode *cn);
|
|
|
static int journal_join(struct reiserfs_transaction_handle *th,
|
|
static int journal_join(struct reiserfs_transaction_handle *th,
|
|
|
- struct super_block *sb, unsigned long nblocks);
|
|
|
|
|
|
|
+ struct super_block *sb);
|
|
|
static void release_journal_dev(struct super_block *super,
|
|
static void release_journal_dev(struct super_block *super,
|
|
|
struct reiserfs_journal *journal);
|
|
struct reiserfs_journal *journal);
|
|
|
static int dirty_one_transaction(struct super_block *s,
|
|
static int dirty_one_transaction(struct super_block *s,
|
|
@@ -107,8 +101,10 @@ static void queue_log_writer(struct super_block *s);
|
|
|
/* values for join in do_journal_begin_r */
|
|
/* values for join in do_journal_begin_r */
|
|
|
enum {
|
|
enum {
|
|
|
JBEGIN_REG = 0, /* regular journal begin */
|
|
JBEGIN_REG = 0, /* regular journal begin */
|
|
|
- JBEGIN_JOIN = 1, /* join the running transaction if at all possible */
|
|
|
|
|
- JBEGIN_ABORT = 2, /* called from cleanup code, ignores aborted flag */
|
|
|
|
|
|
|
+ /* join the running transaction if at all possible */
|
|
|
|
|
+ JBEGIN_JOIN = 1,
|
|
|
|
|
+ /* called from cleanup code, ignores aborted flag */
|
|
|
|
|
+ JBEGIN_ABORT = 2,
|
|
|
};
|
|
};
|
|
|
|
|
|
|
|
static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
|
|
static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
|
|
@@ -123,10 +119,11 @@ static void init_journal_hash(struct super_block *sb)
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** clears BH_Dirty and sticks the buffer on the clean list. Called because I can't allow refile_buffer to
|
|
|
|
|
-** make schedule happen after I've freed a block. Look at remove_from_transaction and journal_mark_freed for
|
|
|
|
|
-** more details.
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * clears BH_Dirty and sticks the buffer on the clean list. Called because
|
|
|
|
|
+ * I can't allow refile_buffer to make schedule happen after I've freed a
|
|
|
|
|
+ * block. Look at remove_from_transaction and journal_mark_freed for
|
|
|
|
|
+ * more details.
|
|
|
|
|
+ */
|
|
|
static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
|
|
static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
|
|
|
{
|
|
{
|
|
|
if (bh) {
|
|
if (bh) {
|
|
@@ -163,7 +160,7 @@ static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
|
|
|
struct list_head *entry = journal->j_bitmap_nodes.next;
|
|
struct list_head *entry = journal->j_bitmap_nodes.next;
|
|
|
|
|
|
|
|
journal->j_used_bitmap_nodes++;
|
|
journal->j_used_bitmap_nodes++;
|
|
|
- repeat:
|
|
|
|
|
|
|
+repeat:
|
|
|
|
|
|
|
|
if (entry != &journal->j_bitmap_nodes) {
|
|
if (entry != &journal->j_bitmap_nodes) {
|
|
|
bn = list_entry(entry, struct reiserfs_bitmap_node, list);
|
|
bn = list_entry(entry, struct reiserfs_bitmap_node, list);
|
|
@@ -204,7 +201,8 @@ static void allocate_bitmap_nodes(struct super_block *sb)
|
|
|
list_add(&bn->list, &journal->j_bitmap_nodes);
|
|
list_add(&bn->list, &journal->j_bitmap_nodes);
|
|
|
journal->j_free_bitmap_nodes++;
|
|
journal->j_free_bitmap_nodes++;
|
|
|
} else {
|
|
} else {
|
|
|
- break; /* this is ok, we'll try again when more are needed */
|
|
|
|
|
|
|
+ /* this is ok, we'll try again when more are needed */
|
|
|
|
|
+ break;
|
|
|
}
|
|
}
|
|
|
}
|
|
}
|
|
|
}
|
|
}
|
|
@@ -239,8 +237,8 @@ static void cleanup_bitmap_list(struct super_block *sb,
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** only call this on FS unmount.
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * only call this on FS unmount.
|
|
|
|
|
+ */
|
|
|
static int free_list_bitmaps(struct super_block *sb,
|
|
static int free_list_bitmaps(struct super_block *sb,
|
|
|
struct reiserfs_list_bitmap *jb_array)
|
|
struct reiserfs_list_bitmap *jb_array)
|
|
|
{
|
|
{
|
|
@@ -275,9 +273,9 @@ static int free_bitmap_nodes(struct super_block *sb)
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
|
|
|
|
|
-** jb_array is the array to be filled in.
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
|
|
|
|
|
+ * jb_array is the array to be filled in.
|
|
|
|
|
+ */
|
|
|
int reiserfs_allocate_list_bitmaps(struct super_block *sb,
|
|
int reiserfs_allocate_list_bitmaps(struct super_block *sb,
|
|
|
struct reiserfs_list_bitmap *jb_array,
|
|
struct reiserfs_list_bitmap *jb_array,
|
|
|
unsigned int bmap_nr)
|
|
unsigned int bmap_nr)
|
|
@@ -306,9 +304,9 @@ int reiserfs_allocate_list_bitmaps(struct super_block *sb,
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** find an available list bitmap. If you can't find one, flush a commit list
|
|
|
|
|
-** and try again
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * find an available list bitmap. If you can't find one, flush a commit list
|
|
|
|
|
+ * and try again
|
|
|
|
|
+ */
|
|
|
static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
|
|
static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
|
|
|
struct reiserfs_journal_list
|
|
struct reiserfs_journal_list
|
|
|
*jl)
|
|
*jl)
|
|
@@ -332,18 +330,18 @@ static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
|
|
|
break;
|
|
break;
|
|
|
}
|
|
}
|
|
|
}
|
|
}
|
|
|
- if (jb->journal_list) { /* double check to make sure if flushed correctly */
|
|
|
|
|
|
|
+ /* double check to make sure if flushed correctly */
|
|
|
|
|
+ if (jb->journal_list)
|
|
|
return NULL;
|
|
return NULL;
|
|
|
- }
|
|
|
|
|
jb->journal_list = jl;
|
|
jb->journal_list = jl;
|
|
|
return jb;
|
|
return jb;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** allocates a new chunk of X nodes, and links them all together as a list.
|
|
|
|
|
-** Uses the cnode->next and cnode->prev pointers
|
|
|
|
|
-** returns NULL on failure
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * allocates a new chunk of X nodes, and links them all together as a list.
|
|
|
|
|
+ * Uses the cnode->next and cnode->prev pointers
|
|
|
|
|
+ * returns NULL on failure
|
|
|
|
|
+ */
|
|
|
static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
|
|
static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
|
|
|
{
|
|
{
|
|
|
struct reiserfs_journal_cnode *head;
|
|
struct reiserfs_journal_cnode *head;
|
|
@@ -365,9 +363,7 @@ static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
|
|
|
return head;
|
|
return head;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/*
|
|
|
|
|
-** pulls a cnode off the free list, or returns NULL on failure
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+/* pulls a cnode off the free list, or returns NULL on failure */
|
|
|
static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
|
|
static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
|
|
|
{
|
|
{
|
|
|
struct reiserfs_journal_cnode *cn;
|
|
struct reiserfs_journal_cnode *cn;
|
|
@@ -393,8 +389,8 @@ static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** returns a cnode to the free list
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * returns a cnode to the free list
|
|
|
|
|
+ */
|
|
|
static void free_cnode(struct super_block *sb,
|
|
static void free_cnode(struct super_block *sb,
|
|
|
struct reiserfs_journal_cnode *cn)
|
|
struct reiserfs_journal_cnode *cn)
|
|
|
{
|
|
{
|
|
@@ -419,7 +415,10 @@ static void clear_prepared_bits(struct buffer_head *bh)
|
|
|
clear_buffer_journal_restore_dirty(bh);
|
|
clear_buffer_journal_restore_dirty(bh);
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/* return a cnode with same dev, block number and size in table, or null if not found */
|
|
|
|
|
|
|
+/*
|
|
|
|
|
+ * return a cnode with same dev, block number and size in table,
|
|
|
|
|
+ * or null if not found
|
|
|
|
|
+ */
|
|
|
static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
|
|
static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
|
|
|
super_block
|
|
super_block
|
|
|
*sb,
|
|
*sb,
|
|
@@ -439,23 +438,24 @@ static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated
|
|
|
|
|
-** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
|
|
|
|
|
-** being overwritten by a replay after crashing.
|
|
|
|
|
-**
|
|
|
|
|
-** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting
|
|
|
|
|
-** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make
|
|
|
|
|
-** sure you never write the block without logging it.
|
|
|
|
|
-**
|
|
|
|
|
-** next_zero_bit is a suggestion about the next block to try for find_forward.
|
|
|
|
|
-** when bl is rejected because it is set in a journal list bitmap, we search
|
|
|
|
|
-** for the next zero bit in the bitmap that rejected bl. Then, we return that
|
|
|
|
|
-** through next_zero_bit for find_forward to try.
|
|
|
|
|
-**
|
|
|
|
|
-** Just because we return something in next_zero_bit does not mean we won't
|
|
|
|
|
-** reject it on the next call to reiserfs_in_journal
|
|
|
|
|
-**
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * this actually means 'can this block be reallocated yet?'. If you set
|
|
|
|
|
+ * search_all, a block can only be allocated if it is not in the current
|
|
|
|
|
+ * transaction, was not freed by the current transaction, and has no chance
|
|
|
|
|
+ * of ever being overwritten by a replay after crashing.
|
|
|
|
|
+ *
|
|
|
|
|
+ * If you don't set search_all, a block can only be allocated if it is not
|
|
|
|
|
+ * in the current transaction. Since deleting a block removes it from the
|
|
|
|
|
+ * current transaction, this case should never happen. If you don't set
|
|
|
|
|
+ * search_all, make sure you never write the block without logging it.
|
|
|
|
|
+ *
|
|
|
|
|
+ * next_zero_bit is a suggestion about the next block to try for find_forward.
|
|
|
|
|
+ * when bl is rejected because it is set in a journal list bitmap, we search
|
|
|
|
|
+ * for the next zero bit in the bitmap that rejected bl. Then, we return
|
|
|
|
|
+ * that through next_zero_bit for find_forward to try.
|
|
|
|
|
+ *
|
|
|
|
|
+ * Just because we return something in next_zero_bit does not mean we won't
|
|
|
|
|
+ * reject it on the next call to reiserfs_in_journal
|
|
|
|
|
+ */
|
|
|
int reiserfs_in_journal(struct super_block *sb,
|
|
int reiserfs_in_journal(struct super_block *sb,
|
|
|
unsigned int bmap_nr, int bit_nr, int search_all,
|
|
unsigned int bmap_nr, int bit_nr, int search_all,
|
|
|
b_blocknr_t * next_zero_bit)
|
|
b_blocknr_t * next_zero_bit)
|
|
@@ -469,9 +469,11 @@ int reiserfs_in_journal(struct super_block *sb,
|
|
|
*next_zero_bit = 0; /* always start this at zero. */
|
|
*next_zero_bit = 0; /* always start this at zero. */
|
|
|
|
|
|
|
|
PROC_INFO_INC(sb, journal.in_journal);
|
|
PROC_INFO_INC(sb, journal.in_journal);
|
|
|
- /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
|
|
|
|
|
- ** if we crash before the transaction that freed it commits, this transaction won't
|
|
|
|
|
- ** have committed either, and the block will never be written
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * If we aren't doing a search_all, this is a metablock, and it
|
|
|
|
|
+ * will be logged before use. if we crash before the transaction
|
|
|
|
|
+ * that freed it commits, this transaction won't have committed
|
|
|
|
|
+ * either, and the block will never be written
|
|
|
*/
|
|
*/
|
|
|
if (search_all) {
|
|
if (search_all) {
|
|
|
for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
|
|
for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
|
|
@@ -511,8 +513,7 @@ int reiserfs_in_journal(struct super_block *sb,
|
|
|
return 0;
|
|
return 0;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/* insert cn into table
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+/* insert cn into table */
|
|
|
static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
|
|
static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
|
|
|
struct reiserfs_journal_cnode *cn)
|
|
struct reiserfs_journal_cnode *cn)
|
|
|
{
|
|
{
|
|
@@ -558,10 +559,10 @@ static inline void put_journal_list(struct super_block *s,
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** this used to be much more involved, and I'm keeping it just in case things get ugly again.
|
|
|
|
|
-** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
|
|
|
|
|
-** transaction.
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * this used to be much more involved, and I'm keeping it just in case
|
|
|
|
|
+ * things get ugly again. it gets called by flush_commit_list, and
|
|
|
|
|
+ * cleans up any data stored about blocks freed during a transaction.
|
|
|
|
|
+ */
|
|
|
static void cleanup_freed_for_journal_list(struct super_block *sb,
|
|
static void cleanup_freed_for_journal_list(struct super_block *sb,
|
|
|
struct reiserfs_journal_list *jl)
|
|
struct reiserfs_journal_list *jl)
|
|
|
{
|
|
{
|
|
@@ -756,11 +757,12 @@ static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
|
|
|
jh = bh->b_private;
|
|
jh = bh->b_private;
|
|
|
list_del_init(&jh->list);
|
|
list_del_init(&jh->list);
|
|
|
} else {
|
|
} else {
|
|
|
- no_jh:
|
|
|
|
|
|
|
+no_jh:
|
|
|
get_bh(bh);
|
|
get_bh(bh);
|
|
|
jh = alloc_jh();
|
|
jh = alloc_jh();
|
|
|
spin_lock(&j->j_dirty_buffers_lock);
|
|
spin_lock(&j->j_dirty_buffers_lock);
|
|
|
- /* buffer must be locked for __add_jh, should be able to have
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * buffer must be locked for __add_jh, should be able to have
|
|
|
* two adds at the same time
|
|
* two adds at the same time
|
|
|
*/
|
|
*/
|
|
|
BUG_ON(bh->b_private);
|
|
BUG_ON(bh->b_private);
|
|
@@ -818,7 +820,8 @@ static int write_ordered_buffers(spinlock_t * lock,
|
|
|
spin_lock(lock);
|
|
spin_lock(lock);
|
|
|
goto loop_next;
|
|
goto loop_next;
|
|
|
}
|
|
}
|
|
|
- /* in theory, dirty non-uptodate buffers should never get here,
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * in theory, dirty non-uptodate buffers should never get here,
|
|
|
* but the upper layer io error paths still have a few quirks.
|
|
* but the upper layer io error paths still have a few quirks.
|
|
|
* Handle them here as gracefully as we can
|
|
* Handle them here as gracefully as we can
|
|
|
*/
|
|
*/
|
|
@@ -833,7 +836,7 @@ static int write_ordered_buffers(spinlock_t * lock,
|
|
|
reiserfs_free_jh(bh);
|
|
reiserfs_free_jh(bh);
|
|
|
unlock_buffer(bh);
|
|
unlock_buffer(bh);
|
|
|
}
|
|
}
|
|
|
- loop_next:
|
|
|
|
|
|
|
+loop_next:
|
|
|
put_bh(bh);
|
|
put_bh(bh);
|
|
|
cond_resched_lock(lock);
|
|
cond_resched_lock(lock);
|
|
|
}
|
|
}
|
|
@@ -856,13 +859,14 @@ static int write_ordered_buffers(spinlock_t * lock,
|
|
|
if (!buffer_uptodate(bh)) {
|
|
if (!buffer_uptodate(bh)) {
|
|
|
ret = -EIO;
|
|
ret = -EIO;
|
|
|
}
|
|
}
|
|
|
- /* ugly interaction with invalidatepage here.
|
|
|
|
|
- * reiserfs_invalidate_page will pin any buffer that has a valid
|
|
|
|
|
- * journal head from an older transaction. If someone else sets
|
|
|
|
|
- * our buffer dirty after we write it in the first loop, and
|
|
|
|
|
- * then someone truncates the page away, nobody will ever write
|
|
|
|
|
- * the buffer. We're safe if we write the page one last time
|
|
|
|
|
- * after freeing the journal header.
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * ugly interaction with invalidatepage here.
|
|
|
|
|
+ * reiserfs_invalidate_page will pin any buffer that has a
|
|
|
|
|
+ * valid journal head from an older transaction. If someone
|
|
|
|
|
+ * else sets our buffer dirty after we write it in the first
|
|
|
|
|
+ * loop, and then someone truncates the page away, nobody
|
|
|
|
|
+ * will ever write the buffer. We're safe if we write the
|
|
|
|
|
+ * page one last time after freeing the journal header.
|
|
|
*/
|
|
*/
|
|
|
if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
|
|
if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
|
|
|
spin_unlock(lock);
|
|
spin_unlock(lock);
|
|
@@ -887,7 +891,7 @@ static int flush_older_commits(struct super_block *s,
|
|
|
unsigned int other_trans_id;
|
|
unsigned int other_trans_id;
|
|
|
unsigned int first_trans_id;
|
|
unsigned int first_trans_id;
|
|
|
|
|
|
|
|
- find_first:
|
|
|
|
|
|
|
+find_first:
|
|
|
/*
|
|
/*
|
|
|
* first we walk backwards to find the oldest uncommitted transation
|
|
* first we walk backwards to find the oldest uncommitted transation
|
|
|
*/
|
|
*/
|
|
@@ -923,9 +927,11 @@ static int flush_older_commits(struct super_block *s,
|
|
|
if (!journal_list_still_alive(s, trans_id))
|
|
if (!journal_list_still_alive(s, trans_id))
|
|
|
return 1;
|
|
return 1;
|
|
|
|
|
|
|
|
- /* the one we just flushed is gone, this means all
|
|
|
|
|
- * older lists are also gone, so first_jl is no longer
|
|
|
|
|
- * valid either. Go back to the beginning.
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * the one we just flushed is gone, this means
|
|
|
|
|
+ * all older lists are also gone, so first_jl
|
|
|
|
|
+ * is no longer valid either. Go back to the
|
|
|
|
|
+ * beginning.
|
|
|
*/
|
|
*/
|
|
|
if (!journal_list_still_alive
|
|
if (!journal_list_still_alive
|
|
|
(s, other_trans_id)) {
|
|
(s, other_trans_id)) {
|
|
@@ -958,12 +964,12 @@ static int reiserfs_async_progress_wait(struct super_block *s)
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** if this journal list still has commit blocks unflushed, send them to disk.
|
|
|
|
|
-**
|
|
|
|
|
-** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
|
|
|
|
|
-** Before the commit block can by written, every other log block must be safely on disk
|
|
|
|
|
-**
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * if this journal list still has commit blocks unflushed, send them to disk.
|
|
|
|
|
+ *
|
|
|
|
|
+ * log areas must be flushed in order (transaction 2 can't commit before
|
|
|
|
|
+ * transaction 1) Before the commit block can by written, every other log
|
|
|
|
|
+ * block must be safely on disk
|
|
|
|
|
+ */
|
|
|
static int flush_commit_list(struct super_block *s,
|
|
static int flush_commit_list(struct super_block *s,
|
|
|
struct reiserfs_journal_list *jl, int flushall)
|
|
struct reiserfs_journal_list *jl, int flushall)
|
|
|
{
|
|
{
|
|
@@ -982,8 +988,9 @@ static int flush_commit_list(struct super_block *s,
|
|
|
return 0;
|
|
return 0;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- /* before we can put our commit blocks on disk, we have to make sure everyone older than
|
|
|
|
|
- ** us is on disk too
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * before we can put our commit blocks on disk, we have to make
|
|
|
|
|
+ * sure everyone older than us is on disk too
|
|
|
*/
|
|
*/
|
|
|
BUG_ON(jl->j_len <= 0);
|
|
BUG_ON(jl->j_len <= 0);
|
|
|
BUG_ON(trans_id == journal->j_trans_id);
|
|
BUG_ON(trans_id == journal->j_trans_id);
|
|
@@ -991,7 +998,10 @@ static int flush_commit_list(struct super_block *s,
|
|
|
get_journal_list(jl);
|
|
get_journal_list(jl);
|
|
|
if (flushall) {
|
|
if (flushall) {
|
|
|
if (flush_older_commits(s, jl) == 1) {
|
|
if (flush_older_commits(s, jl) == 1) {
|
|
|
- /* list disappeared during flush_older_commits. return */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * list disappeared during flush_older_commits.
|
|
|
|
|
+ * return
|
|
|
|
|
+ */
|
|
|
goto put_jl;
|
|
goto put_jl;
|
|
|
}
|
|
}
|
|
|
}
|
|
}
|
|
@@ -1006,9 +1016,9 @@ static int flush_commit_list(struct super_block *s,
|
|
|
BUG_ON(jl->j_trans_id == 0);
|
|
BUG_ON(jl->j_trans_id == 0);
|
|
|
|
|
|
|
|
/* this commit is done, exit */
|
|
/* this commit is done, exit */
|
|
|
- if (atomic_read(&(jl->j_commit_left)) <= 0) {
|
|
|
|
|
|
|
+ if (atomic_read(&jl->j_commit_left) <= 0) {
|
|
|
if (flushall) {
|
|
if (flushall) {
|
|
|
- atomic_set(&(jl->j_older_commits_done), 1);
|
|
|
|
|
|
|
+ atomic_set(&jl->j_older_commits_done, 1);
|
|
|
}
|
|
}
|
|
|
mutex_unlock(&jl->j_commit_mutex);
|
|
mutex_unlock(&jl->j_commit_mutex);
|
|
|
goto put_jl;
|
|
goto put_jl;
|
|
@@ -1063,9 +1073,10 @@ static int flush_commit_list(struct super_block *s,
|
|
|
depth = reiserfs_write_unlock_nested(s);
|
|
depth = reiserfs_write_unlock_nested(s);
|
|
|
__wait_on_buffer(tbh);
|
|
__wait_on_buffer(tbh);
|
|
|
reiserfs_write_lock_nested(s, depth);
|
|
reiserfs_write_lock_nested(s, depth);
|
|
|
- // since we're using ll_rw_blk above, it might have skipped over
|
|
|
|
|
- // a locked buffer. Double check here
|
|
|
|
|
- //
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * since we're using ll_rw_blk above, it might have skipped
|
|
|
|
|
+ * over a locked buffer. Double check here
|
|
|
|
|
+ */
|
|
|
/* redundant, sync_dirty_buffer() checks */
|
|
/* redundant, sync_dirty_buffer() checks */
|
|
|
if (buffer_dirty(tbh)) {
|
|
if (buffer_dirty(tbh)) {
|
|
|
depth = reiserfs_write_unlock_nested(s);
|
|
depth = reiserfs_write_unlock_nested(s);
|
|
@@ -1079,17 +1090,21 @@ static int flush_commit_list(struct super_block *s,
|
|
|
#endif
|
|
#endif
|
|
|
retval = -EIO;
|
|
retval = -EIO;
|
|
|
}
|
|
}
|
|
|
- put_bh(tbh); /* once for journal_find_get_block */
|
|
|
|
|
- put_bh(tbh); /* once due to original getblk in do_journal_end */
|
|
|
|
|
- atomic_dec(&(jl->j_commit_left));
|
|
|
|
|
|
|
+ /* once for journal_find_get_block */
|
|
|
|
|
+ put_bh(tbh);
|
|
|
|
|
+ /* once due to original getblk in do_journal_end */
|
|
|
|
|
+ put_bh(tbh);
|
|
|
|
|
+ atomic_dec(&jl->j_commit_left);
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
|
|
|
|
|
|
|
+ BUG_ON(atomic_read(&jl->j_commit_left) != 1);
|
|
|
|
|
|
|
|
- /* If there was a write error in the journal - we can't commit
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * If there was a write error in the journal - we can't commit
|
|
|
* this transaction - it will be invalid and, if successful,
|
|
* this transaction - it will be invalid and, if successful,
|
|
|
* will just end up propagating the write error out to
|
|
* will just end up propagating the write error out to
|
|
|
- * the file system. */
|
|
|
|
|
|
|
+ * the file system.
|
|
|
|
|
+ */
|
|
|
if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
|
|
if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
|
|
|
if (buffer_dirty(jl->j_commit_bh))
|
|
if (buffer_dirty(jl->j_commit_bh))
|
|
|
BUG();
|
|
BUG();
|
|
@@ -1102,9 +1117,11 @@ static int flush_commit_list(struct super_block *s,
|
|
|
reiserfs_write_lock_nested(s, depth);
|
|
reiserfs_write_lock_nested(s, depth);
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- /* If there was a write error in the journal - we can't commit this
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * If there was a write error in the journal - we can't commit this
|
|
|
* transaction - it will be invalid and, if successful, will just end
|
|
* transaction - it will be invalid and, if successful, will just end
|
|
|
- * up propagating the write error out to the filesystem. */
|
|
|
|
|
|
|
+ * up propagating the write error out to the filesystem.
|
|
|
|
|
+ */
|
|
|
if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
|
|
if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
|
|
|
#ifdef CONFIG_REISERFS_CHECK
|
|
#ifdef CONFIG_REISERFS_CHECK
|
|
|
reiserfs_warning(s, "journal-615", "buffer write failed");
|
|
reiserfs_warning(s, "journal-615", "buffer write failed");
|
|
@@ -1119,7 +1136,10 @@ static int flush_commit_list(struct super_block *s,
|
|
|
}
|
|
}
|
|
|
journal->j_last_commit_id = jl->j_trans_id;
|
|
journal->j_last_commit_id = jl->j_trans_id;
|
|
|
|
|
|
|
|
- /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * now, every commit block is on the disk. It is safe to allow
|
|
|
|
|
+ * blocks freed during this transaction to be reallocated
|
|
|
|
|
+ */
|
|
|
cleanup_freed_for_journal_list(s, jl);
|
|
cleanup_freed_for_journal_list(s, jl);
|
|
|
|
|
|
|
|
retval = retval ? retval : journal->j_errno;
|
|
retval = retval ? retval : journal->j_errno;
|
|
@@ -1127,13 +1147,13 @@ static int flush_commit_list(struct super_block *s,
|
|
|
/* mark the metadata dirty */
|
|
/* mark the metadata dirty */
|
|
|
if (!retval)
|
|
if (!retval)
|
|
|
dirty_one_transaction(s, jl);
|
|
dirty_one_transaction(s, jl);
|
|
|
- atomic_dec(&(jl->j_commit_left));
|
|
|
|
|
|
|
+ atomic_dec(&jl->j_commit_left);
|
|
|
|
|
|
|
|
if (flushall) {
|
|
if (flushall) {
|
|
|
- atomic_set(&(jl->j_older_commits_done), 1);
|
|
|
|
|
|
|
+ atomic_set(&jl->j_older_commits_done, 1);
|
|
|
}
|
|
}
|
|
|
mutex_unlock(&jl->j_commit_mutex);
|
|
mutex_unlock(&jl->j_commit_mutex);
|
|
|
- put_jl:
|
|
|
|
|
|
|
+put_jl:
|
|
|
put_journal_list(s, jl);
|
|
put_journal_list(s, jl);
|
|
|
|
|
|
|
|
if (retval)
|
|
if (retval)
|
|
@@ -1143,9 +1163,9 @@ static int flush_commit_list(struct super_block *s,
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or
|
|
|
|
|
-** returns NULL if it can't find anything
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * flush_journal_list frequently needs to find a newer transaction for a
|
|
|
|
|
+ * given block. This does that, or returns NULL if it can't find anything
|
|
|
|
|
+ */
|
|
|
static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
|
|
static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
|
|
|
reiserfs_journal_cnode
|
|
reiserfs_journal_cnode
|
|
|
*cn)
|
|
*cn)
|
|
@@ -1169,10 +1189,11 @@ static void remove_journal_hash(struct super_block *,
|
|
|
int);
|
|
int);
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** once all the real blocks have been flushed, it is safe to remove them from the
|
|
|
|
|
-** journal list for this transaction. Aside from freeing the cnode, this also allows the
|
|
|
|
|
-** block to be reallocated for data blocks if it had been deleted.
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * once all the real blocks have been flushed, it is safe to remove them
|
|
|
|
|
+ * from the journal list for this transaction. Aside from freeing the
|
|
|
|
|
+ * cnode, this also allows the block to be reallocated for data blocks
|
|
|
|
|
+ * if it had been deleted.
|
|
|
|
|
+ */
|
|
|
static void remove_all_from_journal_list(struct super_block *sb,
|
|
static void remove_all_from_journal_list(struct super_block *sb,
|
|
|
struct reiserfs_journal_list *jl,
|
|
struct reiserfs_journal_list *jl,
|
|
|
int debug)
|
|
int debug)
|
|
@@ -1181,8 +1202,9 @@ static void remove_all_from_journal_list(struct super_block *sb,
|
|
|
struct reiserfs_journal_cnode *cn, *last;
|
|
struct reiserfs_journal_cnode *cn, *last;
|
|
|
cn = jl->j_realblock;
|
|
cn = jl->j_realblock;
|
|
|
|
|
|
|
|
- /* which is better, to lock once around the whole loop, or
|
|
|
|
|
- ** to lock for each call to remove_journal_hash?
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * which is better, to lock once around the whole loop, or
|
|
|
|
|
+ * to lock for each call to remove_journal_hash?
|
|
|
*/
|
|
*/
|
|
|
while (cn) {
|
|
while (cn) {
|
|
|
if (cn->blocknr != 0) {
|
|
if (cn->blocknr != 0) {
|
|
@@ -1204,12 +1226,13 @@ static void remove_all_from_journal_list(struct super_block *sb,
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
|
|
|
|
|
-** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
|
|
|
|
|
-** releasing blocks in this transaction for reuse as data blocks.
|
|
|
|
|
-** called by flush_journal_list, before it calls remove_all_from_journal_list
|
|
|
|
|
-**
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * if this timestamp is greater than the timestamp we wrote last to the
|
|
|
|
|
+ * header block, write it to the header block. once this is done, I can
|
|
|
|
|
+ * safely say the log area for this transaction won't ever be replayed,
|
|
|
|
|
+ * and I can start releasing blocks in this transaction for reuse as data
|
|
|
|
|
+ * blocks. called by flush_journal_list, before it calls
|
|
|
|
|
+ * remove_all_from_journal_list
|
|
|
|
|
+ */
|
|
|
static int _update_journal_header_block(struct super_block *sb,
|
|
static int _update_journal_header_block(struct super_block *sb,
|
|
|
unsigned long offset,
|
|
unsigned long offset,
|
|
|
unsigned int trans_id)
|
|
unsigned int trans_id)
|
|
@@ -1279,10 +1302,11 @@ static int flush_older_journal_lists(struct super_block *sb,
|
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
|
unsigned int trans_id = jl->j_trans_id;
|
|
unsigned int trans_id = jl->j_trans_id;
|
|
|
|
|
|
|
|
- /* we know we are the only ones flushing things, no extra race
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * we know we are the only ones flushing things, no extra race
|
|
|
* protection is required.
|
|
* protection is required.
|
|
|
*/
|
|
*/
|
|
|
- restart:
|
|
|
|
|
|
|
+restart:
|
|
|
entry = journal->j_journal_list.next;
|
|
entry = journal->j_journal_list.next;
|
|
|
/* Did we wrap? */
|
|
/* Did we wrap? */
|
|
|
if (entry == &journal->j_journal_list)
|
|
if (entry == &journal->j_journal_list)
|
|
@@ -1309,15 +1333,16 @@ static void del_from_work_list(struct super_block *s,
|
|
|
}
|
|
}
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/* flush a journal list, both commit and real blocks
|
|
|
|
|
-**
|
|
|
|
|
-** always set flushall to 1, unless you are calling from inside
|
|
|
|
|
-** flush_journal_list
|
|
|
|
|
-**
|
|
|
|
|
-** IMPORTANT. This can only be called while there are no journal writers,
|
|
|
|
|
-** and the journal is locked. That means it can only be called from
|
|
|
|
|
-** do_journal_end, or by journal_release
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+/*
|
|
|
|
|
+ * flush a journal list, both commit and real blocks
|
|
|
|
|
+ *
|
|
|
|
|
+ * always set flushall to 1, unless you are calling from inside
|
|
|
|
|
+ * flush_journal_list
|
|
|
|
|
+ *
|
|
|
|
|
+ * IMPORTANT. This can only be called while there are no journal writers,
|
|
|
|
|
+ * and the journal is locked. That means it can only be called from
|
|
|
|
|
+ * do_journal_end, or by journal_release
|
|
|
|
|
+ */
|
|
|
static int flush_journal_list(struct super_block *s,
|
|
static int flush_journal_list(struct super_block *s,
|
|
|
struct reiserfs_journal_list *jl, int flushall)
|
|
struct reiserfs_journal_list *jl, int flushall)
|
|
|
{
|
|
{
|
|
@@ -1354,13 +1379,14 @@ static int flush_journal_list(struct super_block *s,
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/* if all the work is already done, get out of here */
|
|
/* if all the work is already done, get out of here */
|
|
|
- if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
|
|
|
|
|
- atomic_read(&(jl->j_commit_left)) <= 0) {
|
|
|
|
|
|
|
+ if (atomic_read(&jl->j_nonzerolen) <= 0 &&
|
|
|
|
|
+ atomic_read(&jl->j_commit_left) <= 0) {
|
|
|
goto flush_older_and_return;
|
|
goto flush_older_and_return;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- /* start by putting the commit list on disk. This will also flush
|
|
|
|
|
- ** the commit lists of any olders transactions
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * start by putting the commit list on disk. This will also flush
|
|
|
|
|
+ * the commit lists of any olders transactions
|
|
|
*/
|
|
*/
|
|
|
flush_commit_list(s, jl, 1);
|
|
flush_commit_list(s, jl, 1);
|
|
|
|
|
|
|
@@ -1369,15 +1395,16 @@ static int flush_journal_list(struct super_block *s,
|
|
|
BUG();
|
|
BUG();
|
|
|
|
|
|
|
|
/* are we done now? */
|
|
/* are we done now? */
|
|
|
- if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
|
|
|
|
|
- atomic_read(&(jl->j_commit_left)) <= 0) {
|
|
|
|
|
|
|
+ if (atomic_read(&jl->j_nonzerolen) <= 0 &&
|
|
|
|
|
+ atomic_read(&jl->j_commit_left) <= 0) {
|
|
|
goto flush_older_and_return;
|
|
goto flush_older_and_return;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- /* loop through each cnode, see if we need to write it,
|
|
|
|
|
- ** or wait on a more recent transaction, or just ignore it
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * loop through each cnode, see if we need to write it,
|
|
|
|
|
+ * or wait on a more recent transaction, or just ignore it
|
|
|
*/
|
|
*/
|
|
|
- if (atomic_read(&(journal->j_wcount)) != 0) {
|
|
|
|
|
|
|
+ if (atomic_read(&journal->j_wcount) != 0) {
|
|
|
reiserfs_panic(s, "journal-844", "journal list is flushing, "
|
|
reiserfs_panic(s, "journal-844", "journal list is flushing, "
|
|
|
"wcount is not 0");
|
|
"wcount is not 0");
|
|
|
}
|
|
}
|
|
@@ -1391,20 +1418,25 @@ static int flush_journal_list(struct super_block *s,
|
|
|
goto free_cnode;
|
|
goto free_cnode;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- /* This transaction failed commit. Don't write out to the disk */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * This transaction failed commit.
|
|
|
|
|
+ * Don't write out to the disk
|
|
|
|
|
+ */
|
|
|
if (!(jl->j_state & LIST_DIRTY))
|
|
if (!(jl->j_state & LIST_DIRTY))
|
|
|
goto free_cnode;
|
|
goto free_cnode;
|
|
|
|
|
|
|
|
pjl = find_newer_jl_for_cn(cn);
|
|
pjl = find_newer_jl_for_cn(cn);
|
|
|
- /* the order is important here. We check pjl to make sure we
|
|
|
|
|
- ** don't clear BH_JDirty_wait if we aren't the one writing this
|
|
|
|
|
- ** block to disk
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * the order is important here. We check pjl to make sure we
|
|
|
|
|
+ * don't clear BH_JDirty_wait if we aren't the one writing this
|
|
|
|
|
+ * block to disk
|
|
|
*/
|
|
*/
|
|
|
if (!pjl && cn->bh) {
|
|
if (!pjl && cn->bh) {
|
|
|
saved_bh = cn->bh;
|
|
saved_bh = cn->bh;
|
|
|
|
|
|
|
|
- /* we do this to make sure nobody releases the buffer while
|
|
|
|
|
- ** we are working with it
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * we do this to make sure nobody releases the
|
|
|
|
|
+ * buffer while we are working with it
|
|
|
*/
|
|
*/
|
|
|
get_bh(saved_bh);
|
|
get_bh(saved_bh);
|
|
|
|
|
|
|
@@ -1413,13 +1445,17 @@ static int flush_journal_list(struct super_block *s,
|
|
|
was_jwait = 1;
|
|
was_jwait = 1;
|
|
|
was_dirty = 1;
|
|
was_dirty = 1;
|
|
|
} else if (can_dirty(cn)) {
|
|
} else if (can_dirty(cn)) {
|
|
|
- /* everything with !pjl && jwait should be writable */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * everything with !pjl && jwait
|
|
|
|
|
+ * should be writable
|
|
|
|
|
+ */
|
|
|
BUG();
|
|
BUG();
|
|
|
}
|
|
}
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- /* if someone has this block in a newer transaction, just make
|
|
|
|
|
- ** sure they are committed, and don't try writing it to disk
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * if someone has this block in a newer transaction, just make
|
|
|
|
|
+ * sure they are committed, and don't try writing it to disk
|
|
|
*/
|
|
*/
|
|
|
if (pjl) {
|
|
if (pjl) {
|
|
|
if (atomic_read(&pjl->j_commit_left))
|
|
if (atomic_read(&pjl->j_commit_left))
|
|
@@ -1427,16 +1463,18 @@ static int flush_journal_list(struct super_block *s,
|
|
|
goto free_cnode;
|
|
goto free_cnode;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- /* bh == NULL when the block got to disk on its own, OR,
|
|
|
|
|
- ** the block got freed in a future transaction
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * bh == NULL when the block got to disk on its own, OR,
|
|
|
|
|
+ * the block got freed in a future transaction
|
|
|
*/
|
|
*/
|
|
|
if (saved_bh == NULL) {
|
|
if (saved_bh == NULL) {
|
|
|
goto free_cnode;
|
|
goto free_cnode;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- /* this should never happen. kupdate_one_transaction has this list
|
|
|
|
|
- ** locked while it works, so we should never see a buffer here that
|
|
|
|
|
- ** is not marked JDirty_wait
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * this should never happen. kupdate_one_transaction has
|
|
|
|
|
+ * this list locked while it works, so we should never see a
|
|
|
|
|
+ * buffer here that is not marked JDirty_wait
|
|
|
*/
|
|
*/
|
|
|
if ((!was_jwait) && !buffer_locked(saved_bh)) {
|
|
if ((!was_jwait) && !buffer_locked(saved_bh)) {
|
|
|
reiserfs_warning(s, "journal-813",
|
|
reiserfs_warning(s, "journal-813",
|
|
@@ -1447,7 +1485,10 @@ static int flush_journal_list(struct super_block *s,
|
|
|
was_jwait ? ' ' : '!');
|
|
was_jwait ? ' ' : '!');
|
|
|
}
|
|
}
|
|
|
if (was_dirty) {
|
|
if (was_dirty) {
|
|
|
- /* we inc again because saved_bh gets decremented at free_cnode */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * we inc again because saved_bh gets decremented
|
|
|
|
|
+ * at free_cnode
|
|
|
|
|
+ */
|
|
|
get_bh(saved_bh);
|
|
get_bh(saved_bh);
|
|
|
set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
|
|
set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
|
|
|
lock_buffer(saved_bh);
|
|
lock_buffer(saved_bh);
|
|
@@ -1463,13 +1504,16 @@ static int flush_journal_list(struct super_block *s,
|
|
|
(unsigned long long)saved_bh->
|
|
(unsigned long long)saved_bh->
|
|
|
b_blocknr, __func__);
|
|
b_blocknr, __func__);
|
|
|
}
|
|
}
|
|
|
- free_cnode:
|
|
|
|
|
|
|
+free_cnode:
|
|
|
last = cn;
|
|
last = cn;
|
|
|
cn = cn->next;
|
|
cn = cn->next;
|
|
|
if (saved_bh) {
|
|
if (saved_bh) {
|
|
|
- /* we incremented this to keep others from taking the buffer head away */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * we incremented this to keep others from
|
|
|
|
|
+ * taking the buffer head away
|
|
|
|
|
+ */
|
|
|
put_bh(saved_bh);
|
|
put_bh(saved_bh);
|
|
|
- if (atomic_read(&(saved_bh->b_count)) < 0) {
|
|
|
|
|
|
|
+ if (atomic_read(&saved_bh->b_count) < 0) {
|
|
|
reiserfs_warning(s, "journal-945",
|
|
reiserfs_warning(s, "journal-945",
|
|
|
"saved_bh->b_count < 0");
|
|
"saved_bh->b_count < 0");
|
|
|
}
|
|
}
|
|
@@ -1499,8 +1543,10 @@ static int flush_journal_list(struct super_block *s,
|
|
|
#endif
|
|
#endif
|
|
|
err = -EIO;
|
|
err = -EIO;
|
|
|
}
|
|
}
|
|
|
- /* note, we must clear the JDirty_wait bit after the up to date
|
|
|
|
|
- ** check, otherwise we race against our flushpage routine
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * note, we must clear the JDirty_wait bit
|
|
|
|
|
+ * after the up to date check, otherwise we
|
|
|
|
|
+ * race against our flushpage routine
|
|
|
*/
|
|
*/
|
|
|
BUG_ON(!test_clear_buffer_journal_dirty
|
|
BUG_ON(!test_clear_buffer_journal_dirty
|
|
|
(cn->bh));
|
|
(cn->bh));
|
|
@@ -1518,25 +1564,27 @@ static int flush_journal_list(struct super_block *s,
|
|
|
reiserfs_abort(s, -EIO,
|
|
reiserfs_abort(s, -EIO,
|
|
|
"Write error while pushing transaction to disk in %s",
|
|
"Write error while pushing transaction to disk in %s",
|
|
|
__func__);
|
|
__func__);
|
|
|
- flush_older_and_return:
|
|
|
|
|
|
|
+flush_older_and_return:
|
|
|
|
|
|
|
|
- /* before we can update the journal header block, we _must_ flush all
|
|
|
|
|
- ** real blocks from all older transactions to disk. This is because
|
|
|
|
|
- ** once the header block is updated, this transaction will not be
|
|
|
|
|
- ** replayed after a crash
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * before we can update the journal header block, we _must_ flush all
|
|
|
|
|
+ * real blocks from all older transactions to disk. This is because
|
|
|
|
|
+ * once the header block is updated, this transaction will not be
|
|
|
|
|
+ * replayed after a crash
|
|
|
*/
|
|
*/
|
|
|
if (flushall) {
|
|
if (flushall) {
|
|
|
flush_older_journal_lists(s, jl);
|
|
flush_older_journal_lists(s, jl);
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
err = journal->j_errno;
|
|
err = journal->j_errno;
|
|
|
- /* before we can remove everything from the hash tables for this
|
|
|
|
|
- ** transaction, we must make sure it can never be replayed
|
|
|
|
|
- **
|
|
|
|
|
- ** since we are only called from do_journal_end, we know for sure there
|
|
|
|
|
- ** are no allocations going on while we are flushing journal lists. So,
|
|
|
|
|
- ** we only need to update the journal header block for the last list
|
|
|
|
|
- ** being flushed
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * before we can remove everything from the hash tables for this
|
|
|
|
|
+ * transaction, we must make sure it can never be replayed
|
|
|
|
|
+ *
|
|
|
|
|
+ * since we are only called from do_journal_end, we know for sure there
|
|
|
|
|
+ * are no allocations going on while we are flushing journal lists. So,
|
|
|
|
|
+ * we only need to update the journal header block for the last list
|
|
|
|
|
+ * being flushed
|
|
|
*/
|
|
*/
|
|
|
if (!err && flushall) {
|
|
if (!err && flushall) {
|
|
|
err =
|
|
err =
|
|
@@ -1561,11 +1609,12 @@ static int flush_journal_list(struct super_block *s,
|
|
|
}
|
|
}
|
|
|
journal->j_last_flush_id = jl->j_trans_id;
|
|
journal->j_last_flush_id = jl->j_trans_id;
|
|
|
|
|
|
|
|
- /* not strictly required since we are freeing the list, but it should
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * not strictly required since we are freeing the list, but it should
|
|
|
* help find code using dead lists later on
|
|
* help find code using dead lists later on
|
|
|
*/
|
|
*/
|
|
|
jl->j_len = 0;
|
|
jl->j_len = 0;
|
|
|
- atomic_set(&(jl->j_nonzerolen), 0);
|
|
|
|
|
|
|
+ atomic_set(&jl->j_nonzerolen, 0);
|
|
|
jl->j_start = 0;
|
|
jl->j_start = 0;
|
|
|
jl->j_realblock = NULL;
|
|
jl->j_realblock = NULL;
|
|
|
jl->j_commit_bh = NULL;
|
|
jl->j_commit_bh = NULL;
|
|
@@ -1592,15 +1641,17 @@ static int write_one_transaction(struct super_block *s,
|
|
|
|
|
|
|
|
cn = jl->j_realblock;
|
|
cn = jl->j_realblock;
|
|
|
while (cn) {
|
|
while (cn) {
|
|
|
- /* if the blocknr == 0, this has been cleared from the hash,
|
|
|
|
|
- ** skip it
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * if the blocknr == 0, this has been cleared from the hash,
|
|
|
|
|
+ * skip it
|
|
|
*/
|
|
*/
|
|
|
if (cn->blocknr == 0) {
|
|
if (cn->blocknr == 0) {
|
|
|
goto next;
|
|
goto next;
|
|
|
}
|
|
}
|
|
|
if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
|
|
if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
|
|
|
struct buffer_head *tmp_bh;
|
|
struct buffer_head *tmp_bh;
|
|
|
- /* we can race against journal_mark_freed when we try
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * we can race against journal_mark_freed when we try
|
|
|
* to lock_buffer(cn->bh), so we have to inc the buffer
|
|
* to lock_buffer(cn->bh), so we have to inc the buffer
|
|
|
* count, and recheck things after locking
|
|
* count, and recheck things after locking
|
|
|
*/
|
|
*/
|
|
@@ -1619,7 +1670,7 @@ static int write_one_transaction(struct super_block *s,
|
|
|
}
|
|
}
|
|
|
put_bh(tmp_bh);
|
|
put_bh(tmp_bh);
|
|
|
}
|
|
}
|
|
|
- next:
|
|
|
|
|
|
|
+next:
|
|
|
cn = cn->next;
|
|
cn = cn->next;
|
|
|
cond_resched();
|
|
cond_resched();
|
|
|
}
|
|
}
|
|
@@ -1637,15 +1688,17 @@ static int dirty_one_transaction(struct super_block *s,
|
|
|
jl->j_state |= LIST_DIRTY;
|
|
jl->j_state |= LIST_DIRTY;
|
|
|
cn = jl->j_realblock;
|
|
cn = jl->j_realblock;
|
|
|
while (cn) {
|
|
while (cn) {
|
|
|
- /* look for a more recent transaction that logged this
|
|
|
|
|
- ** buffer. Only the most recent transaction with a buffer in
|
|
|
|
|
- ** it is allowed to send that buffer to disk
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * look for a more recent transaction that logged this
|
|
|
|
|
+ * buffer. Only the most recent transaction with a buffer in
|
|
|
|
|
+ * it is allowed to send that buffer to disk
|
|
|
*/
|
|
*/
|
|
|
pjl = find_newer_jl_for_cn(cn);
|
|
pjl = find_newer_jl_for_cn(cn);
|
|
|
if (!pjl && cn->blocknr && cn->bh
|
|
if (!pjl && cn->blocknr && cn->bh
|
|
|
&& buffer_journal_dirty(cn->bh)) {
|
|
&& buffer_journal_dirty(cn->bh)) {
|
|
|
BUG_ON(!can_dirty(cn));
|
|
BUG_ON(!can_dirty(cn));
|
|
|
- /* if the buffer is prepared, it will either be logged
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * if the buffer is prepared, it will either be logged
|
|
|
* or restored. If restored, we need to make sure
|
|
* or restored. If restored, we need to make sure
|
|
|
* it actually gets marked dirty
|
|
* it actually gets marked dirty
|
|
|
*/
|
|
*/
|
|
@@ -1682,7 +1735,8 @@ static int kupdate_transactions(struct super_block *s,
|
|
|
goto done;
|
|
goto done;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- /* we've got j_flush_mutex held, nobody is going to delete any
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * we've got j_flush_mutex held, nobody is going to delete any
|
|
|
* of these lists out from underneath us
|
|
* of these lists out from underneath us
|
|
|
*/
|
|
*/
|
|
|
while ((num_trans && transactions_flushed < num_trans) ||
|
|
while ((num_trans && transactions_flushed < num_trans) ||
|
|
@@ -1716,20 +1770,21 @@ static int kupdate_transactions(struct super_block *s,
|
|
|
write_chunk(&chunk);
|
|
write_chunk(&chunk);
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- done:
|
|
|
|
|
|
|
+done:
|
|
|
mutex_unlock(&journal->j_flush_mutex);
|
|
mutex_unlock(&journal->j_flush_mutex);
|
|
|
return ret;
|
|
return ret;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/* for o_sync and fsync heavy applications, they tend to use
|
|
|
|
|
-** all the journa list slots with tiny transactions. These
|
|
|
|
|
-** trigger lots and lots of calls to update the header block, which
|
|
|
|
|
-** adds seeks and slows things down.
|
|
|
|
|
-**
|
|
|
|
|
-** This function tries to clear out a large chunk of the journal lists
|
|
|
|
|
-** at once, which makes everything faster since only the newest journal
|
|
|
|
|
-** list updates the header block
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+/*
|
|
|
|
|
+ * for o_sync and fsync heavy applications, they tend to use
|
|
|
|
|
+ * all the journa list slots with tiny transactions. These
|
|
|
|
|
+ * trigger lots and lots of calls to update the header block, which
|
|
|
|
|
+ * adds seeks and slows things down.
|
|
|
|
|
+ *
|
|
|
|
|
+ * This function tries to clear out a large chunk of the journal lists
|
|
|
|
|
+ * at once, which makes everything faster since only the newest journal
|
|
|
|
|
+ * list updates the header block
|
|
|
|
|
+ */
|
|
|
static int flush_used_journal_lists(struct super_block *s,
|
|
static int flush_used_journal_lists(struct super_block *s,
|
|
|
struct reiserfs_journal_list *jl)
|
|
struct reiserfs_journal_list *jl)
|
|
|
{
|
|
{
|
|
@@ -1766,9 +1821,11 @@ static int flush_used_journal_lists(struct super_block *s,
|
|
|
}
|
|
}
|
|
|
get_journal_list(jl);
|
|
get_journal_list(jl);
|
|
|
get_journal_list(flush_jl);
|
|
get_journal_list(flush_jl);
|
|
|
- /* try to find a group of blocks we can flush across all the
|
|
|
|
|
- ** transactions, but only bother if we've actually spanned
|
|
|
|
|
- ** across multiple lists
|
|
|
|
|
|
|
+
|
|
|
|
|
+ /*
|
|
|
|
|
+ * try to find a group of blocks we can flush across all the
|
|
|
|
|
+ * transactions, but only bother if we've actually spanned
|
|
|
|
|
+ * across multiple lists
|
|
|
*/
|
|
*/
|
|
|
if (flush_jl != jl) {
|
|
if (flush_jl != jl) {
|
|
|
ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
|
|
ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
|
|
@@ -1780,9 +1837,9 @@ static int flush_used_journal_lists(struct super_block *s,
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** removes any nodes in table with name block and dev as bh.
|
|
|
|
|
-** only touchs the hnext and hprev pointers.
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * removes any nodes in table with name block and dev as bh.
|
|
|
|
|
+ * only touchs the hnext and hprev pointers.
|
|
|
|
|
+ */
|
|
|
void remove_journal_hash(struct super_block *sb,
|
|
void remove_journal_hash(struct super_block *sb,
|
|
|
struct reiserfs_journal_cnode **table,
|
|
struct reiserfs_journal_cnode **table,
|
|
|
struct reiserfs_journal_list *jl,
|
|
struct reiserfs_journal_list *jl,
|
|
@@ -1811,8 +1868,12 @@ void remove_journal_hash(struct super_block *sb,
|
|
|
cur->blocknr = 0;
|
|
cur->blocknr = 0;
|
|
|
cur->sb = NULL;
|
|
cur->sb = NULL;
|
|
|
cur->state = 0;
|
|
cur->state = 0;
|
|
|
- if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */
|
|
|
|
|
- atomic_dec(&(cur->jlist->j_nonzerolen));
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * anybody who clears the cur->bh will also
|
|
|
|
|
+ * dec the nonzerolen
|
|
|
|
|
+ */
|
|
|
|
|
+ if (cur->bh && cur->jlist)
|
|
|
|
|
+ atomic_dec(&cur->jlist->j_nonzerolen);
|
|
|
cur->bh = NULL;
|
|
cur->bh = NULL;
|
|
|
cur->jlist = NULL;
|
|
cur->jlist = NULL;
|
|
|
}
|
|
}
|
|
@@ -1832,17 +1893,18 @@ static void free_journal_ram(struct super_block *sb)
|
|
|
if (journal->j_header_bh) {
|
|
if (journal->j_header_bh) {
|
|
|
brelse(journal->j_header_bh);
|
|
brelse(journal->j_header_bh);
|
|
|
}
|
|
}
|
|
|
- /* j_header_bh is on the journal dev, make sure not to release the journal
|
|
|
|
|
- * dev until we brelse j_header_bh
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * j_header_bh is on the journal dev, make sure
|
|
|
|
|
+ * not to release the journal dev until we brelse j_header_bh
|
|
|
*/
|
|
*/
|
|
|
release_journal_dev(sb, journal);
|
|
release_journal_dev(sb, journal);
|
|
|
vfree(journal);
|
|
vfree(journal);
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** call on unmount. Only set error to 1 if you haven't made your way out
|
|
|
|
|
-** of read_super() yet. Any other caller must keep error at 0.
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * call on unmount. Only set error to 1 if you haven't made your way out
|
|
|
|
|
+ * of read_super() yet. Any other caller must keep error at 0.
|
|
|
|
|
+ */
|
|
|
static int do_journal_release(struct reiserfs_transaction_handle *th,
|
|
static int do_journal_release(struct reiserfs_transaction_handle *th,
|
|
|
struct super_block *sb, int error)
|
|
struct super_block *sb, int error)
|
|
|
{
|
|
{
|
|
@@ -1850,21 +1912,25 @@ static int do_journal_release(struct reiserfs_transaction_handle *th,
|
|
|
int flushed = 0;
|
|
int flushed = 0;
|
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
|
|
|
|
|
|
- /* we only want to flush out transactions if we were called with error == 0
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * we only want to flush out transactions if we were
|
|
|
|
|
+ * called with error == 0
|
|
|
*/
|
|
*/
|
|
|
if (!error && !(sb->s_flags & MS_RDONLY)) {
|
|
if (!error && !(sb->s_flags & MS_RDONLY)) {
|
|
|
/* end the current trans */
|
|
/* end the current trans */
|
|
|
BUG_ON(!th->t_trans_id);
|
|
BUG_ON(!th->t_trans_id);
|
|
|
- do_journal_end(th, sb, 10, FLUSH_ALL);
|
|
|
|
|
|
|
+ do_journal_end(th, FLUSH_ALL);
|
|
|
|
|
|
|
|
- /* make sure something gets logged to force our way into the flush code */
|
|
|
|
|
- if (!journal_join(&myth, sb, 1)) {
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * make sure something gets logged to force
|
|
|
|
|
+ * our way into the flush code
|
|
|
|
|
+ */
|
|
|
|
|
+ if (!journal_join(&myth, sb)) {
|
|
|
reiserfs_prepare_for_journal(sb,
|
|
reiserfs_prepare_for_journal(sb,
|
|
|
SB_BUFFER_WITH_SB(sb),
|
|
SB_BUFFER_WITH_SB(sb),
|
|
|
1);
|
|
1);
|
|
|
- journal_mark_dirty(&myth, sb,
|
|
|
|
|
- SB_BUFFER_WITH_SB(sb));
|
|
|
|
|
- do_journal_end(&myth, sb, 1, FLUSH_ALL);
|
|
|
|
|
|
|
+ journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
|
|
|
|
|
+ do_journal_end(&myth, FLUSH_ALL);
|
|
|
flushed = 1;
|
|
flushed = 1;
|
|
|
}
|
|
}
|
|
|
}
|
|
}
|
|
@@ -1872,17 +1938,15 @@ static int do_journal_release(struct reiserfs_transaction_handle *th,
|
|
|
/* this also catches errors during the do_journal_end above */
|
|
/* this also catches errors during the do_journal_end above */
|
|
|
if (!error && reiserfs_is_journal_aborted(journal)) {
|
|
if (!error && reiserfs_is_journal_aborted(journal)) {
|
|
|
memset(&myth, 0, sizeof(myth));
|
|
memset(&myth, 0, sizeof(myth));
|
|
|
- if (!journal_join_abort(&myth, sb, 1)) {
|
|
|
|
|
|
|
+ if (!journal_join_abort(&myth, sb)) {
|
|
|
reiserfs_prepare_for_journal(sb,
|
|
reiserfs_prepare_for_journal(sb,
|
|
|
SB_BUFFER_WITH_SB(sb),
|
|
SB_BUFFER_WITH_SB(sb),
|
|
|
1);
|
|
1);
|
|
|
- journal_mark_dirty(&myth, sb,
|
|
|
|
|
- SB_BUFFER_WITH_SB(sb));
|
|
|
|
|
- do_journal_end(&myth, sb, 1, FLUSH_ALL);
|
|
|
|
|
|
|
+ journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
|
|
|
|
|
+ do_journal_end(&myth, FLUSH_ALL);
|
|
|
}
|
|
}
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- reiserfs_mounted_fs_count--;
|
|
|
|
|
/* wait for all commits to finish */
|
|
/* wait for all commits to finish */
|
|
|
cancel_delayed_work(&SB_JOURNAL(sb)->j_work);
|
|
cancel_delayed_work(&SB_JOURNAL(sb)->j_work);
|
|
|
|
|
|
|
@@ -1893,12 +1957,7 @@ static int do_journal_release(struct reiserfs_transaction_handle *th,
|
|
|
reiserfs_write_unlock(sb);
|
|
reiserfs_write_unlock(sb);
|
|
|
|
|
|
|
|
cancel_delayed_work_sync(&REISERFS_SB(sb)->old_work);
|
|
cancel_delayed_work_sync(&REISERFS_SB(sb)->old_work);
|
|
|
- flush_workqueue(commit_wq);
|
|
|
|
|
-
|
|
|
|
|
- if (!reiserfs_mounted_fs_count) {
|
|
|
|
|
- destroy_workqueue(commit_wq);
|
|
|
|
|
- commit_wq = NULL;
|
|
|
|
|
- }
|
|
|
|
|
|
|
+ flush_workqueue(REISERFS_SB(sb)->commit_wq);
|
|
|
|
|
|
|
|
free_journal_ram(sb);
|
|
free_journal_ram(sb);
|
|
|
|
|
|
|
@@ -1907,25 +1966,24 @@ static int do_journal_release(struct reiserfs_transaction_handle *th,
|
|
|
return 0;
|
|
return 0;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/*
|
|
|
|
|
-** call on unmount. flush all journal trans, release all alloc'd ram
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+/* * call on unmount. flush all journal trans, release all alloc'd ram */
|
|
|
int journal_release(struct reiserfs_transaction_handle *th,
|
|
int journal_release(struct reiserfs_transaction_handle *th,
|
|
|
struct super_block *sb)
|
|
struct super_block *sb)
|
|
|
{
|
|
{
|
|
|
return do_journal_release(th, sb, 0);
|
|
return do_journal_release(th, sb, 0);
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/*
|
|
|
|
|
-** only call from an error condition inside reiserfs_read_super!
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+/* only call from an error condition inside reiserfs_read_super! */
|
|
|
int journal_release_error(struct reiserfs_transaction_handle *th,
|
|
int journal_release_error(struct reiserfs_transaction_handle *th,
|
|
|
struct super_block *sb)
|
|
struct super_block *sb)
|
|
|
{
|
|
{
|
|
|
return do_journal_release(th, sb, 1);
|
|
return do_journal_release(th, sb, 1);
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/* compares description block with commit block. returns 1 if they differ, 0 if they are the same */
|
|
|
|
|
|
|
+/*
|
|
|
|
|
+ * compares description block with commit block.
|
|
|
|
|
+ * returns 1 if they differ, 0 if they are the same
|
|
|
|
|
+ */
|
|
|
static int journal_compare_desc_commit(struct super_block *sb,
|
|
static int journal_compare_desc_commit(struct super_block *sb,
|
|
|
struct reiserfs_journal_desc *desc,
|
|
struct reiserfs_journal_desc *desc,
|
|
|
struct reiserfs_journal_commit *commit)
|
|
struct reiserfs_journal_commit *commit)
|
|
@@ -1939,11 +1997,12 @@ static int journal_compare_desc_commit(struct super_block *sb,
|
|
|
return 0;
|
|
return 0;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/* returns 0 if it did not find a description block
|
|
|
|
|
-** returns -1 if it found a corrupt commit block
|
|
|
|
|
-** returns 1 if both desc and commit were valid
|
|
|
|
|
-** NOTE: only called during fs mount
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+/*
|
|
|
|
|
+ * returns 0 if it did not find a description block
|
|
|
|
|
+ * returns -1 if it found a corrupt commit block
|
|
|
|
|
+ * returns 1 if both desc and commit were valid
|
|
|
|
|
+ * NOTE: only called during fs mount
|
|
|
|
|
+ */
|
|
|
static int journal_transaction_is_valid(struct super_block *sb,
|
|
static int journal_transaction_is_valid(struct super_block *sb,
|
|
|
struct buffer_head *d_bh,
|
|
struct buffer_head *d_bh,
|
|
|
unsigned int *oldest_invalid_trans_id,
|
|
unsigned int *oldest_invalid_trans_id,
|
|
@@ -1989,7 +2048,10 @@ static int journal_transaction_is_valid(struct super_block *sb,
|
|
|
}
|
|
}
|
|
|
offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
|
|
offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
|
|
|
|
|
|
|
|
- /* ok, we have a journal description block, lets see if the transaction was valid */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * ok, we have a journal description block,
|
|
|
|
|
+ * let's see if the transaction was valid
|
|
|
|
|
+ */
|
|
|
c_bh =
|
|
c_bh =
|
|
|
journal_bread(sb,
|
|
journal_bread(sb,
|
|
|
SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
|
|
SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
|
|
@@ -2041,11 +2103,11 @@ static void brelse_array(struct buffer_head **heads, int num)
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** given the start, and values for the oldest acceptable transactions,
|
|
|
|
|
-** this either reads in a replays a transaction, or returns because the
|
|
|
|
|
-** transaction is invalid, or too old.
|
|
|
|
|
-** NOTE: only called during fs mount
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * given the start, and values for the oldest acceptable transactions,
|
|
|
|
|
+ * this either reads in a replays a transaction, or returns because the
|
|
|
|
|
+ * transaction is invalid, or too old.
|
|
|
|
|
+ * NOTE: only called during fs mount
|
|
|
|
|
+ */
|
|
|
static int journal_read_transaction(struct super_block *sb,
|
|
static int journal_read_transaction(struct super_block *sb,
|
|
|
unsigned long cur_dblock,
|
|
unsigned long cur_dblock,
|
|
|
unsigned long oldest_start,
|
|
unsigned long oldest_start,
|
|
@@ -2119,7 +2181,10 @@ static int journal_read_transaction(struct super_block *sb,
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
trans_id = get_desc_trans_id(desc);
|
|
trans_id = get_desc_trans_id(desc);
|
|
|
- /* now we know we've got a good transaction, and it was inside the valid time ranges */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * now we know we've got a good transaction, and it was
|
|
|
|
|
+ * inside the valid time ranges
|
|
|
|
|
+ */
|
|
|
log_blocks = kmalloc(get_desc_trans_len(desc) *
|
|
log_blocks = kmalloc(get_desc_trans_len(desc) *
|
|
|
sizeof(struct buffer_head *), GFP_NOFS);
|
|
sizeof(struct buffer_head *), GFP_NOFS);
|
|
|
real_blocks = kmalloc(get_desc_trans_len(desc) *
|
|
real_blocks = kmalloc(get_desc_trans_len(desc) *
|
|
@@ -2164,7 +2229,7 @@ static int journal_read_transaction(struct super_block *sb,
|
|
|
reiserfs_warning(sb, "journal-1204",
|
|
reiserfs_warning(sb, "journal-1204",
|
|
|
"REPLAY FAILURE fsck required! "
|
|
"REPLAY FAILURE fsck required! "
|
|
|
"Trying to replay onto a log block");
|
|
"Trying to replay onto a log block");
|
|
|
- abort_replay:
|
|
|
|
|
|
|
+abort_replay:
|
|
|
brelse_array(log_blocks, i);
|
|
brelse_array(log_blocks, i);
|
|
|
brelse_array(real_blocks, i);
|
|
brelse_array(real_blocks, i);
|
|
|
brelse(c_bh);
|
|
brelse(c_bh);
|
|
@@ -2226,7 +2291,10 @@ static int journal_read_transaction(struct super_block *sb,
|
|
|
"journal-1095: setting journal " "start to offset %ld",
|
|
"journal-1095: setting journal " "start to offset %ld",
|
|
|
cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
|
|
cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
|
|
|
|
|
|
|
|
- /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * init starting values for the first transaction, in case
|
|
|
|
|
+ * this is the last transaction to be replayed.
|
|
|
|
|
+ */
|
|
|
journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
|
|
journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
|
|
|
journal->j_last_flush_trans_id = trans_id;
|
|
journal->j_last_flush_trans_id = trans_id;
|
|
|
journal->j_trans_id = trans_id + 1;
|
|
journal->j_trans_id = trans_id + 1;
|
|
@@ -2240,12 +2308,14 @@ static int journal_read_transaction(struct super_block *sb,
|
|
|
return 0;
|
|
return 0;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/* This function reads blocks starting from block and to max_block of bufsize
|
|
|
|
|
- size (but no more than BUFNR blocks at a time). This proved to improve
|
|
|
|
|
- mounting speed on self-rebuilding raid5 arrays at least.
|
|
|
|
|
- Right now it is only used from journal code. But later we might use it
|
|
|
|
|
- from other places.
|
|
|
|
|
- Note: Do not use journal_getblk/sb_getblk functions here! */
|
|
|
|
|
|
|
+/*
|
|
|
|
|
+ * This function reads blocks starting from block and to max_block of bufsize
|
|
|
|
|
+ * size (but no more than BUFNR blocks at a time). This proved to improve
|
|
|
|
|
+ * mounting speed on self-rebuilding raid5 arrays at least.
|
|
|
|
|
+ * Right now it is only used from journal code. But later we might use it
|
|
|
|
|
+ * from other places.
|
|
|
|
|
+ * Note: Do not use journal_getblk/sb_getblk functions here!
|
|
|
|
|
+ */
|
|
|
static struct buffer_head *reiserfs_breada(struct block_device *dev,
|
|
static struct buffer_head *reiserfs_breada(struct block_device *dev,
|
|
|
b_blocknr_t block, int bufsize,
|
|
b_blocknr_t block, int bufsize,
|
|
|
b_blocknr_t max_block)
|
|
b_blocknr_t max_block)
|
|
@@ -2284,15 +2354,17 @@ static struct buffer_head *reiserfs_breada(struct block_device *dev,
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** read and replay the log
|
|
|
|
|
-** on a clean unmount, the journal header's next unflushed pointer will
|
|
|
|
|
-** be to an invalid transaction. This tests that before finding all the
|
|
|
|
|
-** transactions in the log, which makes normal mount times fast.
|
|
|
|
|
-** After a crash, this starts with the next unflushed transaction, and
|
|
|
|
|
-** replays until it finds one too old, or invalid.
|
|
|
|
|
-** On exit, it sets things up so the first transaction will work correctly.
|
|
|
|
|
-** NOTE: only called during fs mount
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * read and replay the log
|
|
|
|
|
+ * on a clean unmount, the journal header's next unflushed pointer will be
|
|
|
|
|
+ * to an invalid transaction. This tests that before finding all the
|
|
|
|
|
+ * transactions in the log, which makes normal mount times fast.
|
|
|
|
|
+ *
|
|
|
|
|
+ * After a crash, this starts with the next unflushed transaction, and
|
|
|
|
|
+ * replays until it finds one too old, or invalid.
|
|
|
|
|
+ *
|
|
|
|
|
+ * On exit, it sets things up so the first transaction will work correctly.
|
|
|
|
|
+ * NOTE: only called during fs mount
|
|
|
|
|
+ */
|
|
|
static int journal_read(struct super_block *sb)
|
|
static int journal_read(struct super_block *sb)
|
|
|
{
|
|
{
|
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
@@ -2316,9 +2388,10 @@ static int journal_read(struct super_block *sb)
|
|
|
bdevname(journal->j_dev_bd, b));
|
|
bdevname(journal->j_dev_bd, b));
|
|
|
start = get_seconds();
|
|
start = get_seconds();
|
|
|
|
|
|
|
|
- /* step 1, read in the journal header block. Check the transaction it says
|
|
|
|
|
- ** is the first unflushed, and if that transaction is not valid,
|
|
|
|
|
- ** replay is done
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * step 1, read in the journal header block. Check the transaction
|
|
|
|
|
+ * it says is the first unflushed, and if that transaction is not
|
|
|
|
|
+ * valid, replay is done
|
|
|
*/
|
|
*/
|
|
|
journal->j_header_bh = journal_bread(sb,
|
|
journal->j_header_bh = journal_bread(sb,
|
|
|
SB_ONDISK_JOURNAL_1st_BLOCK(sb)
|
|
SB_ONDISK_JOURNAL_1st_BLOCK(sb)
|
|
@@ -2342,9 +2415,10 @@ static int journal_read(struct super_block *sb)
|
|
|
le32_to_cpu(jh->j_last_flush_trans_id));
|
|
le32_to_cpu(jh->j_last_flush_trans_id));
|
|
|
valid_journal_header = 1;
|
|
valid_journal_header = 1;
|
|
|
|
|
|
|
|
- /* now, we try to read the first unflushed offset. If it is not valid,
|
|
|
|
|
- ** there is nothing more we can do, and it makes no sense to read
|
|
|
|
|
- ** through the whole log.
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * now, we try to read the first unflushed offset. If it
|
|
|
|
|
+ * is not valid, there is nothing more we can do, and it
|
|
|
|
|
+ * makes no sense to read through the whole log.
|
|
|
*/
|
|
*/
|
|
|
d_bh =
|
|
d_bh =
|
|
|
journal_bread(sb,
|
|
journal_bread(sb,
|
|
@@ -2358,15 +2432,19 @@ static int journal_read(struct super_block *sb)
|
|
|
goto start_log_replay;
|
|
goto start_log_replay;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- /* ok, there are transactions that need to be replayed. start with the first log block, find
|
|
|
|
|
- ** all the valid transactions, and pick out the oldest.
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * ok, there are transactions that need to be replayed. start
|
|
|
|
|
+ * with the first log block, find all the valid transactions, and
|
|
|
|
|
+ * pick out the oldest.
|
|
|
*/
|
|
*/
|
|
|
while (continue_replay
|
|
while (continue_replay
|
|
|
&& cur_dblock <
|
|
&& cur_dblock <
|
|
|
(SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
|
|
(SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
|
|
|
SB_ONDISK_JOURNAL_SIZE(sb))) {
|
|
SB_ONDISK_JOURNAL_SIZE(sb))) {
|
|
|
- /* Note that it is required for blocksize of primary fs device and journal
|
|
|
|
|
- device to be the same */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * Note that it is required for blocksize of primary fs
|
|
|
|
|
+ * device and journal device to be the same
|
|
|
|
|
+ */
|
|
|
d_bh =
|
|
d_bh =
|
|
|
reiserfs_breada(journal->j_dev_bd, cur_dblock,
|
|
reiserfs_breada(journal->j_dev_bd, cur_dblock,
|
|
|
sb->s_blocksize,
|
|
sb->s_blocksize,
|
|
@@ -2413,7 +2491,7 @@ static int journal_read(struct super_block *sb)
|
|
|
brelse(d_bh);
|
|
brelse(d_bh);
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- start_log_replay:
|
|
|
|
|
|
|
+start_log_replay:
|
|
|
cur_dblock = oldest_start;
|
|
cur_dblock = oldest_start;
|
|
|
if (oldest_trans_id) {
|
|
if (oldest_trans_id) {
|
|
|
reiserfs_debug(sb, REISERFS_DEBUG_CODE,
|
|
reiserfs_debug(sb, REISERFS_DEBUG_CODE,
|
|
@@ -2444,9 +2522,11 @@ static int journal_read(struct super_block *sb)
|
|
|
reiserfs_debug(sb, REISERFS_DEBUG_CODE,
|
|
reiserfs_debug(sb, REISERFS_DEBUG_CODE,
|
|
|
"journal-1225: No valid " "transactions found");
|
|
"journal-1225: No valid " "transactions found");
|
|
|
}
|
|
}
|
|
|
- /* j_start does not get set correctly if we don't replay any transactions.
|
|
|
|
|
- ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
|
|
|
|
|
- ** copy the trans_id from the header
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * j_start does not get set correctly if we don't replay any
|
|
|
|
|
+ * transactions. if we had a valid journal_header, set j_start
|
|
|
|
|
+ * to the first unflushed transaction value, copy the trans_id
|
|
|
|
|
+ * from the header
|
|
|
*/
|
|
*/
|
|
|
if (valid_journal_header && replay_count == 0) {
|
|
if (valid_journal_header && replay_count == 0) {
|
|
|
journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
|
|
journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
|
|
@@ -2475,8 +2555,9 @@ static int journal_read(struct super_block *sb)
|
|
|
_update_journal_header_block(sb, journal->j_start,
|
|
_update_journal_header_block(sb, journal->j_start,
|
|
|
journal->j_last_flush_trans_id)) {
|
|
journal->j_last_flush_trans_id)) {
|
|
|
reiserfs_write_unlock(sb);
|
|
reiserfs_write_unlock(sb);
|
|
|
- /* replay failed, caller must call free_journal_ram and abort
|
|
|
|
|
- ** the mount
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * replay failed, caller must call free_journal_ram and abort
|
|
|
|
|
+ * the mount
|
|
|
*/
|
|
*/
|
|
|
return -1;
|
|
return -1;
|
|
|
}
|
|
}
|
|
@@ -2569,7 +2650,7 @@ static int journal_init_dev(struct super_block *super,
|
|
|
return 0;
|
|
return 0;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/**
|
|
|
|
|
|
|
+/*
|
|
|
* When creating/tuning a file system user can assign some
|
|
* When creating/tuning a file system user can assign some
|
|
|
* journal params within boundaries which depend on the ratio
|
|
* journal params within boundaries which depend on the ratio
|
|
|
* blocksize/standard_blocksize.
|
|
* blocksize/standard_blocksize.
|
|
@@ -2587,8 +2668,7 @@ static int check_advise_trans_params(struct super_block *sb,
|
|
|
struct reiserfs_journal *journal)
|
|
struct reiserfs_journal *journal)
|
|
|
{
|
|
{
|
|
|
if (journal->j_trans_max) {
|
|
if (journal->j_trans_max) {
|
|
|
- /* Non-default journal params.
|
|
|
|
|
- Do sanity check for them. */
|
|
|
|
|
|
|
+ /* Non-default journal params. Do sanity check for them. */
|
|
|
int ratio = 1;
|
|
int ratio = 1;
|
|
|
if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
|
|
if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
|
|
|
ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
|
|
ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
|
|
@@ -2610,10 +2690,12 @@ static int check_advise_trans_params(struct super_block *sb,
|
|
|
return 1;
|
|
return 1;
|
|
|
}
|
|
}
|
|
|
} else {
|
|
} else {
|
|
|
- /* Default journal params.
|
|
|
|
|
- The file system was created by old version
|
|
|
|
|
- of mkreiserfs, so some fields contain zeros,
|
|
|
|
|
- and we need to advise proper values for them */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * Default journal params.
|
|
|
|
|
+ * The file system was created by old version
|
|
|
|
|
+ * of mkreiserfs, so some fields contain zeros,
|
|
|
|
|
+ * and we need to advise proper values for them
|
|
|
|
|
+ */
|
|
|
if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
|
|
if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
|
|
|
reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
|
|
reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
|
|
|
sb->s_blocksize);
|
|
sb->s_blocksize);
|
|
@@ -2626,9 +2708,7 @@ static int check_advise_trans_params(struct super_block *sb,
|
|
|
return 0;
|
|
return 0;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/*
|
|
|
|
|
-** must be called once on fs mount. calls journal_read for you
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+/* must be called once on fs mount. calls journal_read for you */
|
|
|
int journal_init(struct super_block *sb, const char *j_dev_name,
|
|
int journal_init(struct super_block *sb, const char *j_dev_name,
|
|
|
int old_format, unsigned int commit_max_age)
|
|
int old_format, unsigned int commit_max_age)
|
|
|
{
|
|
{
|
|
@@ -2667,8 +2747,10 @@ int journal_init(struct super_block *sb, const char *j_dev_name,
|
|
|
REISERFS_DISK_OFFSET_IN_BYTES /
|
|
REISERFS_DISK_OFFSET_IN_BYTES /
|
|
|
sb->s_blocksize + 2);
|
|
sb->s_blocksize + 2);
|
|
|
|
|
|
|
|
- /* Sanity check to see is the standard journal fitting within first bitmap
|
|
|
|
|
- (actual for small blocksizes) */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * Sanity check to see is the standard journal fitting
|
|
|
|
|
+ * within first bitmap (actual for small blocksizes)
|
|
|
|
|
+ */
|
|
|
if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
|
|
if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
|
|
|
(SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
|
|
(SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
|
|
|
SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
|
|
SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
|
|
@@ -2754,20 +2836,20 @@ int journal_init(struct super_block *sb, const char *j_dev_name,
|
|
|
journal->j_start = 0;
|
|
journal->j_start = 0;
|
|
|
journal->j_len = 0;
|
|
journal->j_len = 0;
|
|
|
journal->j_len_alloc = 0;
|
|
journal->j_len_alloc = 0;
|
|
|
- atomic_set(&(journal->j_wcount), 0);
|
|
|
|
|
- atomic_set(&(journal->j_async_throttle), 0);
|
|
|
|
|
|
|
+ atomic_set(&journal->j_wcount, 0);
|
|
|
|
|
+ atomic_set(&journal->j_async_throttle, 0);
|
|
|
journal->j_bcount = 0;
|
|
journal->j_bcount = 0;
|
|
|
journal->j_trans_start_time = 0;
|
|
journal->j_trans_start_time = 0;
|
|
|
journal->j_last = NULL;
|
|
journal->j_last = NULL;
|
|
|
journal->j_first = NULL;
|
|
journal->j_first = NULL;
|
|
|
- init_waitqueue_head(&(journal->j_join_wait));
|
|
|
|
|
|
|
+ init_waitqueue_head(&journal->j_join_wait);
|
|
|
mutex_init(&journal->j_mutex);
|
|
mutex_init(&journal->j_mutex);
|
|
|
mutex_init(&journal->j_flush_mutex);
|
|
mutex_init(&journal->j_flush_mutex);
|
|
|
|
|
|
|
|
journal->j_trans_id = 10;
|
|
journal->j_trans_id = 10;
|
|
|
journal->j_mount_id = 10;
|
|
journal->j_mount_id = 10;
|
|
|
journal->j_state = 0;
|
|
journal->j_state = 0;
|
|
|
- atomic_set(&(journal->j_jlock), 0);
|
|
|
|
|
|
|
+ atomic_set(&journal->j_jlock, 0);
|
|
|
journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
|
|
journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
|
|
|
journal->j_cnode_free_orig = journal->j_cnode_free_list;
|
|
journal->j_cnode_free_orig = journal->j_cnode_free_list;
|
|
|
journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
|
|
journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
|
|
@@ -2807,23 +2889,19 @@ int journal_init(struct super_block *sb, const char *j_dev_name,
|
|
|
goto free_and_return;
|
|
goto free_and_return;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- reiserfs_mounted_fs_count++;
|
|
|
|
|
- if (reiserfs_mounted_fs_count <= 1)
|
|
|
|
|
- commit_wq = alloc_workqueue("reiserfs", WQ_MEM_RECLAIM, 0);
|
|
|
|
|
-
|
|
|
|
|
INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
|
|
INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
|
|
|
journal->j_work_sb = sb;
|
|
journal->j_work_sb = sb;
|
|
|
return 0;
|
|
return 0;
|
|
|
- free_and_return:
|
|
|
|
|
|
|
+free_and_return:
|
|
|
free_journal_ram(sb);
|
|
free_journal_ram(sb);
|
|
|
return 1;
|
|
return 1;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** test for a polite end of the current transaction. Used by file_write, and should
|
|
|
|
|
-** be used by delete to make sure they don't write more than can fit inside a single
|
|
|
|
|
-** transaction
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * test for a polite end of the current transaction. Used by file_write,
|
|
|
|
|
+ * and should be used by delete to make sure they don't write more than
|
|
|
|
|
+ * can fit inside a single transaction
|
|
|
|
|
+ */
|
|
|
int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
|
|
int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
|
|
|
int new_alloc)
|
|
int new_alloc)
|
|
|
{
|
|
{
|
|
@@ -2835,7 +2913,7 @@ int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
|
|
|
return 0;
|
|
return 0;
|
|
|
if (journal->j_must_wait > 0 ||
|
|
if (journal->j_must_wait > 0 ||
|
|
|
(journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
|
|
(journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
|
|
|
- atomic_read(&(journal->j_jlock)) ||
|
|
|
|
|
|
|
+ atomic_read(&journal->j_jlock) ||
|
|
|
(now - journal->j_trans_start_time) > journal->j_max_trans_age ||
|
|
(now - journal->j_trans_start_time) > journal->j_max_trans_age ||
|
|
|
journal->j_cnode_free < (journal->j_trans_max * 3)) {
|
|
journal->j_cnode_free < (journal->j_trans_max * 3)) {
|
|
|
return 1;
|
|
return 1;
|
|
@@ -2846,8 +2924,7 @@ int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
|
|
|
return 0;
|
|
return 0;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/* this must be called inside a transaction
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+/* this must be called inside a transaction */
|
|
|
void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
|
|
void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
|
|
|
{
|
|
{
|
|
|
struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
|
|
struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
|
|
@@ -2857,8 +2934,7 @@ void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
|
|
|
return;
|
|
return;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/* this must be called without a transaction started
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+/* this must be called without a transaction started */
|
|
|
void reiserfs_allow_writes(struct super_block *s)
|
|
void reiserfs_allow_writes(struct super_block *s)
|
|
|
{
|
|
{
|
|
|
struct reiserfs_journal *journal = SB_JOURNAL(s);
|
|
struct reiserfs_journal *journal = SB_JOURNAL(s);
|
|
@@ -2866,8 +2942,7 @@ void reiserfs_allow_writes(struct super_block *s)
|
|
|
wake_up(&journal->j_join_wait);
|
|
wake_up(&journal->j_join_wait);
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/* this must be called without a transaction started
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+/* this must be called without a transaction started */
|
|
|
void reiserfs_wait_on_write_block(struct super_block *s)
|
|
void reiserfs_wait_on_write_block(struct super_block *s)
|
|
|
{
|
|
{
|
|
|
struct reiserfs_journal *journal = SB_JOURNAL(s);
|
|
struct reiserfs_journal *journal = SB_JOURNAL(s);
|
|
@@ -2929,11 +3004,12 @@ static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
|
|
|
}
|
|
}
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/* join == true if you must join an existing transaction.
|
|
|
|
|
-** join == false if you can deal with waiting for others to finish
|
|
|
|
|
-**
|
|
|
|
|
-** this will block until the transaction is joinable. send the number of blocks you
|
|
|
|
|
-** expect to use in nblocks.
|
|
|
|
|
|
|
+/*
|
|
|
|
|
+ * join == true if you must join an existing transaction.
|
|
|
|
|
+ * join == false if you can deal with waiting for others to finish
|
|
|
|
|
+ *
|
|
|
|
|
+ * this will block until the transaction is joinable. send the number of
|
|
|
|
|
+ * blocks you expect to use in nblocks.
|
|
|
*/
|
|
*/
|
|
|
static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
|
|
static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
|
|
|
struct super_block *sb, unsigned long nblocks,
|
|
struct super_block *sb, unsigned long nblocks,
|
|
@@ -2955,7 +3031,7 @@ static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
|
|
|
th->t_refcount = 1;
|
|
th->t_refcount = 1;
|
|
|
th->t_super = sb;
|
|
th->t_super = sb;
|
|
|
|
|
|
|
|
- relock:
|
|
|
|
|
|
|
+relock:
|
|
|
lock_journal(sb);
|
|
lock_journal(sb);
|
|
|
if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
|
|
if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
|
|
|
unlock_journal(sb);
|
|
unlock_journal(sb);
|
|
@@ -2974,9 +3050,11 @@ static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
|
|
|
}
|
|
}
|
|
|
now = get_seconds();
|
|
now = get_seconds();
|
|
|
|
|
|
|
|
- /* if there is no room in the journal OR
|
|
|
|
|
- ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
|
|
|
|
|
- ** we don't sleep if there aren't other writers
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * if there is no room in the journal OR
|
|
|
|
|
+ * if this transaction is too old, and we weren't called joinable,
|
|
|
|
|
+ * wait for it to finish before beginning we don't sleep if there
|
|
|
|
|
+ * aren't other writers
|
|
|
*/
|
|
*/
|
|
|
|
|
|
|
|
if ((!join && journal->j_must_wait > 0) ||
|
|
if ((!join && journal->j_must_wait > 0) ||
|
|
@@ -2990,7 +3068,8 @@ static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
|
|
|
|| (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
|
|
|| (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
|
|
|
|
|
|
|
|
old_trans_id = journal->j_trans_id;
|
|
old_trans_id = journal->j_trans_id;
|
|
|
- unlock_journal(sb); /* allow others to finish this transaction */
|
|
|
|
|
|
|
+ /* allow others to finish this transaction */
|
|
|
|
|
+ unlock_journal(sb);
|
|
|
|
|
|
|
|
if (!join && (journal->j_len_alloc + nblocks + 2) >=
|
|
if (!join && (journal->j_len_alloc + nblocks + 2) >=
|
|
|
journal->j_max_batch &&
|
|
journal->j_max_batch &&
|
|
@@ -3002,8 +3081,9 @@ static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
|
|
|
goto relock;
|
|
goto relock;
|
|
|
}
|
|
}
|
|
|
}
|
|
}
|
|
|
- /* don't mess with joining the transaction if all we have to do is
|
|
|
|
|
- * wait for someone else to do a commit
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * don't mess with joining the transaction if all we
|
|
|
|
|
+ * have to do is wait for someone else to do a commit
|
|
|
*/
|
|
*/
|
|
|
if (atomic_read(&journal->j_jlock)) {
|
|
if (atomic_read(&journal->j_jlock)) {
|
|
|
while (journal->j_trans_id == old_trans_id &&
|
|
while (journal->j_trans_id == old_trans_id &&
|
|
@@ -3012,15 +3092,15 @@ static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
|
|
|
}
|
|
}
|
|
|
goto relock;
|
|
goto relock;
|
|
|
}
|
|
}
|
|
|
- retval = journal_join(&myth, sb, 1);
|
|
|
|
|
|
|
+ retval = journal_join(&myth, sb);
|
|
|
if (retval)
|
|
if (retval)
|
|
|
goto out_fail;
|
|
goto out_fail;
|
|
|
|
|
|
|
|
/* someone might have ended the transaction while we joined */
|
|
/* someone might have ended the transaction while we joined */
|
|
|
if (old_trans_id != journal->j_trans_id) {
|
|
if (old_trans_id != journal->j_trans_id) {
|
|
|
- retval = do_journal_end(&myth, sb, 1, 0);
|
|
|
|
|
|
|
+ retval = do_journal_end(&myth, 0);
|
|
|
} else {
|
|
} else {
|
|
|
- retval = do_journal_end(&myth, sb, 1, COMMIT_NOW);
|
|
|
|
|
|
|
+ retval = do_journal_end(&myth, COMMIT_NOW);
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
if (retval)
|
|
if (retval)
|
|
@@ -3033,7 +3113,7 @@ static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
|
|
|
if (journal->j_trans_start_time == 0) {
|
|
if (journal->j_trans_start_time == 0) {
|
|
|
journal->j_trans_start_time = get_seconds();
|
|
journal->j_trans_start_time = get_seconds();
|
|
|
}
|
|
}
|
|
|
- atomic_inc(&(journal->j_wcount));
|
|
|
|
|
|
|
+ atomic_inc(&journal->j_wcount);
|
|
|
journal->j_len_alloc += nblocks;
|
|
journal->j_len_alloc += nblocks;
|
|
|
th->t_blocks_logged = 0;
|
|
th->t_blocks_logged = 0;
|
|
|
th->t_blocks_allocated = nblocks;
|
|
th->t_blocks_allocated = nblocks;
|
|
@@ -3042,11 +3122,13 @@ static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
|
|
|
INIT_LIST_HEAD(&th->t_list);
|
|
INIT_LIST_HEAD(&th->t_list);
|
|
|
return 0;
|
|
return 0;
|
|
|
|
|
|
|
|
- out_fail:
|
|
|
|
|
|
|
+out_fail:
|
|
|
memset(th, 0, sizeof(*th));
|
|
memset(th, 0, sizeof(*th));
|
|
|
- /* Re-set th->t_super, so we can properly keep track of how many
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * Re-set th->t_super, so we can properly keep track of how many
|
|
|
* persistent transactions there are. We need to do this so if this
|
|
* persistent transactions there are. We need to do this so if this
|
|
|
- * call is part of a failed restart_transaction, we can free it later */
|
|
|
|
|
|
|
+ * call is part of a failed restart_transaction, we can free it later
|
|
|
|
|
+ */
|
|
|
th->t_super = sb;
|
|
th->t_super = sb;
|
|
|
return retval;
|
|
return retval;
|
|
|
}
|
|
}
|
|
@@ -3059,14 +3141,15 @@ struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
|
|
|
int ret;
|
|
int ret;
|
|
|
struct reiserfs_transaction_handle *th;
|
|
struct reiserfs_transaction_handle *th;
|
|
|
|
|
|
|
|
- /* if we're nesting into an existing transaction. It will be
|
|
|
|
|
- ** persistent on its own
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * if we're nesting into an existing transaction. It will be
|
|
|
|
|
+ * persistent on its own
|
|
|
*/
|
|
*/
|
|
|
if (reiserfs_transaction_running(s)) {
|
|
if (reiserfs_transaction_running(s)) {
|
|
|
th = current->journal_info;
|
|
th = current->journal_info;
|
|
|
th->t_refcount++;
|
|
th->t_refcount++;
|
|
|
BUG_ON(th->t_refcount < 2);
|
|
BUG_ON(th->t_refcount < 2);
|
|
|
-
|
|
|
|
|
|
|
+
|
|
|
return th;
|
|
return th;
|
|
|
}
|
|
}
|
|
|
th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
|
|
th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
|
|
@@ -3087,7 +3170,7 @@ int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
|
|
|
struct super_block *s = th->t_super;
|
|
struct super_block *s = th->t_super;
|
|
|
int ret = 0;
|
|
int ret = 0;
|
|
|
if (th->t_trans_id)
|
|
if (th->t_trans_id)
|
|
|
- ret = journal_end(th, th->t_super, th->t_blocks_allocated);
|
|
|
|
|
|
|
+ ret = journal_end(th);
|
|
|
else
|
|
else
|
|
|
ret = -EIO;
|
|
ret = -EIO;
|
|
|
if (th->t_refcount == 0) {
|
|
if (th->t_refcount == 0) {
|
|
@@ -3098,29 +3181,31 @@ int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
static int journal_join(struct reiserfs_transaction_handle *th,
|
|
static int journal_join(struct reiserfs_transaction_handle *th,
|
|
|
- struct super_block *sb, unsigned long nblocks)
|
|
|
|
|
|
|
+ struct super_block *sb)
|
|
|
{
|
|
{
|
|
|
struct reiserfs_transaction_handle *cur_th = current->journal_info;
|
|
struct reiserfs_transaction_handle *cur_th = current->journal_info;
|
|
|
|
|
|
|
|
- /* this keeps do_journal_end from NULLing out the current->journal_info
|
|
|
|
|
- ** pointer
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * this keeps do_journal_end from NULLing out the
|
|
|
|
|
+ * current->journal_info pointer
|
|
|
*/
|
|
*/
|
|
|
th->t_handle_save = cur_th;
|
|
th->t_handle_save = cur_th;
|
|
|
BUG_ON(cur_th && cur_th->t_refcount > 1);
|
|
BUG_ON(cur_th && cur_th->t_refcount > 1);
|
|
|
- return do_journal_begin_r(th, sb, nblocks, JBEGIN_JOIN);
|
|
|
|
|
|
|
+ return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN);
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
int journal_join_abort(struct reiserfs_transaction_handle *th,
|
|
int journal_join_abort(struct reiserfs_transaction_handle *th,
|
|
|
- struct super_block *sb, unsigned long nblocks)
|
|
|
|
|
|
|
+ struct super_block *sb)
|
|
|
{
|
|
{
|
|
|
struct reiserfs_transaction_handle *cur_th = current->journal_info;
|
|
struct reiserfs_transaction_handle *cur_th = current->journal_info;
|
|
|
|
|
|
|
|
- /* this keeps do_journal_end from NULLing out the current->journal_info
|
|
|
|
|
- ** pointer
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * this keeps do_journal_end from NULLing out the
|
|
|
|
|
+ * current->journal_info pointer
|
|
|
*/
|
|
*/
|
|
|
th->t_handle_save = cur_th;
|
|
th->t_handle_save = cur_th;
|
|
|
BUG_ON(cur_th && cur_th->t_refcount > 1);
|
|
BUG_ON(cur_th && cur_th->t_refcount > 1);
|
|
|
- return do_journal_begin_r(th, sb, nblocks, JBEGIN_ABORT);
|
|
|
|
|
|
|
+ return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT);
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
int journal_begin(struct reiserfs_transaction_handle *th,
|
|
int journal_begin(struct reiserfs_transaction_handle *th,
|
|
@@ -3142,9 +3227,10 @@ int journal_begin(struct reiserfs_transaction_handle *th,
|
|
|
"journal_info != 0");
|
|
"journal_info != 0");
|
|
|
return 0;
|
|
return 0;
|
|
|
} else {
|
|
} else {
|
|
|
- /* we've ended up with a handle from a different filesystem.
|
|
|
|
|
- ** save it and restore on journal_end. This should never
|
|
|
|
|
- ** really happen...
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * we've ended up with a handle from a different
|
|
|
|
|
+ * filesystem. save it and restore on journal_end.
|
|
|
|
|
+ * This should never really happen...
|
|
|
*/
|
|
*/
|
|
|
reiserfs_warning(sb, "clm-2100",
|
|
reiserfs_warning(sb, "clm-2100",
|
|
|
"nesting info a different FS");
|
|
"nesting info a different FS");
|
|
@@ -3157,9 +3243,10 @@ int journal_begin(struct reiserfs_transaction_handle *th,
|
|
|
ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
|
|
ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
|
|
|
BUG_ON(current->journal_info != th);
|
|
BUG_ON(current->journal_info != th);
|
|
|
|
|
|
|
|
- /* I guess this boils down to being the reciprocal of clm-2100 above.
|
|
|
|
|
- * If do_journal_begin_r fails, we need to put it back, since journal_end
|
|
|
|
|
- * won't be called to do it. */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * I guess this boils down to being the reciprocal of clm-2100 above.
|
|
|
|
|
+ * If do_journal_begin_r fails, we need to put it back, since
|
|
|
|
|
+ * journal_end won't be called to do it. */
|
|
|
if (ret)
|
|
if (ret)
|
|
|
current->journal_info = th->t_handle_save;
|
|
current->journal_info = th->t_handle_save;
|
|
|
else
|
|
else
|
|
@@ -3169,17 +3256,19 @@ int journal_begin(struct reiserfs_transaction_handle *th,
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** puts bh into the current transaction. If it was already there, reorders removes the
|
|
|
|
|
-** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
|
|
|
|
|
-**
|
|
|
|
|
-** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the
|
|
|
|
|
-** transaction is committed.
|
|
|
|
|
-**
|
|
|
|
|
-** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * puts bh into the current transaction. If it was already there, reorders
|
|
|
|
|
+ * removes the old pointers from the hash, and puts new ones in (to make
|
|
|
|
|
+ * sure replay happen in the right order).
|
|
|
|
|
+ *
|
|
|
|
|
+ * if it was dirty, cleans and files onto the clean list. I can't let it
|
|
|
|
|
+ * be dirty again until the transaction is committed.
|
|
|
|
|
+ *
|
|
|
|
|
+ * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
|
|
|
|
|
+ */
|
|
|
int journal_mark_dirty(struct reiserfs_transaction_handle *th,
|
|
int journal_mark_dirty(struct reiserfs_transaction_handle *th,
|
|
|
- struct super_block *sb, struct buffer_head *bh)
|
|
|
|
|
|
|
+ struct buffer_head *bh)
|
|
|
{
|
|
{
|
|
|
|
|
+ struct super_block *sb = th->t_super;
|
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
|
struct reiserfs_journal_cnode *cn = NULL;
|
|
struct reiserfs_journal_cnode *cn = NULL;
|
|
|
int count_already_incd = 0;
|
|
int count_already_incd = 0;
|
|
@@ -3201,9 +3290,10 @@ int journal_mark_dirty(struct reiserfs_transaction_handle *th,
|
|
|
return 0;
|
|
return 0;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- /* this must be turned into a panic instead of a warning. We can't allow
|
|
|
|
|
- ** a dirty or journal_dirty or locked buffer to be logged, as some changes
|
|
|
|
|
- ** could get to disk too early. NOT GOOD.
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * this must be turned into a panic instead of a warning. We can't
|
|
|
|
|
+ * allow a dirty or journal_dirty or locked buffer to be logged, as
|
|
|
|
|
+ * some changes could get to disk too early. NOT GOOD.
|
|
|
*/
|
|
*/
|
|
|
if (!prepared || buffer_dirty(bh)) {
|
|
if (!prepared || buffer_dirty(bh)) {
|
|
|
reiserfs_warning(sb, "journal-1777",
|
|
reiserfs_warning(sb, "journal-1777",
|
|
@@ -3216,14 +3306,16 @@ int journal_mark_dirty(struct reiserfs_transaction_handle *th,
|
|
|
buffer_journal_dirty(bh) ? ' ' : '!');
|
|
buffer_journal_dirty(bh) ? ' ' : '!');
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- if (atomic_read(&(journal->j_wcount)) <= 0) {
|
|
|
|
|
|
|
+ if (atomic_read(&journal->j_wcount) <= 0) {
|
|
|
reiserfs_warning(sb, "journal-1409",
|
|
reiserfs_warning(sb, "journal-1409",
|
|
|
"returning because j_wcount was %d",
|
|
"returning because j_wcount was %d",
|
|
|
- atomic_read(&(journal->j_wcount)));
|
|
|
|
|
|
|
+ atomic_read(&journal->j_wcount));
|
|
|
return 1;
|
|
return 1;
|
|
|
}
|
|
}
|
|
|
- /* this error means I've screwed up, and we've overflowed the transaction.
|
|
|
|
|
- ** Nothing can be done here, except make the FS readonly or panic.
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * this error means I've screwed up, and we've overflowed
|
|
|
|
|
+ * the transaction. Nothing can be done here, except make the
|
|
|
|
|
+ * FS readonly or panic.
|
|
|
*/
|
|
*/
|
|
|
if (journal->j_len >= journal->j_trans_max) {
|
|
if (journal->j_len >= journal->j_trans_max) {
|
|
|
reiserfs_panic(th->t_super, "journal-1413",
|
|
reiserfs_panic(th->t_super, "journal-1413",
|
|
@@ -3280,9 +3372,9 @@ int journal_mark_dirty(struct reiserfs_transaction_handle *th,
|
|
|
return 0;
|
|
return 0;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-int journal_end(struct reiserfs_transaction_handle *th,
|
|
|
|
|
- struct super_block *sb, unsigned long nblocks)
|
|
|
|
|
|
|
+int journal_end(struct reiserfs_transaction_handle *th)
|
|
|
{
|
|
{
|
|
|
|
|
+ struct super_block *sb = th->t_super;
|
|
|
if (!current->journal_info && th->t_refcount > 1)
|
|
if (!current->journal_info && th->t_refcount > 1)
|
|
|
reiserfs_warning(sb, "REISER-NESTING",
|
|
reiserfs_warning(sb, "REISER-NESTING",
|
|
|
"th NULL, refcount %d", th->t_refcount);
|
|
"th NULL, refcount %d", th->t_refcount);
|
|
@@ -3297,8 +3389,9 @@ int journal_end(struct reiserfs_transaction_handle *th,
|
|
|
struct reiserfs_transaction_handle *cur_th =
|
|
struct reiserfs_transaction_handle *cur_th =
|
|
|
current->journal_info;
|
|
current->journal_info;
|
|
|
|
|
|
|
|
- /* we aren't allowed to close a nested transaction on a different
|
|
|
|
|
- ** filesystem from the one in the task struct
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * we aren't allowed to close a nested transaction on a
|
|
|
|
|
+ * different filesystem from the one in the task struct
|
|
|
*/
|
|
*/
|
|
|
BUG_ON(cur_th->t_super != th->t_super);
|
|
BUG_ON(cur_th->t_super != th->t_super);
|
|
|
|
|
|
|
@@ -3308,17 +3401,18 @@ int journal_end(struct reiserfs_transaction_handle *th,
|
|
|
}
|
|
}
|
|
|
return 0;
|
|
return 0;
|
|
|
} else {
|
|
} else {
|
|
|
- return do_journal_end(th, sb, nblocks, 0);
|
|
|
|
|
|
|
+ return do_journal_end(th, 0);
|
|
|
}
|
|
}
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/* removes from the current transaction, relsing and descrementing any counters.
|
|
|
|
|
-** also files the removed buffer directly onto the clean list
|
|
|
|
|
-**
|
|
|
|
|
-** called by journal_mark_freed when a block has been deleted
|
|
|
|
|
-**
|
|
|
|
|
-** returns 1 if it cleaned and relsed the buffer. 0 otherwise
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+/*
|
|
|
|
|
+ * removes from the current transaction, relsing and descrementing any counters.
|
|
|
|
|
+ * also files the removed buffer directly onto the clean list
|
|
|
|
|
+ *
|
|
|
|
|
+ * called by journal_mark_freed when a block has been deleted
|
|
|
|
|
+ *
|
|
|
|
|
+ * returns 1 if it cleaned and relsed the buffer. 0 otherwise
|
|
|
|
|
+ */
|
|
|
static int remove_from_transaction(struct super_block *sb,
|
|
static int remove_from_transaction(struct super_block *sb,
|
|
|
b_blocknr_t blocknr, int already_cleaned)
|
|
b_blocknr_t blocknr, int already_cleaned)
|
|
|
{
|
|
{
|
|
@@ -3354,7 +3448,7 @@ static int remove_from_transaction(struct super_block *sb,
|
|
|
clear_buffer_dirty(bh);
|
|
clear_buffer_dirty(bh);
|
|
|
clear_buffer_journal_test(bh);
|
|
clear_buffer_journal_test(bh);
|
|
|
put_bh(bh);
|
|
put_bh(bh);
|
|
|
- if (atomic_read(&(bh->b_count)) < 0) {
|
|
|
|
|
|
|
+ if (atomic_read(&bh->b_count) < 0) {
|
|
|
reiserfs_warning(sb, "journal-1752",
|
|
reiserfs_warning(sb, "journal-1752",
|
|
|
"b_count < 0");
|
|
"b_count < 0");
|
|
|
}
|
|
}
|
|
@@ -3367,15 +3461,16 @@ static int remove_from_transaction(struct super_block *sb,
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** for any cnode in a journal list, it can only be dirtied of all the
|
|
|
|
|
-** transactions that include it are committed to disk.
|
|
|
|
|
-** this checks through each transaction, and returns 1 if you are allowed to dirty,
|
|
|
|
|
-** and 0 if you aren't
|
|
|
|
|
-**
|
|
|
|
|
-** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
|
|
|
|
|
-** blocks for a given transaction on disk
|
|
|
|
|
-**
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * for any cnode in a journal list, it can only be dirtied of all the
|
|
|
|
|
+ * transactions that include it are committed to disk.
|
|
|
|
|
+ * this checks through each transaction, and returns 1 if you are allowed
|
|
|
|
|
+ * to dirty, and 0 if you aren't
|
|
|
|
|
+ *
|
|
|
|
|
+ * it is called by dirty_journal_list, which is called after
|
|
|
|
|
+ * flush_commit_list has gotten all the log blocks for a given
|
|
|
|
|
+ * transaction on disk
|
|
|
|
|
+ *
|
|
|
|
|
+ */
|
|
|
static int can_dirty(struct reiserfs_journal_cnode *cn)
|
|
static int can_dirty(struct reiserfs_journal_cnode *cn)
|
|
|
{
|
|
{
|
|
|
struct super_block *sb = cn->sb;
|
|
struct super_block *sb = cn->sb;
|
|
@@ -3383,9 +3478,10 @@ static int can_dirty(struct reiserfs_journal_cnode *cn)
|
|
|
struct reiserfs_journal_cnode *cur = cn->hprev;
|
|
struct reiserfs_journal_cnode *cur = cn->hprev;
|
|
|
int can_dirty = 1;
|
|
int can_dirty = 1;
|
|
|
|
|
|
|
|
- /* first test hprev. These are all newer than cn, so any node here
|
|
|
|
|
- ** with the same block number and dev means this node can't be sent
|
|
|
|
|
- ** to disk right now.
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * first test hprev. These are all newer than cn, so any node here
|
|
|
|
|
+ * with the same block number and dev means this node can't be sent
|
|
|
|
|
+ * to disk right now.
|
|
|
*/
|
|
*/
|
|
|
while (cur && can_dirty) {
|
|
while (cur && can_dirty) {
|
|
|
if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
|
|
if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
|
|
@@ -3394,13 +3490,14 @@ static int can_dirty(struct reiserfs_journal_cnode *cn)
|
|
|
}
|
|
}
|
|
|
cur = cur->hprev;
|
|
cur = cur->hprev;
|
|
|
}
|
|
}
|
|
|
- /* then test hnext. These are all older than cn. As long as they
|
|
|
|
|
- ** are committed to the log, it is safe to write cn to disk
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * then test hnext. These are all older than cn. As long as they
|
|
|
|
|
+ * are committed to the log, it is safe to write cn to disk
|
|
|
*/
|
|
*/
|
|
|
cur = cn->hnext;
|
|
cur = cn->hnext;
|
|
|
while (cur && can_dirty) {
|
|
while (cur && can_dirty) {
|
|
|
if (cur->jlist && cur->jlist->j_len > 0 &&
|
|
if (cur->jlist && cur->jlist->j_len > 0 &&
|
|
|
- atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
|
|
|
|
|
|
|
+ atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh &&
|
|
|
cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
|
|
cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
|
|
|
can_dirty = 0;
|
|
can_dirty = 0;
|
|
|
}
|
|
}
|
|
@@ -3409,12 +3506,13 @@ static int can_dirty(struct reiserfs_journal_cnode *cn)
|
|
|
return can_dirty;
|
|
return can_dirty;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/* syncs the commit blocks, but does not force the real buffers to disk
|
|
|
|
|
-** will wait until the current transaction is done/committed before returning
|
|
|
|
|
-*/
|
|
|
|
|
-int journal_end_sync(struct reiserfs_transaction_handle *th,
|
|
|
|
|
- struct super_block *sb, unsigned long nblocks)
|
|
|
|
|
|
|
+/*
|
|
|
|
|
+ * syncs the commit blocks, but does not force the real buffers to disk
|
|
|
|
|
+ * will wait until the current transaction is done/committed before returning
|
|
|
|
|
+ */
|
|
|
|
|
+int journal_end_sync(struct reiserfs_transaction_handle *th)
|
|
|
{
|
|
{
|
|
|
|
|
+ struct super_block *sb = th->t_super;
|
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
|
|
|
|
|
|
BUG_ON(!th->t_trans_id);
|
|
BUG_ON(!th->t_trans_id);
|
|
@@ -3423,14 +3521,12 @@ int journal_end_sync(struct reiserfs_transaction_handle *th,
|
|
|
if (journal->j_len == 0) {
|
|
if (journal->j_len == 0) {
|
|
|
reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
|
|
reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
|
|
|
1);
|
|
1);
|
|
|
- journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
|
|
|
|
|
|
|
+ journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
|
|
|
}
|
|
}
|
|
|
- return do_journal_end(th, sb, nblocks, COMMIT_NOW | WAIT);
|
|
|
|
|
|
|
+ return do_journal_end(th, COMMIT_NOW | WAIT);
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
-/*
|
|
|
|
|
-** writeback the pending async commits to disk
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+/* writeback the pending async commits to disk */
|
|
|
static void flush_async_commits(struct work_struct *work)
|
|
static void flush_async_commits(struct work_struct *work)
|
|
|
{
|
|
{
|
|
|
struct reiserfs_journal *journal =
|
|
struct reiserfs_journal *journal =
|
|
@@ -3450,9 +3546,9 @@ static void flush_async_commits(struct work_struct *work)
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** flushes any old transactions to disk
|
|
|
|
|
-** ends the current transaction if it is too old
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * flushes any old transactions to disk
|
|
|
|
|
+ * ends the current transaction if it is too old
|
|
|
|
|
+ */
|
|
|
void reiserfs_flush_old_commits(struct super_block *sb)
|
|
void reiserfs_flush_old_commits(struct super_block *sb)
|
|
|
{
|
|
{
|
|
|
time_t now;
|
|
time_t now;
|
|
@@ -3460,48 +3556,53 @@ void reiserfs_flush_old_commits(struct super_block *sb)
|
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
|
|
|
|
|
|
now = get_seconds();
|
|
now = get_seconds();
|
|
|
- /* safety check so we don't flush while we are replaying the log during
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * safety check so we don't flush while we are replaying the log during
|
|
|
* mount
|
|
* mount
|
|
|
*/
|
|
*/
|
|
|
if (list_empty(&journal->j_journal_list))
|
|
if (list_empty(&journal->j_journal_list))
|
|
|
return;
|
|
return;
|
|
|
|
|
|
|
|
- /* check the current transaction. If there are no writers, and it is
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * check the current transaction. If there are no writers, and it is
|
|
|
* too old, finish it, and force the commit blocks to disk
|
|
* too old, finish it, and force the commit blocks to disk
|
|
|
*/
|
|
*/
|
|
|
if (atomic_read(&journal->j_wcount) <= 0 &&
|
|
if (atomic_read(&journal->j_wcount) <= 0 &&
|
|
|
journal->j_trans_start_time > 0 &&
|
|
journal->j_trans_start_time > 0 &&
|
|
|
journal->j_len > 0 &&
|
|
journal->j_len > 0 &&
|
|
|
(now - journal->j_trans_start_time) > journal->j_max_trans_age) {
|
|
(now - journal->j_trans_start_time) > journal->j_max_trans_age) {
|
|
|
- if (!journal_join(&th, sb, 1)) {
|
|
|
|
|
|
|
+ if (!journal_join(&th, sb)) {
|
|
|
reiserfs_prepare_for_journal(sb,
|
|
reiserfs_prepare_for_journal(sb,
|
|
|
SB_BUFFER_WITH_SB(sb),
|
|
SB_BUFFER_WITH_SB(sb),
|
|
|
1);
|
|
1);
|
|
|
- journal_mark_dirty(&th, sb,
|
|
|
|
|
- SB_BUFFER_WITH_SB(sb));
|
|
|
|
|
|
|
+ journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
|
|
|
|
|
|
|
|
- /* we're only being called from kreiserfsd, it makes no sense to do
|
|
|
|
|
- ** an async commit so that kreiserfsd can do it later
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * we're only being called from kreiserfsd, it makes
|
|
|
|
|
+ * no sense to do an async commit so that kreiserfsd
|
|
|
|
|
+ * can do it later
|
|
|
*/
|
|
*/
|
|
|
- do_journal_end(&th, sb, 1, COMMIT_NOW | WAIT);
|
|
|
|
|
|
|
+ do_journal_end(&th, COMMIT_NOW | WAIT);
|
|
|
}
|
|
}
|
|
|
}
|
|
}
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
|
|
|
|
|
-**
|
|
|
|
|
-** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
|
|
|
|
|
-** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just
|
|
|
|
|
-** flushes the commit list and returns 0.
|
|
|
|
|
-**
|
|
|
|
|
-** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait.
|
|
|
|
|
-**
|
|
|
|
|
-** Note, we can't allow the journal_end to proceed while there are still writers in the log.
|
|
|
|
|
-*/
|
|
|
|
|
-static int check_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
|
|
- struct super_block *sb, unsigned long nblocks,
|
|
|
|
|
- int flags)
|
|
|
|
|
|
|
+ * returns 0 if do_journal_end should return right away, returns 1 if
|
|
|
|
|
+ * do_journal_end should finish the commit
|
|
|
|
|
+ *
|
|
|
|
|
+ * if the current transaction is too old, but still has writers, this will
|
|
|
|
|
+ * wait on j_join_wait until all the writers are done. By the time it
|
|
|
|
|
+ * wakes up, the transaction it was called has already ended, so it just
|
|
|
|
|
+ * flushes the commit list and returns 0.
|
|
|
|
|
+ *
|
|
|
|
|
+ * Won't batch when flush or commit_now is set. Also won't batch when
|
|
|
|
|
+ * others are waiting on j_join_wait.
|
|
|
|
|
+ *
|
|
|
|
|
+ * Note, we can't allow the journal_end to proceed while there are still
|
|
|
|
|
+ * writers in the log.
|
|
|
|
|
+ */
|
|
|
|
|
+static int check_journal_end(struct reiserfs_transaction_handle *th, int flags)
|
|
|
{
|
|
{
|
|
|
|
|
|
|
|
time_t now;
|
|
time_t now;
|
|
@@ -3509,6 +3610,7 @@ static int check_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
int commit_now = flags & COMMIT_NOW;
|
|
int commit_now = flags & COMMIT_NOW;
|
|
|
int wait_on_commit = flags & WAIT;
|
|
int wait_on_commit = flags & WAIT;
|
|
|
struct reiserfs_journal_list *jl;
|
|
struct reiserfs_journal_list *jl;
|
|
|
|
|
+ struct super_block *sb = th->t_super;
|
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
|
|
|
|
|
|
BUG_ON(!th->t_trans_id);
|
|
BUG_ON(!th->t_trans_id);
|
|
@@ -3520,23 +3622,27 @@ static int check_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
|
|
journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
|
|
|
- if (atomic_read(&(journal->j_wcount)) > 0) { /* <= 0 is allowed. unmounting might not call begin */
|
|
|
|
|
- atomic_dec(&(journal->j_wcount));
|
|
|
|
|
- }
|
|
|
|
|
|
|
+ /* <= 0 is allowed. unmounting might not call begin */
|
|
|
|
|
+ if (atomic_read(&journal->j_wcount) > 0)
|
|
|
|
|
+ atomic_dec(&journal->j_wcount);
|
|
|
|
|
|
|
|
- /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
|
|
|
|
|
- ** will be dealt with by next transaction that actually writes something, but should be taken
|
|
|
|
|
- ** care of in this trans
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * BUG, deal with case where j_len is 0, but people previously
|
|
|
|
|
+ * freed blocks need to be released will be dealt with by next
|
|
|
|
|
+ * transaction that actually writes something, but should be taken
|
|
|
|
|
+ * care of in this trans
|
|
|
*/
|
|
*/
|
|
|
BUG_ON(journal->j_len == 0);
|
|
BUG_ON(journal->j_len == 0);
|
|
|
|
|
|
|
|
- /* if wcount > 0, and we are called to with flush or commit_now,
|
|
|
|
|
- ** we wait on j_join_wait. We will wake up when the last writer has
|
|
|
|
|
- ** finished the transaction, and started it on its way to the disk.
|
|
|
|
|
- ** Then, we flush the commit or journal list, and just return 0
|
|
|
|
|
- ** because the rest of journal end was already done for this transaction.
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * if wcount > 0, and we are called to with flush or commit_now,
|
|
|
|
|
+ * we wait on j_join_wait. We will wake up when the last writer has
|
|
|
|
|
+ * finished the transaction, and started it on its way to the disk.
|
|
|
|
|
+ * Then, we flush the commit or journal list, and just return 0
|
|
|
|
|
+ * because the rest of journal end was already done for this
|
|
|
|
|
+ * transaction.
|
|
|
*/
|
|
*/
|
|
|
- if (atomic_read(&(journal->j_wcount)) > 0) {
|
|
|
|
|
|
|
+ if (atomic_read(&journal->j_wcount) > 0) {
|
|
|
if (flush || commit_now) {
|
|
if (flush || commit_now) {
|
|
|
unsigned trans_id;
|
|
unsigned trans_id;
|
|
|
|
|
|
|
@@ -3544,27 +3650,30 @@ static int check_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
trans_id = jl->j_trans_id;
|
|
trans_id = jl->j_trans_id;
|
|
|
if (wait_on_commit)
|
|
if (wait_on_commit)
|
|
|
jl->j_state |= LIST_COMMIT_PENDING;
|
|
jl->j_state |= LIST_COMMIT_PENDING;
|
|
|
- atomic_set(&(journal->j_jlock), 1);
|
|
|
|
|
|
|
+ atomic_set(&journal->j_jlock, 1);
|
|
|
if (flush) {
|
|
if (flush) {
|
|
|
journal->j_next_full_flush = 1;
|
|
journal->j_next_full_flush = 1;
|
|
|
}
|
|
}
|
|
|
unlock_journal(sb);
|
|
unlock_journal(sb);
|
|
|
|
|
|
|
|
- /* sleep while the current transaction is still j_jlocked */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * sleep while the current transaction is
|
|
|
|
|
+ * still j_jlocked
|
|
|
|
|
+ */
|
|
|
while (journal->j_trans_id == trans_id) {
|
|
while (journal->j_trans_id == trans_id) {
|
|
|
if (atomic_read(&journal->j_jlock)) {
|
|
if (atomic_read(&journal->j_jlock)) {
|
|
|
queue_log_writer(sb);
|
|
queue_log_writer(sb);
|
|
|
} else {
|
|
} else {
|
|
|
lock_journal(sb);
|
|
lock_journal(sb);
|
|
|
if (journal->j_trans_id == trans_id) {
|
|
if (journal->j_trans_id == trans_id) {
|
|
|
- atomic_set(&(journal->j_jlock),
|
|
|
|
|
|
|
+ atomic_set(&journal->j_jlock,
|
|
|
1);
|
|
1);
|
|
|
}
|
|
}
|
|
|
unlock_journal(sb);
|
|
unlock_journal(sb);
|
|
|
}
|
|
}
|
|
|
}
|
|
}
|
|
|
BUG_ON(journal->j_trans_id == trans_id);
|
|
BUG_ON(journal->j_trans_id == trans_id);
|
|
|
-
|
|
|
|
|
|
|
+
|
|
|
if (commit_now
|
|
if (commit_now
|
|
|
&& journal_list_still_alive(sb, trans_id)
|
|
&& journal_list_still_alive(sb, trans_id)
|
|
|
&& wait_on_commit) {
|
|
&& wait_on_commit) {
|
|
@@ -3584,7 +3693,7 @@ static int check_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
}
|
|
}
|
|
|
/* don't batch when someone is waiting on j_join_wait */
|
|
/* don't batch when someone is waiting on j_join_wait */
|
|
|
/* don't batch when syncing the commit or flushing the whole trans */
|
|
/* don't batch when syncing the commit or flushing the whole trans */
|
|
|
- if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
|
|
|
|
|
|
|
+ if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock))
|
|
|
&& !flush && !commit_now && (journal->j_len < journal->j_max_batch)
|
|
&& !flush && !commit_now && (journal->j_len < journal->j_max_batch)
|
|
|
&& journal->j_len_alloc < journal->j_max_batch
|
|
&& journal->j_len_alloc < journal->j_max_batch
|
|
|
&& journal->j_cnode_free > (journal->j_trans_max * 3)) {
|
|
&& journal->j_cnode_free > (journal->j_trans_max * 3)) {
|
|
@@ -3602,19 +3711,22 @@ static int check_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** Does all the work that makes deleting blocks safe.
|
|
|
|
|
-** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
|
|
|
|
|
-**
|
|
|
|
|
-** otherwise:
|
|
|
|
|
-** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes
|
|
|
|
|
-** before this transaction has finished.
|
|
|
|
|
-**
|
|
|
|
|
-** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with
|
|
|
|
|
-** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash,
|
|
|
|
|
-** the block can't be reallocated yet.
|
|
|
|
|
-**
|
|
|
|
|
-** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * Does all the work that makes deleting blocks safe.
|
|
|
|
|
+ * when deleting a block mark BH_JNew, just remove it from the current
|
|
|
|
|
+ * transaction, clean it's buffer_head and move on.
|
|
|
|
|
+ *
|
|
|
|
|
+ * otherwise:
|
|
|
|
|
+ * set a bit for the block in the journal bitmap. That will prevent it from
|
|
|
|
|
+ * being allocated for unformatted nodes before this transaction has finished.
|
|
|
|
|
+ *
|
|
|
|
|
+ * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.
|
|
|
|
|
+ * That will prevent any old transactions with this block from trying to flush
|
|
|
|
|
+ * to the real location. Since we aren't removing the cnode from the
|
|
|
|
|
+ * journal_list_hash, *the block can't be reallocated yet.
|
|
|
|
|
+ *
|
|
|
|
|
+ * Then remove it from the current transaction, decrementing any counters and
|
|
|
|
|
+ * filing it on the clean list.
|
|
|
|
|
+ */
|
|
|
int journal_mark_freed(struct reiserfs_transaction_handle *th,
|
|
int journal_mark_freed(struct reiserfs_transaction_handle *th,
|
|
|
struct super_block *sb, b_blocknr_t blocknr)
|
|
struct super_block *sb, b_blocknr_t blocknr)
|
|
|
{
|
|
{
|
|
@@ -3637,7 +3749,10 @@ int journal_mark_freed(struct reiserfs_transaction_handle *th,
|
|
|
reiserfs_clean_and_file_buffer(bh);
|
|
reiserfs_clean_and_file_buffer(bh);
|
|
|
cleaned = remove_from_transaction(sb, blocknr, cleaned);
|
|
cleaned = remove_from_transaction(sb, blocknr, cleaned);
|
|
|
} else {
|
|
} else {
|
|
|
- /* set the bit for this block in the journal bitmap for this transaction */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * set the bit for this block in the journal bitmap
|
|
|
|
|
+ * for this transaction
|
|
|
|
|
+ */
|
|
|
jb = journal->j_current_jl->j_list_bitmap;
|
|
jb = journal->j_current_jl->j_list_bitmap;
|
|
|
if (!jb) {
|
|
if (!jb) {
|
|
|
reiserfs_panic(sb, "journal-1702",
|
|
reiserfs_panic(sb, "journal-1702",
|
|
@@ -3653,17 +3768,22 @@ int journal_mark_freed(struct reiserfs_transaction_handle *th,
|
|
|
}
|
|
}
|
|
|
cleaned = remove_from_transaction(sb, blocknr, cleaned);
|
|
cleaned = remove_from_transaction(sb, blocknr, cleaned);
|
|
|
|
|
|
|
|
- /* find all older transactions with this block, make sure they don't try to write it out */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * find all older transactions with this block,
|
|
|
|
|
+ * make sure they don't try to write it out
|
|
|
|
|
+ */
|
|
|
cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
|
|
cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
|
|
|
blocknr);
|
|
blocknr);
|
|
|
while (cn) {
|
|
while (cn) {
|
|
|
if (sb == cn->sb && blocknr == cn->blocknr) {
|
|
if (sb == cn->sb && blocknr == cn->blocknr) {
|
|
|
set_bit(BLOCK_FREED, &cn->state);
|
|
set_bit(BLOCK_FREED, &cn->state);
|
|
|
if (cn->bh) {
|
|
if (cn->bh) {
|
|
|
|
|
+ /*
|
|
|
|
|
+ * remove_from_transaction will brelse
|
|
|
|
|
+ * the buffer if it was in the current
|
|
|
|
|
+ * trans
|
|
|
|
|
+ */
|
|
|
if (!cleaned) {
|
|
if (!cleaned) {
|
|
|
- /* remove_from_transaction will brelse the buffer if it was
|
|
|
|
|
- ** in the current trans
|
|
|
|
|
- */
|
|
|
|
|
clear_buffer_journal_dirty(cn->
|
|
clear_buffer_journal_dirty(cn->
|
|
|
bh);
|
|
bh);
|
|
|
clear_buffer_dirty(cn->bh);
|
|
clear_buffer_dirty(cn->bh);
|
|
@@ -3672,16 +3792,19 @@ int journal_mark_freed(struct reiserfs_transaction_handle *th,
|
|
|
cleaned = 1;
|
|
cleaned = 1;
|
|
|
put_bh(cn->bh);
|
|
put_bh(cn->bh);
|
|
|
if (atomic_read
|
|
if (atomic_read
|
|
|
- (&(cn->bh->b_count)) < 0) {
|
|
|
|
|
|
|
+ (&cn->bh->b_count) < 0) {
|
|
|
reiserfs_warning(sb,
|
|
reiserfs_warning(sb,
|
|
|
"journal-2138",
|
|
"journal-2138",
|
|
|
"cn->bh->b_count < 0");
|
|
"cn->bh->b_count < 0");
|
|
|
}
|
|
}
|
|
|
}
|
|
}
|
|
|
- if (cn->jlist) { /* since we are clearing the bh, we MUST dec nonzerolen */
|
|
|
|
|
- atomic_dec(&
|
|
|
|
|
- (cn->jlist->
|
|
|
|
|
- j_nonzerolen));
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * since we are clearing the bh,
|
|
|
|
|
+ * we MUST dec nonzerolen
|
|
|
|
|
+ */
|
|
|
|
|
+ if (cn->jlist) {
|
|
|
|
|
+ atomic_dec(&cn->jlist->
|
|
|
|
|
+ j_nonzerolen);
|
|
|
}
|
|
}
|
|
|
cn->bh = NULL;
|
|
cn->bh = NULL;
|
|
|
}
|
|
}
|
|
@@ -3714,10 +3837,16 @@ static int __commit_trans_jl(struct inode *inode, unsigned long id,
|
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
|
int ret = 0;
|
|
int ret = 0;
|
|
|
|
|
|
|
|
- /* is it from the current transaction, or from an unknown transaction? */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * is it from the current transaction,
|
|
|
|
|
+ * or from an unknown transaction?
|
|
|
|
|
+ */
|
|
|
if (id == journal->j_trans_id) {
|
|
if (id == journal->j_trans_id) {
|
|
|
jl = journal->j_current_jl;
|
|
jl = journal->j_current_jl;
|
|
|
- /* try to let other writers come in and grow this transaction */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * try to let other writers come in and
|
|
|
|
|
+ * grow this transaction
|
|
|
|
|
+ */
|
|
|
let_transaction_grow(sb, id);
|
|
let_transaction_grow(sb, id);
|
|
|
if (journal->j_trans_id != id) {
|
|
if (journal->j_trans_id != id) {
|
|
|
goto flush_commit_only;
|
|
goto flush_commit_only;
|
|
@@ -3731,21 +3860,22 @@ static int __commit_trans_jl(struct inode *inode, unsigned long id,
|
|
|
if (journal->j_trans_id != id) {
|
|
if (journal->j_trans_id != id) {
|
|
|
reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
|
|
reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
|
|
|
1);
|
|
1);
|
|
|
- journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
|
|
|
|
|
- ret = journal_end(&th, sb, 1);
|
|
|
|
|
|
|
+ journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
|
|
|
|
|
+ ret = journal_end(&th);
|
|
|
goto flush_commit_only;
|
|
goto flush_commit_only;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- ret = journal_end_sync(&th, sb, 1);
|
|
|
|
|
|
|
+ ret = journal_end_sync(&th);
|
|
|
if (!ret)
|
|
if (!ret)
|
|
|
ret = 1;
|
|
ret = 1;
|
|
|
|
|
|
|
|
} else {
|
|
} else {
|
|
|
- /* this gets tricky, we have to make sure the journal list in
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * this gets tricky, we have to make sure the journal list in
|
|
|
* the inode still exists. We know the list is still around
|
|
* the inode still exists. We know the list is still around
|
|
|
* if we've got a larger transaction id than the oldest list
|
|
* if we've got a larger transaction id than the oldest list
|
|
|
*/
|
|
*/
|
|
|
- flush_commit_only:
|
|
|
|
|
|
|
+flush_commit_only:
|
|
|
if (journal_list_still_alive(inode->i_sb, id)) {
|
|
if (journal_list_still_alive(inode->i_sb, id)) {
|
|
|
/*
|
|
/*
|
|
|
* we only set ret to 1 when we know for sure
|
|
* we only set ret to 1 when we know for sure
|
|
@@ -3768,7 +3898,8 @@ int reiserfs_commit_for_inode(struct inode *inode)
|
|
|
unsigned int id = REISERFS_I(inode)->i_trans_id;
|
|
unsigned int id = REISERFS_I(inode)->i_trans_id;
|
|
|
struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
|
|
struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
|
|
|
|
|
|
|
|
- /* for the whole inode, assume unset id means it was
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * for the whole inode, assume unset id means it was
|
|
|
* changed in the current transaction. More conservative
|
|
* changed in the current transaction. More conservative
|
|
|
*/
|
|
*/
|
|
|
if (!id || !jl) {
|
|
if (!id || !jl) {
|
|
@@ -3806,12 +3937,11 @@ void reiserfs_restore_prepared_buffer(struct super_block *sb,
|
|
|
|
|
|
|
|
extern struct tree_balance *cur_tb;
|
|
extern struct tree_balance *cur_tb;
|
|
|
/*
|
|
/*
|
|
|
-** before we can change a metadata block, we have to make sure it won't
|
|
|
|
|
-** be written to disk while we are altering it. So, we must:
|
|
|
|
|
-** clean it
|
|
|
|
|
-** wait on it.
|
|
|
|
|
-**
|
|
|
|
|
-*/
|
|
|
|
|
|
|
+ * before we can change a metadata block, we have to make sure it won't
|
|
|
|
|
+ * be written to disk while we are altering it. So, we must:
|
|
|
|
|
+ * clean it
|
|
|
|
|
+ * wait on it.
|
|
|
|
|
+ */
|
|
|
int reiserfs_prepare_for_journal(struct super_block *sb,
|
|
int reiserfs_prepare_for_journal(struct super_block *sb,
|
|
|
struct buffer_head *bh, int wait)
|
|
struct buffer_head *bh, int wait)
|
|
|
{
|
|
{
|
|
@@ -3832,19 +3962,18 @@ int reiserfs_prepare_for_journal(struct super_block *sb,
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
-** long and ugly. If flush, will not return until all commit
|
|
|
|
|
-** blocks and all real buffers in the trans are on disk.
|
|
|
|
|
-** If no_async, won't return until all commit blocks are on disk.
|
|
|
|
|
-**
|
|
|
|
|
-** keep reading, there are comments as you go along
|
|
|
|
|
-**
|
|
|
|
|
-** If the journal is aborted, we just clean up. Things like flushing
|
|
|
|
|
-** journal lists, etc just won't happen.
|
|
|
|
|
-*/
|
|
|
|
|
-static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
|
|
- struct super_block *sb, unsigned long nblocks,
|
|
|
|
|
- int flags)
|
|
|
|
|
|
|
+ * long and ugly. If flush, will not return until all commit
|
|
|
|
|
+ * blocks and all real buffers in the trans are on disk.
|
|
|
|
|
+ * If no_async, won't return until all commit blocks are on disk.
|
|
|
|
|
+ *
|
|
|
|
|
+ * keep reading, there are comments as you go along
|
|
|
|
|
+ *
|
|
|
|
|
+ * If the journal is aborted, we just clean up. Things like flushing
|
|
|
|
|
+ * journal lists, etc just won't happen.
|
|
|
|
|
+ */
|
|
|
|
|
+static int do_journal_end(struct reiserfs_transaction_handle *th, int flags)
|
|
|
{
|
|
{
|
|
|
|
|
+ struct super_block *sb = th->t_super;
|
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
struct reiserfs_journal *journal = SB_JOURNAL(sb);
|
|
|
struct reiserfs_journal_cnode *cn, *next, *jl_cn;
|
|
struct reiserfs_journal_cnode *cn, *next, *jl_cn;
|
|
|
struct reiserfs_journal_cnode *last_cn = NULL;
|
|
struct reiserfs_journal_cnode *last_cn = NULL;
|
|
@@ -3866,9 +3995,12 @@ static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
|
|
|
|
|
BUG_ON(th->t_refcount > 1);
|
|
BUG_ON(th->t_refcount > 1);
|
|
|
BUG_ON(!th->t_trans_id);
|
|
BUG_ON(!th->t_trans_id);
|
|
|
|
|
+ BUG_ON(!th->t_super);
|
|
|
|
|
|
|
|
- /* protect flush_older_commits from doing mistakes if the
|
|
|
|
|
- transaction ID counter gets overflowed. */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * protect flush_older_commits from doing mistakes if the
|
|
|
|
|
+ * transaction ID counter gets overflowed.
|
|
|
|
|
+ */
|
|
|
if (th->t_trans_id == ~0U)
|
|
if (th->t_trans_id == ~0U)
|
|
|
flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
|
|
flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
|
|
|
flush = flags & FLUSH_ALL;
|
|
flush = flags & FLUSH_ALL;
|
|
@@ -3879,7 +4011,7 @@ static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
if (journal->j_len == 0) {
|
|
if (journal->j_len == 0) {
|
|
|
reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
|
|
reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
|
|
|
1);
|
|
1);
|
|
|
- journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
|
|
|
|
|
|
|
+ journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
lock_journal(sb);
|
|
lock_journal(sb);
|
|
@@ -3892,10 +4024,12 @@ static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
wait_on_commit = 1;
|
|
wait_on_commit = 1;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- /* check_journal_end locks the journal, and unlocks if it does not return 1
|
|
|
|
|
- ** it tells us if we should continue with the journal_end, or just return
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * check_journal_end locks the journal, and unlocks if it does
|
|
|
|
|
+ * not return 1 it tells us if we should continue with the
|
|
|
|
|
+ * journal_end, or just return
|
|
|
*/
|
|
*/
|
|
|
- if (!check_journal_end(th, sb, nblocks, flags)) {
|
|
|
|
|
|
|
+ if (!check_journal_end(th, flags)) {
|
|
|
reiserfs_schedule_old_flush(sb);
|
|
reiserfs_schedule_old_flush(sb);
|
|
|
wake_queued_writers(sb);
|
|
wake_queued_writers(sb);
|
|
|
reiserfs_async_progress_wait(sb);
|
|
reiserfs_async_progress_wait(sb);
|
|
@@ -3908,19 +4042,23 @@ static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
/*
|
|
|
- ** j must wait means we have to flush the log blocks, and the real blocks for
|
|
|
|
|
- ** this transaction
|
|
|
|
|
|
|
+ * j must wait means we have to flush the log blocks, and the
|
|
|
|
|
+ * real blocks for this transaction
|
|
|
*/
|
|
*/
|
|
|
if (journal->j_must_wait > 0) {
|
|
if (journal->j_must_wait > 0) {
|
|
|
flush = 1;
|
|
flush = 1;
|
|
|
}
|
|
}
|
|
|
#ifdef REISERFS_PREALLOCATE
|
|
#ifdef REISERFS_PREALLOCATE
|
|
|
- /* quota ops might need to nest, setup the journal_info pointer for them
|
|
|
|
|
- * and raise the refcount so that it is > 0. */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * quota ops might need to nest, setup the journal_info pointer
|
|
|
|
|
+ * for them and raise the refcount so that it is > 0.
|
|
|
|
|
+ */
|
|
|
current->journal_info = th;
|
|
current->journal_info = th;
|
|
|
th->t_refcount++;
|
|
th->t_refcount++;
|
|
|
- reiserfs_discard_all_prealloc(th); /* it should not involve new blocks into
|
|
|
|
|
- * the transaction */
|
|
|
|
|
|
|
+
|
|
|
|
|
+ /* it should not involve new blocks into the transaction */
|
|
|
|
|
+ reiserfs_discard_all_prealloc(th);
|
|
|
|
|
+
|
|
|
th->t_refcount--;
|
|
th->t_refcount--;
|
|
|
current->journal_info = th->t_handle_save;
|
|
current->journal_info = th->t_handle_save;
|
|
|
#endif
|
|
#endif
|
|
@@ -3936,7 +4074,10 @@ static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
|
|
memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
|
|
|
set_desc_trans_id(desc, journal->j_trans_id);
|
|
set_desc_trans_id(desc, journal->j_trans_id);
|
|
|
|
|
|
|
|
- /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * setup commit block. Don't write (keep it clean too) this one
|
|
|
|
|
+ * until after everyone else is written
|
|
|
|
|
+ */
|
|
|
c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
|
|
c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
|
|
|
((journal->j_start + journal->j_len +
|
|
((journal->j_start + journal->j_len +
|
|
|
1) % SB_ONDISK_JOURNAL_SIZE(sb)));
|
|
1) % SB_ONDISK_JOURNAL_SIZE(sb)));
|
|
@@ -3948,7 +4089,8 @@ static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
/* init this journal list */
|
|
/* init this journal list */
|
|
|
jl = journal->j_current_jl;
|
|
jl = journal->j_current_jl;
|
|
|
|
|
|
|
|
- /* we lock the commit before doing anything because
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * we lock the commit before doing anything because
|
|
|
* we want to make sure nobody tries to run flush_commit_list until
|
|
* we want to make sure nobody tries to run flush_commit_list until
|
|
|
* the new transaction is fully setup, and we've already flushed the
|
|
* the new transaction is fully setup, and we've already flushed the
|
|
|
* ordered bh list
|
|
* ordered bh list
|
|
@@ -3968,9 +4110,10 @@ static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
atomic_set(&jl->j_commit_left, journal->j_len + 2);
|
|
atomic_set(&jl->j_commit_left, journal->j_len + 2);
|
|
|
jl->j_realblock = NULL;
|
|
jl->j_realblock = NULL;
|
|
|
|
|
|
|
|
- /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
|
|
|
|
|
- ** for each real block, add it to the journal list hash,
|
|
|
|
|
- ** copy into real block index array in the commit or desc block
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * The ENTIRE FOR LOOP MUST not cause schedule to occur.
|
|
|
|
|
+ * for each real block, add it to the journal list hash,
|
|
|
|
|
+ * copy into real block index array in the commit or desc block
|
|
|
*/
|
|
*/
|
|
|
trans_half = journal_trans_half(sb->s_blocksize);
|
|
trans_half = journal_trans_half(sb->s_blocksize);
|
|
|
for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
|
|
for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
|
|
@@ -3989,9 +4132,10 @@ static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
last_cn->next = jl_cn;
|
|
last_cn->next = jl_cn;
|
|
|
}
|
|
}
|
|
|
last_cn = jl_cn;
|
|
last_cn = jl_cn;
|
|
|
- /* make sure the block we are trying to log is not a block
|
|
|
|
|
- of journal or reserved area */
|
|
|
|
|
-
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * make sure the block we are trying to log
|
|
|
|
|
+ * is not a block of journal or reserved area
|
|
|
|
|
+ */
|
|
|
if (is_block_in_log_or_reserved_area
|
|
if (is_block_in_log_or_reserved_area
|
|
|
(sb, cn->bh->b_blocknr)) {
|
|
(sb, cn->bh->b_blocknr)) {
|
|
|
reiserfs_panic(sb, "journal-2332",
|
|
reiserfs_panic(sb, "journal-2332",
|
|
@@ -4021,19 +4165,26 @@ static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
set_desc_trans_id(desc, journal->j_trans_id);
|
|
set_desc_trans_id(desc, journal->j_trans_id);
|
|
|
set_commit_trans_len(commit, journal->j_len);
|
|
set_commit_trans_len(commit, journal->j_len);
|
|
|
|
|
|
|
|
- /* special check in case all buffers in the journal were marked for not logging */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * special check in case all buffers in the journal
|
|
|
|
|
+ * were marked for not logging
|
|
|
|
|
+ */
|
|
|
BUG_ON(journal->j_len == 0);
|
|
BUG_ON(journal->j_len == 0);
|
|
|
|
|
|
|
|
- /* we're about to dirty all the log blocks, mark the description block
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * we're about to dirty all the log blocks, mark the description block
|
|
|
* dirty now too. Don't mark the commit block dirty until all the
|
|
* dirty now too. Don't mark the commit block dirty until all the
|
|
|
* others are on disk
|
|
* others are on disk
|
|
|
*/
|
|
*/
|
|
|
mark_buffer_dirty(d_bh);
|
|
mark_buffer_dirty(d_bh);
|
|
|
|
|
|
|
|
- /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * first data block is j_start + 1, so add one to
|
|
|
|
|
+ * cur_write_start wherever you use it
|
|
|
|
|
+ */
|
|
|
cur_write_start = journal->j_start;
|
|
cur_write_start = journal->j_start;
|
|
|
cn = journal->j_first;
|
|
cn = journal->j_first;
|
|
|
- jindex = 1; /* start at one so we don't get the desc again */
|
|
|
|
|
|
|
+ jindex = 1; /* start at one so we don't get the desc again */
|
|
|
while (cn) {
|
|
while (cn) {
|
|
|
clear_buffer_journal_new(cn->bh);
|
|
clear_buffer_journal_new(cn->bh);
|
|
|
/* copy all the real blocks into log area. dirty log blocks */
|
|
/* copy all the real blocks into log area. dirty log blocks */
|
|
@@ -4059,7 +4210,10 @@ static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
set_buffer_journal_dirty(cn->bh);
|
|
set_buffer_journal_dirty(cn->bh);
|
|
|
clear_buffer_journaled(cn->bh);
|
|
clear_buffer_journaled(cn->bh);
|
|
|
} else {
|
|
} else {
|
|
|
- /* JDirty cleared sometime during transaction. don't log this one */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * JDirty cleared sometime during transaction.
|
|
|
|
|
+ * don't log this one
|
|
|
|
|
+ */
|
|
|
reiserfs_warning(sb, "journal-2048",
|
|
reiserfs_warning(sb, "journal-2048",
|
|
|
"BAD, buffer in journal hash, "
|
|
"BAD, buffer in journal hash, "
|
|
|
"but not JDirty!");
|
|
"but not JDirty!");
|
|
@@ -4071,9 +4225,10 @@ static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
reiserfs_cond_resched(sb);
|
|
reiserfs_cond_resched(sb);
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- /* we are done with both the c_bh and d_bh, but
|
|
|
|
|
- ** c_bh must be written after all other commit blocks,
|
|
|
|
|
- ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * we are done with both the c_bh and d_bh, but
|
|
|
|
|
+ * c_bh must be written after all other commit blocks,
|
|
|
|
|
+ * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
|
|
|
*/
|
|
*/
|
|
|
|
|
|
|
|
journal->j_current_jl = alloc_journal_list(sb);
|
|
journal->j_current_jl = alloc_journal_list(sb);
|
|
@@ -4088,7 +4243,7 @@ static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
journal->j_start =
|
|
journal->j_start =
|
|
|
(journal->j_start + journal->j_len +
|
|
(journal->j_start + journal->j_len +
|
|
|
2) % SB_ONDISK_JOURNAL_SIZE(sb);
|
|
2) % SB_ONDISK_JOURNAL_SIZE(sb);
|
|
|
- atomic_set(&(journal->j_wcount), 0);
|
|
|
|
|
|
|
+ atomic_set(&journal->j_wcount, 0);
|
|
|
journal->j_bcount = 0;
|
|
journal->j_bcount = 0;
|
|
|
journal->j_last = NULL;
|
|
journal->j_last = NULL;
|
|
|
journal->j_first = NULL;
|
|
journal->j_first = NULL;
|
|
@@ -4104,15 +4259,18 @@ static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
journal->j_next_async_flush = 0;
|
|
journal->j_next_async_flush = 0;
|
|
|
init_journal_hash(sb);
|
|
init_journal_hash(sb);
|
|
|
|
|
|
|
|
- // make sure reiserfs_add_jh sees the new current_jl before we
|
|
|
|
|
- // write out the tails
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * make sure reiserfs_add_jh sees the new current_jl before we
|
|
|
|
|
+ * write out the tails
|
|
|
|
|
+ */
|
|
|
smp_mb();
|
|
smp_mb();
|
|
|
|
|
|
|
|
- /* tail conversion targets have to hit the disk before we end the
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * tail conversion targets have to hit the disk before we end the
|
|
|
* transaction. Otherwise a later transaction might repack the tail
|
|
* transaction. Otherwise a later transaction might repack the tail
|
|
|
- * before this transaction commits, leaving the data block unflushed and
|
|
|
|
|
- * clean, if we crash before the later transaction commits, the data block
|
|
|
|
|
- * is lost.
|
|
|
|
|
|
|
+ * before this transaction commits, leaving the data block unflushed
|
|
|
|
|
+ * and clean, if we crash before the later transaction commits, the
|
|
|
|
|
+ * data block is lost.
|
|
|
*/
|
|
*/
|
|
|
if (!list_empty(&jl->j_tail_bh_list)) {
|
|
if (!list_empty(&jl->j_tail_bh_list)) {
|
|
|
depth = reiserfs_write_unlock_nested(sb);
|
|
depth = reiserfs_write_unlock_nested(sb);
|
|
@@ -4123,24 +4281,27 @@ static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
BUG_ON(!list_empty(&jl->j_tail_bh_list));
|
|
BUG_ON(!list_empty(&jl->j_tail_bh_list));
|
|
|
mutex_unlock(&jl->j_commit_mutex);
|
|
mutex_unlock(&jl->j_commit_mutex);
|
|
|
|
|
|
|
|
- /* honor the flush wishes from the caller, simple commits can
|
|
|
|
|
- ** be done outside the journal lock, they are done below
|
|
|
|
|
- **
|
|
|
|
|
- ** if we don't flush the commit list right now, we put it into
|
|
|
|
|
- ** the work queue so the people waiting on the async progress work
|
|
|
|
|
- ** queue don't wait for this proc to flush journal lists and such.
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * honor the flush wishes from the caller, simple commits can
|
|
|
|
|
+ * be done outside the journal lock, they are done below
|
|
|
|
|
+ *
|
|
|
|
|
+ * if we don't flush the commit list right now, we put it into
|
|
|
|
|
+ * the work queue so the people waiting on the async progress work
|
|
|
|
|
+ * queue don't wait for this proc to flush journal lists and such.
|
|
|
*/
|
|
*/
|
|
|
if (flush) {
|
|
if (flush) {
|
|
|
flush_commit_list(sb, jl, 1);
|
|
flush_commit_list(sb, jl, 1);
|
|
|
flush_journal_list(sb, jl, 1);
|
|
flush_journal_list(sb, jl, 1);
|
|
|
} else if (!(jl->j_state & LIST_COMMIT_PENDING))
|
|
} else if (!(jl->j_state & LIST_COMMIT_PENDING))
|
|
|
- queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
|
|
|
|
|
|
|
+ queue_delayed_work(REISERFS_SB(sb)->commit_wq,
|
|
|
|
|
+ &journal->j_work, HZ / 10);
|
|
|
|
|
|
|
|
- /* if the next transaction has any chance of wrapping, flush
|
|
|
|
|
- ** transactions that might get overwritten. If any journal lists are very
|
|
|
|
|
- ** old flush them as well.
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * if the next transaction has any chance of wrapping, flush
|
|
|
|
|
+ * transactions that might get overwritten. If any journal lists
|
|
|
|
|
+ * are very old flush them as well.
|
|
|
*/
|
|
*/
|
|
|
- first_jl:
|
|
|
|
|
|
|
+first_jl:
|
|
|
list_for_each_safe(entry, safe, &journal->j_journal_list) {
|
|
list_for_each_safe(entry, safe, &journal->j_journal_list) {
|
|
|
temp_jl = JOURNAL_LIST_ENTRY(entry);
|
|
temp_jl = JOURNAL_LIST_ENTRY(entry);
|
|
|
if (journal->j_start <= temp_jl->j_start) {
|
|
if (journal->j_start <= temp_jl->j_start) {
|
|
@@ -4151,8 +4312,10 @@ static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
} else if ((journal->j_start +
|
|
} else if ((journal->j_start +
|
|
|
journal->j_trans_max + 1) <
|
|
journal->j_trans_max + 1) <
|
|
|
SB_ONDISK_JOURNAL_SIZE(sb)) {
|
|
SB_ONDISK_JOURNAL_SIZE(sb)) {
|
|
|
- /* if we don't cross into the next transaction and we don't
|
|
|
|
|
- * wrap, there is no way we can overlap any later transactions
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * if we don't cross into the next
|
|
|
|
|
+ * transaction and we don't wrap, there is
|
|
|
|
|
+ * no way we can overlap any later transactions
|
|
|
* break now
|
|
* break now
|
|
|
*/
|
|
*/
|
|
|
break;
|
|
break;
|
|
@@ -4166,10 +4329,12 @@ static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
flush_used_journal_lists(sb, temp_jl);
|
|
flush_used_journal_lists(sb, temp_jl);
|
|
|
goto first_jl;
|
|
goto first_jl;
|
|
|
} else {
|
|
} else {
|
|
|
- /* we don't overlap anything from out start to the end of the
|
|
|
|
|
- * log, and our wrapped portion doesn't overlap anything at
|
|
|
|
|
- * the start of the log. We can break
|
|
|
|
|
- */
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * we don't overlap anything from out start
|
|
|
|
|
+ * to the end of the log, and our wrapped
|
|
|
|
|
+ * portion doesn't overlap anything at
|
|
|
|
|
+ * the start of the log. We can break
|
|
|
|
|
+ */
|
|
|
break;
|
|
break;
|
|
|
}
|
|
}
|
|
|
}
|
|
}
|
|
@@ -4183,23 +4348,25 @@ static int do_journal_end(struct reiserfs_transaction_handle *th,
|
|
|
"could not get a list bitmap");
|
|
"could not get a list bitmap");
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
- atomic_set(&(journal->j_jlock), 0);
|
|
|
|
|
|
|
+ atomic_set(&journal->j_jlock, 0);
|
|
|
unlock_journal(sb);
|
|
unlock_journal(sb);
|
|
|
/* wake up any body waiting to join. */
|
|
/* wake up any body waiting to join. */
|
|
|
clear_bit(J_WRITERS_QUEUED, &journal->j_state);
|
|
clear_bit(J_WRITERS_QUEUED, &journal->j_state);
|
|
|
- wake_up(&(journal->j_join_wait));
|
|
|
|
|
|
|
+ wake_up(&journal->j_join_wait);
|
|
|
|
|
|
|
|
if (!flush && wait_on_commit &&
|
|
if (!flush && wait_on_commit &&
|
|
|
journal_list_still_alive(sb, commit_trans_id)) {
|
|
journal_list_still_alive(sb, commit_trans_id)) {
|
|
|
flush_commit_list(sb, jl, 1);
|
|
flush_commit_list(sb, jl, 1);
|
|
|
}
|
|
}
|
|
|
- out:
|
|
|
|
|
|
|
+out:
|
|
|
reiserfs_check_lock_depth(sb, "journal end2");
|
|
reiserfs_check_lock_depth(sb, "journal end2");
|
|
|
|
|
|
|
|
memset(th, 0, sizeof(*th));
|
|
memset(th, 0, sizeof(*th));
|
|
|
- /* Re-set th->t_super, so we can properly keep track of how many
|
|
|
|
|
|
|
+ /*
|
|
|
|
|
+ * Re-set th->t_super, so we can properly keep track of how many
|
|
|
* persistent transactions there are. We need to do this so if this
|
|
* persistent transactions there are. We need to do this so if this
|
|
|
- * call is part of a failed restart_transaction, we can free it later */
|
|
|
|
|
|
|
+ * call is part of a failed restart_transaction, we can free it later
|
|
|
|
|
+ */
|
|
|
th->t_super = sb;
|
|
th->t_super = sb;
|
|
|
|
|
|
|
|
return journal->j_errno;
|
|
return journal->j_errno;
|