journal.c

linux 内核源代码

/*
** Write ahead logging implementation copyright Chris Mason 2000
**
** The background commits make this code very interelated, 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 <asm/uaccess.h>
#include <asm/system.h>

#include <linux/time.h>
#include <asm/semaphore.h>

#include <linux/vmalloc.h>
#include <linux/reiserfs_fs.h>

#include <linux/kernel.h>
#include <linux/errno.h>
#include <linux/fcntl.h>
#include <linux/stat.h>
#include <linux/string.h>
#include <linux/smp_lock.h>
#include <linux/buffer_head.h>
#include <linux/workqueue.h>
#include <linux/writeback.h>
#include <linux/blkdev.h>
#include <linux/backing-dev.h>

/* gets a struct reiserfs_journal_list * from a list head */
#define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
                               j_list))
#define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_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 */
#define BUFNR 64		/*read ahead */

/* 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 */

#define BLOCK_NEEDS_FLUSH 4	/* used in flush_journal_list */
#define BLOCK_DIRTIED 5

/* journal list state bits */
#define LIST_TOUCHED 1
#define LIST_DIRTY   2
#define LIST_COMMIT_PENDING  4	/* someone will commit this list */

/* flags for do_journal_end */
#define FLUSH_ALL   1		/* flush commit and real blocks */
#define COMMIT_NOW  2		/* end and commit this transaction */
#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 flush_journal_list(struct super_block *s,
			      struct reiserfs_journal_list *jl, int flushall);
static int flush_commit_list(struct super_block *s,
			     struct reiserfs_journal_list *jl, int flushall);
static int can_dirty(struct reiserfs_journal_cnode *cn);
static int journal_join(struct reiserfs_transaction_handle *th,
			struct super_block *p_s_sb, unsigned long nblocks);
static int release_journal_dev(struct super_block *super,
			       struct reiserfs_journal *journal);
static int dirty_one_transaction(struct super_block *s,
				 struct reiserfs_journal_list *jl);
static void flush_async_commits(struct work_struct *work);
static void queue_log_writer(struct super_block *s);

/* values for join in do_journal_begin_r */
enum {
	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 */
};

static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
			      struct super_block *p_s_sb,
			      unsigned long nblocks, int join);

static void init_journal_hash(struct super_block *p_s_sb)
{
	struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
	memset(journal->j_hash_table, 0,
	       JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
}

/*
** 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)
{
	if (bh) {
		clear_buffer_dirty(bh);
		clear_buffer_journal_test(bh);
	}
	return 0;
}

static void disable_barrier(struct super_block *s)
{
	REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH);
	printk("reiserfs: disabling flush barriers on %s\n",
	       reiserfs_bdevname(s));
}

static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
							 *p_s_sb)
{
	struct reiserfs_bitmap_node *bn;
	static int id;

	bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
	if (!bn) {
		return NULL;
	}
	bn->data = kzalloc(p_s_sb->s_blocksize, GFP_NOFS);
	if (!bn->data) {
		kfree(bn);
		return NULL;
	}
	bn->id = id++;
	INIT_LIST_HEAD(&bn->list);
	return bn;
}

static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *p_s_sb)
{
	struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
	struct reiserfs_bitmap_node *bn = NULL;
	struct list_head *entry = journal->j_bitmap_nodes.next;

	journal->j_used_bitmap_nodes++;
      repeat:

	if (entry != &journal->j_bitmap_nodes) {
		bn = list_entry(entry, struct reiserfs_bitmap_node, list);
		list_del(entry);
		memset(bn->data, 0, p_s_sb->s_blocksize);
		journal->j_free_bitmap_nodes--;
		return bn;
	}
	bn = allocate_bitmap_node(p_s_sb);
	if (!bn) {
		yield();
		goto repeat;
	}
	return bn;
}
static inline void free_bitmap_node(struct super_block *p_s_sb,
				    struct reiserfs_bitmap_node *bn)
{
	struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
	journal->j_used_bitmap_nodes--;
	if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
		kfree(bn->data);
		kfree(bn);
	} else {
		list_add(&bn->list, &journal->j_bitmap_nodes);
		journal->j_free_bitmap_nodes++;
	}
}

static void allocate_bitmap_nodes(struct super_block *p_s_sb)
{
	int i;
	struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
	struct reiserfs_bitmap_node *bn = NULL;
	for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
		bn = allocate_bitmap_node(p_s_sb);
		if (bn) {
			list_add(&bn->list, &journal->j_bitmap_nodes);
			journal->j_free_bitmap_nodes++;
		} else {
			break;	// this is ok, we'll try again when more are needed 
		}
	}
}

static int set_bit_in_list_bitmap(struct super_block *p_s_sb,
				  b_blocknr_t block,
				  struct reiserfs_list_bitmap *jb)
{
	unsigned int bmap_nr = block / (p_s_sb->s_blocksize << 3);
	unsigned int bit_nr = block % (p_s_sb->s_blocksize << 3);

	if (!jb->bitmaps[bmap_nr]) {
		jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb);
	}
	set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
	return 0;
}

static void cleanup_bitmap_list(struct super_block *p_s_sb,
				struct reiserfs_list_bitmap *jb)
{
	int i;
	if (jb->bitmaps == NULL)
		return;

	for (i = 0; i < reiserfs_bmap_count(p_s_sb); i++) {
		if (jb->bitmaps[i]) {
			free_bitmap_node(p_s_sb, jb->bitmaps[i]);
			jb->bitmaps[i] = NULL;
		}
	}
}

/*
** only call this on FS unmount.
*/
static int free_list_bitmaps(struct super_block *p_s_sb,
			     struct reiserfs_list_bitmap *jb_array)
{
	int i;
	struct reiserfs_list_bitmap *jb;
	for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
		jb = jb_array + i;
		jb->journal_list = NULL;
		cleanup_bitmap_list(p_s_sb, jb);
		vfree(jb->bitmaps);
		jb->bitmaps = NULL;
	}
	return 0;
}

static int free_bitmap_nodes(struct super_block *p_s_sb)
{
	struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
	struct list_head *next = journal->j_bitmap_nodes.next;
	struct reiserfs_bitmap_node *bn;

	while (next != &journal->j_bitmap_nodes) {
		bn = list_entry(next, struct reiserfs_bitmap_node, list);
		list_del(next);
		kfree(bn->data);
		kfree(bn);
		next = journal->j_bitmap_nodes.next;
		journal->j_free_bitmap_nodes--;
	}

	return 0;
}

/*
** 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 *p_s_sb,
				   struct reiserfs_list_bitmap *jb_array,
				   unsigned int bmap_nr)
{
	int i;
	int failed = 0;
	struct reiserfs_list_bitmap *jb;
	int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);

	for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
		jb = jb_array + i;
		jb->journal_list = NULL;
		jb->bitmaps = vmalloc(mem);
		if (!jb->bitmaps) {
			reiserfs_warning(p_s_sb,
					 "clm-2000, unable to allocate bitmaps for journal lists");
			failed = 1;
			break;
		}
		memset(jb->bitmaps, 0, mem);
	}
	if (failed) {
		free_list_bitmaps(p_s_sb, jb_array);
		return -1;
	}
	return 0;
}

/*
** 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 *p_s_sb,
						    struct reiserfs_journal_list
						    *jl)
{
	int i, j;
	struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
	struct reiserfs_list_bitmap *jb = NULL;

	for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
		i = journal->j_list_bitmap_index;
		journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
		jb = journal->j_list_bitmap + i;
		if (journal->j_list_bitmap[i].journal_list) {
			flush_commit_list(p_s_sb,
					  journal->j_list_bitmap[i].
					  journal_list, 1);
			if (!journal->j_list_bitmap[i].journal_list) {
				break;
			}
		} else {
			break;
		}
	}
	if (jb->journal_list) {	/* double check to make sure if flushed correctly */
		return NULL;
	}
	jb->journal_list = jl;
	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
*/
static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
{
	struct reiserfs_journal_cnode *head;
	int i;
	if (num_cnodes <= 0) {
		return NULL;
	}
	head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
	if (!head) {
		return NULL;
	}
	memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode));
	head[0].prev = NULL;
	head[0].next = head + 1;
	for (i = 1; i < num_cnodes; i++) {
		head[i].prev = head + (i - 1);
		head[i].next = head + (i + 1);	/* if last one, overwrite it after the if */
	}
	head[num_cnodes - 1].next = NULL;
	return head;
}

/*
** pulls a cnode off the free list, or returns NULL on failure 
*/
static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb)
{
	struct reiserfs_journal_cnode *cn;
	struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);

	reiserfs_check_lock_depth(p_s_sb, "get_cnode");

	if (journal->j_cnode_free <= 0) {
		return NULL;
	}
	journal->j_cnode_used++;
	journal->j_cnode_free--;
	cn = journal->j_cnode_free_list;
	if (!cn) {
		return cn;
	}
	if (cn->next) {
		cn->next->prev = NULL;
	}
	journal->j_cnode_free_list = cn->next;
	memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
	return cn;
}

/*
** returns a cnode to the free list 
*/
static void free_cnode(struct super_block *p_s_sb,
		       struct reiserfs_journal_cnode *cn)
{
	struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);

	reiserfs_check_lock_depth(p_s_sb, "free_cnode");

	journal->j_cnode_used--;
	journal->j_cnode_free++;
	/* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
	cn->next = journal->j_cnode_free_list;
	if (journal->j_cnode_free_list) {
		journal->j_cnode_free_list->prev = cn;
	}
	cn->prev = NULL;	/* not needed with the memset, but I might kill the memset, and forget to do this */
	journal->j_cnode_free_list = cn;
}

static void clear_prepared_bits(struct buffer_head *bh)
{
	clear_buffer_journal_prepared(bh);
	clear_buffer_journal_restore_dirty(bh);
}

/* utility function to force a BUG if it is called without the big
** kernel lock held.  caller is the string printed just before calling BUG()
*/
void reiserfs_check_lock_depth(struct super_block *sb, char *caller)
{
#ifdef CONFIG_SMP
	if (current->lock_depth < 0) {
		reiserfs_panic(sb, "%s called without kernel lock held",
			       caller);
	}
#else
	;
#endif
}

/* 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
								  super_block
								  *sb,
								  struct
								  reiserfs_journal_cnode
								  **table,
								  long bl)
{
	struct reiserfs_journal_cnode *cn;
	cn = journal_hash(table, sb, bl);
	while (cn) {
		if (cn->blocknr == bl && cn->sb == sb)
			return cn;
		cn = cn->hnext;
	}