aboutsummaryrefslogtreecommitdiff
path: root/fs/ocfs2
diff options
context:
space:
mode:
Diffstat (limited to 'fs/ocfs2')
-rw-r--r--fs/ocfs2/Makefile1
-rw-r--r--fs/ocfs2/alloc.c28
-rw-r--r--fs/ocfs2/aops.c83
-rw-r--r--fs/ocfs2/buffer_head_io.c95
-rw-r--r--fs/ocfs2/buffer_head_io.h2
-rw-r--r--fs/ocfs2/cluster/heartbeat.c8
-rw-r--r--fs/ocfs2/cluster/tcp_internal.h8
-rw-r--r--fs/ocfs2/dcache.c359
-rw-r--r--fs/ocfs2/dcache.h27
-rw-r--r--fs/ocfs2/dir.c28
-rw-r--r--fs/ocfs2/dlm/dlmapi.h1
-rw-r--r--fs/ocfs2/dlm/dlmast.c16
-rw-r--r--fs/ocfs2/dlm/dlmcommon.h1
-rw-r--r--fs/ocfs2/dlm/dlmlock.c10
-rw-r--r--fs/ocfs2/dlm/dlmmaster.c4
-rw-r--r--fs/ocfs2/dlm/dlmrecovery.c3
-rw-r--r--fs/ocfs2/dlm/userdlm.c81
-rw-r--r--fs/ocfs2/dlm/userdlm.h1
-rw-r--r--fs/ocfs2/dlmglue.c1103
-rw-r--r--fs/ocfs2/dlmglue.h22
-rw-r--r--fs/ocfs2/export.c8
-rw-r--r--fs/ocfs2/file.c3
-rw-r--r--fs/ocfs2/inode.c188
-rw-r--r--fs/ocfs2/inode.h11
-rw-r--r--fs/ocfs2/ioctl.c136
-rw-r--r--fs/ocfs2/ioctl.h16
-rw-r--r--fs/ocfs2/journal.c3
-rw-r--r--fs/ocfs2/namei.c148
-rw-r--r--fs/ocfs2/ocfs2_fs.h24
-rw-r--r--fs/ocfs2/ocfs2_lockid.h25
-rw-r--r--fs/ocfs2/super.c6
-rw-r--r--fs/ocfs2/sysfile.c6
-rw-r--r--fs/ocfs2/uptodate.c21
-rw-r--r--fs/ocfs2/uptodate.h2
-rw-r--r--fs/ocfs2/vote.c180
-rw-r--r--fs/ocfs2/vote.h5
36 files changed, 1638 insertions, 1025 deletions
diff --git a/fs/ocfs2/Makefile b/fs/ocfs2/Makefile
index 7d3be845a61..9fb8132f19b 100644
--- a/fs/ocfs2/Makefile
+++ b/fs/ocfs2/Makefile
@@ -16,6 +16,7 @@ ocfs2-objs := \
file.o \
heartbeat.o \
inode.o \
+ ioctl.o \
journal.o \
localalloc.o \
mmap.o \
diff --git a/fs/ocfs2/alloc.c b/fs/ocfs2/alloc.c
index edaab05a93e..f43bc5f18a3 100644
--- a/fs/ocfs2/alloc.c
+++ b/fs/ocfs2/alloc.c
@@ -1717,17 +1717,29 @@ static int ocfs2_do_truncate(struct ocfs2_super *osb,
ocfs2_remove_from_cache(inode, eb_bh);
- BUG_ON(eb->h_suballoc_slot);
BUG_ON(el->l_recs[0].e_clusters);
BUG_ON(el->l_recs[0].e_cpos);
BUG_ON(el->l_recs[0].e_blkno);
- status = ocfs2_free_extent_block(handle,
- tc->tc_ext_alloc_inode,
- tc->tc_ext_alloc_bh,
- eb);
- if (status < 0) {
- mlog_errno(status);
- goto bail;
+ if (eb->h_suballoc_slot == 0) {
+ /*
+ * This code only understands how to
+ * lock the suballocator in slot 0,
+ * which is fine because allocation is
+ * only ever done out of that
+ * suballocator too. A future version
+ * might change that however, so avoid
+ * a free if we don't know how to
+ * handle it. This way an fs incompat
+ * bit will not be necessary.
+ */
+ status = ocfs2_free_extent_block(handle,
+ tc->tc_ext_alloc_inode,
+ tc->tc_ext_alloc_bh,
+ eb);
+ if (status < 0) {
+ mlog_errno(status);
+ goto bail;
+ }
}
}
brelse(eb_bh);
diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
index f1d1c342ce0..3d7c082a8f5 100644
--- a/fs/ocfs2/aops.c
+++ b/fs/ocfs2/aops.c
@@ -391,31 +391,28 @@ out:
static int ocfs2_commit_write(struct file *file, struct page *page,
unsigned from, unsigned to)
{
- int ret, extending = 0, locklevel = 0;
- loff_t new_i_size;
+ int ret;
struct buffer_head *di_bh = NULL;
struct inode *inode = page->mapping->host;
struct ocfs2_journal_handle *handle = NULL;
+ struct ocfs2_dinode *di;
mlog_entry("(0x%p, 0x%p, %u, %u)\n", file, page, from, to);
/* NOTE: ocfs2_file_aio_write has ensured that it's safe for
- * us to sample inode->i_size here without the metadata lock:
+ * us to continue here without rechecking the I/O against
+ * changed inode values.
*
* 1) We're currently holding the inode alloc lock, so no
* nodes can change it underneath us.
*
* 2) We've had to take the metadata lock at least once
- * already to check for extending writes, hence insuring
- * that our current copy is also up to date.
+ * already to check for extending writes, suid removal, etc.
+ * The meta data update code then ensures that we don't get a
+ * stale inode allocation image (i_size, i_clusters, etc).
*/
- new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
- if (new_i_size > i_size_read(inode)) {
- extending = 1;
- locklevel = 1;
- }
- ret = ocfs2_meta_lock_with_page(inode, NULL, &di_bh, locklevel, page);
+ ret = ocfs2_meta_lock_with_page(inode, NULL, &di_bh, 1, page);
if (ret != 0) {
mlog_errno(ret);
goto out;
@@ -427,23 +424,20 @@ static int ocfs2_commit_write(struct file *file, struct page *page,
goto out_unlock_meta;
}
- if (extending) {
- handle = ocfs2_start_walk_page_trans(inode, page, from, to);
- if (IS_ERR(handle)) {
- ret = PTR_ERR(handle);
- handle = NULL;
- goto out_unlock_data;
- }
+ handle = ocfs2_start_walk_page_trans(inode, page, from, to);
+ if (IS_ERR(handle)) {
+ ret = PTR_ERR(handle);
+ goto out_unlock_data;
+ }
- /* Mark our buffer early. We'd rather catch this error up here
- * as opposed to after a successful commit_write which would
- * require us to set back inode->i_size. */
- ret = ocfs2_journal_access(handle, inode, di_bh,
- OCFS2_JOURNAL_ACCESS_WRITE);
- if (ret < 0) {
- mlog_errno(ret);
- goto out_commit;
- }
+ /* Mark our buffer early. We'd rather catch this error up here
+ * as opposed to after a successful commit_write which would
+ * require us to set back inode->i_size. */
+ ret = ocfs2_journal_access(handle, inode, di_bh,
+ OCFS2_JOURNAL_ACCESS_WRITE);
+ if (ret < 0) {
+ mlog_errno(ret);
+ goto out_commit;
}
/* might update i_size */
@@ -453,37 +447,28 @@ static int ocfs2_commit_write(struct file *file, struct page *page,
goto out_commit;
}
- if (extending) {
- loff_t size = (u64) i_size_read(inode);
- struct ocfs2_dinode *di =
- (struct ocfs2_dinode *)di_bh->b_data;
+ di = (struct ocfs2_dinode *)di_bh->b_data;
- /* ocfs2_mark_inode_dirty is too heavy to use here. */
- inode->i_blocks = ocfs2_align_bytes_to_sectors(size);
- inode->i_ctime = inode->i_mtime = CURRENT_TIME;
+ /* ocfs2_mark_inode_dirty() is too heavy to use here. */
+ inode->i_mtime = inode->i_ctime = CURRENT_TIME;
+ di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
+ di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
- di->i_size = cpu_to_le64(size);
- di->i_ctime = di->i_mtime =
- cpu_to_le64(inode->i_mtime.tv_sec);
- di->i_ctime_nsec = di->i_mtime_nsec =
- cpu_to_le32(inode->i_mtime.tv_nsec);
+ inode->i_blocks = ocfs2_align_bytes_to_sectors((u64)(i_size_read(inode)));
+ di->i_size = cpu_to_le64((u64)i_size_read(inode));
- ret = ocfs2_journal_dirty(handle, di_bh);
- if (ret < 0) {
- mlog_errno(ret);
- goto out_commit;
- }
+ ret = ocfs2_journal_dirty(handle, di_bh);
+ if (ret < 0) {
+ mlog_errno(ret);
+ goto out_commit;
}
- BUG_ON(extending && (i_size_read(inode) != new_i_size));
-
out_commit:
- if (handle)
- ocfs2_commit_trans(handle);
+ ocfs2_commit_trans(handle);
out_unlock_data:
ocfs2_data_unlock(inode, 1);
out_unlock_meta:
- ocfs2_meta_unlock(inode, locklevel);
+ ocfs2_meta_unlock(inode, 1);
out:
if (di_bh)
brelse(di_bh);
diff --git a/fs/ocfs2/buffer_head_io.c b/fs/ocfs2/buffer_head_io.c
index 9a24adf9be6..c9037414f4f 100644
--- a/fs/ocfs2/buffer_head_io.c
+++ b/fs/ocfs2/buffer_head_io.c
@@ -100,6 +100,9 @@ int ocfs2_read_blocks(struct ocfs2_super *osb, u64 block, int nr,
mlog_entry("(block=(%llu), nr=(%d), flags=%d, inode=%p)\n",
(unsigned long long)block, nr, flags, inode);
+ BUG_ON((flags & OCFS2_BH_READAHEAD) &&
+ (!inode || !(flags & OCFS2_BH_CACHED)));
+
if (osb == NULL || osb->sb == NULL || bhs == NULL) {
status = -EINVAL;
mlog_errno(status);
@@ -140,6 +143,30 @@ int ocfs2_read_blocks(struct ocfs2_super *osb, u64 block, int nr,
bh = bhs[i];
ignore_cache = 0;
+ /* There are three read-ahead cases here which we need to
+ * be concerned with. All three assume a buffer has
+ * previously been submitted with OCFS2_BH_READAHEAD
+ * and it hasn't yet completed I/O.
+ *
+ * 1) The current request is sync to disk. This rarely
+ * happens these days, and never when performance
+ * matters - the code can just wait on the buffer
+ * lock and re-submit.
+ *
+ * 2) The current request is cached, but not
+ * readahead. ocfs2_buffer_uptodate() will return
+ * false anyway, so we'll wind up waiting on the
+ * buffer lock to do I/O. We re-check the request
+ * with after getting the lock to avoid a re-submit.
+ *
+ * 3) The current request is readahead (and so must
+ * also be a caching one). We short circuit if the
+ * buffer is locked (under I/O) and if it's in the
+ * uptodate cache. The re-check from #2 catches the
+ * case that the previous read-ahead completes just
+ * before our is-it-in-flight check.
+ */
+
if (flags & OCFS2_BH_CACHED &&
!ocfs2_buffer_uptodate(inode, bh)) {
mlog(ML_UPTODATE,
@@ -169,6 +196,14 @@ int ocfs2_read_blocks(struct ocfs2_super *osb, u64 block, int nr,
continue;
}
+ /* A read-ahead request was made - if the
+ * buffer is already under read-ahead from a
+ * previously submitted request than we are
+ * done here. */
+ if ((flags & OCFS2_BH_READAHEAD)
+ && ocfs2_buffer_read_ahead(inode, bh))
+ continue;
+
lock_buffer(bh);
if (buffer_jbd(bh)) {
#ifdef CATCH_BH_JBD_RACES
@@ -181,13 +216,22 @@ int ocfs2_read_blocks(struct ocfs2_super *osb, u64 block, int nr,
continue;
#endif
}
+
+ /* Re-check ocfs2_buffer_uptodate() as a
+ * previously read-ahead buffer may have
+ * completed I/O while we were waiting for the
+ * buffer lock. */
+ if ((flags & OCFS2_BH_CACHED)
+ && !(flags & OCFS2_BH_READAHEAD)
+ && ocfs2_buffer_uptodate(inode, bh)) {
+ unlock_buffer(bh);
+ continue;
+ }
+
clear_buffer_uptodate(bh);
get_bh(bh); /* for end_buffer_read_sync() */
bh->b_end_io = end_buffer_read_sync;
- if (flags & OCFS2_BH_READAHEAD)
- submit_bh(READA, bh);
- else
- submit_bh(READ, bh);
+ submit_bh(READ, bh);
continue;
}
}
@@ -197,34 +241,39 @@ int ocfs2_read_blocks(struct ocfs2_super *osb, u64 block, int nr,
for (i = (nr - 1); i >= 0; i--) {
bh = bhs[i];
- /* We know this can't have changed as we hold the
- * inode sem. Avoid doing any work on the bh if the
- * journal has it. */
- if (!buffer_jbd(bh))
- wait_on_buffer(bh);
-
- if (!buffer_uptodate(bh)) {
- /* Status won't be cleared from here on out,
- * so we can safely record this and loop back
- * to cleanup the other buffers. Don't need to
- * remove the clustered uptodate information
- * for this bh as it's not marked locally
- * uptodate. */
- status = -EIO;
- brelse(bh);
- bhs[i] = NULL;
- continue;
+ if (!(flags & OCFS2_BH_READAHEAD)) {
+ /* We know this can't have changed as we hold the
+ * inode sem. Avoid doing any work on the bh if the
+ * journal has it. */
+ if (!buffer_jbd(bh))
+ wait_on_buffer(bh);
+
+ if (!buffer_uptodate(bh)) {
+ /* Status won't be cleared from here on out,
+ * so we can safely record this and loop back
+ * to cleanup the other buffers. Don't need to
+ * remove the clustered uptodate information
+ * for this bh as it's not marked locally
+ * uptodate. */
+ status = -EIO;
+ brelse(bh);
+ bhs[i] = NULL;
+ continue;
+ }
}
+ /* Always set the buffer in the cache, even if it was
+ * a forced read, or read-ahead which hasn't yet
+ * completed. */
if (inode)
ocfs2_set_buffer_uptodate(inode, bh);
}
if (inode)
mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
- mlog(ML_BH_IO, "block=(%llu), nr=(%d), cached=%s\n",
+ mlog(ML_BH_IO, "block=(%llu), nr=(%d), cached=%s, flags=0x%x\n",
(unsigned long long)block, nr,
- (!(flags & OCFS2_BH_CACHED) || ignore_cache) ? "no" : "yes");
+ (!(flags & OCFS2_BH_CACHED) || ignore_cache) ? "no" : "yes", flags);
bail:
diff --git a/fs/ocfs2/buffer_head_io.h b/fs/ocfs2/buffer_head_io.h
index 6ecb90937b6..6cc20930fac 100644
--- a/fs/ocfs2/buffer_head_io.h
+++ b/fs/ocfs2/buffer_head_io.h
@@ -49,7 +49,7 @@ int ocfs2_read_blocks(struct ocfs2_super *osb,
#define OCFS2_BH_CACHED 1
-#define OCFS2_BH_READAHEAD 8 /* use this to pass READA down to submit_bh */
+#define OCFS2_BH_READAHEAD 8
static inline int ocfs2_read_block(struct ocfs2_super * osb, u64 off,
struct buffer_head **bh, int flags,
diff --git a/fs/ocfs2/cluster/heartbeat.c b/fs/ocfs2/cluster/heartbeat.c
index 504595d6cf6..305cba3681f 100644
--- a/fs/ocfs2/cluster/heartbeat.c
+++ b/fs/ocfs2/cluster/heartbeat.c
@@ -320,8 +320,12 @@ static int compute_max_sectors(struct block_device *bdev)
max_pages = q->max_hw_segments;
max_pages--; /* Handle I/Os that straddle a page */
- max_sectors = max_pages << (PAGE_SHIFT - 9);
-
+ if (max_pages) {
+ max_sectors = max_pages << (PAGE_SHIFT - 9);
+ } else {
+ /* If BIO contains 1 or less than 1 page. */
+ max_sectors = q->max_sectors;
+ }
/* Why is fls() 1-based???? */
pow_two_sectors = 1 << (fls(max_sectors) - 1);
diff --git a/fs/ocfs2/cluster/tcp_internal.h b/fs/ocfs2/cluster/tcp_internal.h
index ff9e2e2104c..4b46aac7d24 100644
--- a/fs/ocfs2/cluster/tcp_internal.h
+++ b/fs/ocfs2/cluster/tcp_internal.h
@@ -44,11 +44,17 @@
* locking semantics of the file system using the protocol. It should
* be somewhere else, I'm sure, but right now it isn't.
*
+ * New in version 4:
+ * - Remove i_generation from lock names for better stat performance.
+ *
+ * New in version 3:
+ * - Replace dentry votes with a cluster lock
+ *
* New in version 2:
* - full 64 bit i_size in the metadata lock lvbs
* - introduction of "rw" lock and pushing meta/data locking down
*/
-#define O2NET_PROTOCOL_VERSION 2ULL
+#define O2NET_PROTOCOL_VERSION 4ULL
struct o2net_handshake {
__be64 protocol_version;
__be64 connector_id;
diff --git a/fs/ocfs2/dcache.c b/fs/ocfs2/dcache.c
index 1a01380e387..014e73978da 100644
--- a/fs/ocfs2/dcache.c
+++ b/fs/ocfs2/dcache.c
@@ -35,15 +35,17 @@
#include "alloc.h"
#include "dcache.h"
+#include "dlmglue.h"
#include "file.h"
#include "inode.h"
+
static int ocfs2_dentry_revalidate(struct dentry *dentry,
struct nameidata *nd)
{
struct inode *inode = dentry->d_inode;
int ret = 0; /* if all else fails, just return false */
- struct ocfs2_super *osb;
+ struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
mlog_entry("(0x%p, '%.*s')\n", dentry,
dentry->d_name.len, dentry->d_name.name);
@@ -55,28 +57,31 @@ static int ocfs2_dentry_revalidate(struct dentry *dentry,
goto bail;
}
- osb = OCFS2_SB(inode->i_sb);
-
BUG_ON(!osb);
- if (inode != osb->root_inode) {
- spin_lock(&OCFS2_I(inode)->ip_lock);
- /* did we or someone else delete this inode? */
- if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
- spin_unlock(&OCFS2_I(inode)->ip_lock);
- mlog(0, "inode (%llu) deleted, returning false\n",
- (unsigned long long)OCFS2_I(inode)->ip_blkno);
- goto bail;
- }
+ if (inode == osb->root_inode || is_bad_inode(inode))
+ goto bail;
+
+ spin_lock(&OCFS2_I(inode)->ip_lock);
+ /* did we or someone else delete this inode? */
+ if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
spin_unlock(&OCFS2_I(inode)->ip_lock);
+ mlog(0, "inode (%llu) deleted, returning false\n",
+ (unsigned long long)OCFS2_I(inode)->ip_blkno);
+ goto bail;
+ }
+ spin_unlock(&OCFS2_I(inode)->ip_lock);
- if (!inode->i_nlink) {
- mlog(0, "Inode %llu orphaned, returning false "
- "dir = %d\n",
- (unsigned long long)OCFS2_I(inode)->ip_blkno,
- S_ISDIR(inode->i_mode));
- goto bail;
- }
+ /*
+ * We don't need a cluster lock to test this because once an
+ * inode nlink hits zero, it never goes back.
+ */
+ if (inode->i_nlink == 0) {
+ mlog(0, "Inode %llu orphaned, returning false "
+ "dir = %d\n",
+ (unsigned long long)OCFS2_I(inode)->ip_blkno,
+ S_ISDIR(inode->i_mode));
+ goto bail;
}
ret = 1;
@@ -87,6 +92,322 @@ bail:
return ret;
}
+static int ocfs2_match_dentry(struct dentry *dentry,
+ u64 parent_blkno,
+ int skip_unhashed)
+{
+ struct inode *parent;
+
+ /*
+ * ocfs2_lookup() does a d_splice_alias() _before_ attaching
+ * to the lock data, so we skip those here, otherwise
+ * ocfs2_dentry_attach_lock() will get its original dentry
+ * back.
+ */
+ if (!dentry->d_fsdata)
+ return 0;
+
+ if (!dentry->d_parent)
+ return 0;
+
+ if (skip_unhashed && d_unhashed(dentry))
+ return 0;
+
+ parent = dentry->d_parent->d_inode;
+ /* Negative parent dentry? */
+ if (!parent)
+ return 0;
+
+ /* Name is in a different directory. */
+ if (OCFS2_I(parent)->ip_blkno != parent_blkno)
+ return 0;
+
+ return 1;
+}
+
+/*
+ * Walk the inode alias list, and find a dentry which has a given
+ * parent. ocfs2_dentry_attach_lock() wants to find _any_ alias as it
+ * is looking for a dentry_lock reference. The vote thread is looking
+ * to unhash aliases, so we allow it to skip any that already have
+ * that property.
+ */
+struct dentry *ocfs2_find_local_alias(struct inode *inode,
+ u64 parent_blkno,
+ int skip_unhashed)
+{
+ struct list_head *p;
+ struct dentry *dentry = NULL;
+
+ spin_lock(&dcache_lock);
+
+ list_for_each(p, &inode->i_dentry) {
+ dentry = list_entry(p, struct dentry, d_alias);
+
+ if (ocfs2_match_dentry(dentry, parent_blkno, skip_unhashed)) {
+ mlog(0, "dentry found: %.*s\n",
+ dentry->d_name.len, dentry->d_name.name);
+
+ dget_locked(dentry);
+ break;
+ }
+
+ dentry = NULL;
+ }
+
+ spin_unlock(&dcache_lock);
+
+ return dentry;
+}
+
+DEFINE_SPINLOCK(dentry_attach_lock);
+
+/*
+ * Attach this dentry to a cluster lock.
+ *
+ * Dentry locks cover all links in a given directory to a particular
+ * inode. We do this so that ocfs2 can build a lock name which all
+ * nodes in the cluster can agree on at all times. Shoving full names
+ * in the cluster lock won't work due to size restrictions. Covering
+ * links inside of a directory is a good compromise because it still
+ * allows us to use the parent directory lock to synchronize
+ * operations.
+ *
+ * Call this function with the parent dir semaphore and the parent dir
+ * cluster lock held.
+ *
+ * The dir semaphore will protect us from having to worry about
+ * concurrent processes on our node trying to attach a lock at the
+ * same time.
+ *
+ * The dir cluster lock (held at either PR or EX mode) protects us
+ * from unlink and rename on other nodes.
+ *
+ * A dput() can happen asynchronously due to pruning, so we cover
+ * attaching and detaching the dentry lock with a
+ * dentry_attach_lock.
+ *
+ * A node which has done lookup on a name retains a protected read
+ * lock until final dput. If the user requests and unlink or rename,
+ * the protected read is upgraded to an exclusive lock. Other nodes
+ * who have seen the dentry will then be informed that they need to
+ * downgrade their lock, which will involve d_delete on the
+ * dentry. This happens in ocfs2_dentry_convert_worker().
+ */
+int ocfs2_dentry_attach_lock(struct dentry *dentry,
+ struct inode *inode,
+ u64 parent_blkno)
+{
+ int ret;
+ struct dentry *alias;
+ struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
+
+ mlog(0, "Attach \"%.*s\", parent %llu, fsdata: %p\n",
+ dentry->d_name.len, dentry->d_name.name,
+ (unsigned long long)parent_blkno, dl);
+
+ /*
+ * Negative dentry. We ignore these for now.
+ *
+ * XXX: Could we can improve ocfs2_dentry_revalidate() by
+ * tracking these?
+ */
+ if (!inode)
+ return 0;
+
+ if (dl) {
+ mlog_bug_on_msg(dl->dl_parent_blkno != parent_blkno,
+ " \"%.*s\": old parent: %llu, new: %llu\n",
+ dentry->d_name.len, dentry->d_name.name,
+ (unsigned long long)parent_blkno,
+ (unsigned long long)dl->dl_parent_blkno);
+ return 0;
+ }
+
+ alias = ocfs2_find_local_alias(inode, parent_blkno, 0);
+ if (alias) {
+ /*
+ * Great, an alias exists, which means we must have a
+ * dentry lock already. We can just grab the lock off
+ * the alias and add it to the list.
+ *
+ * We're depending here on the fact that this dentry
+ * was found and exists in the dcache and so must have
+ * a reference to the dentry_lock because we can't
+ * race creates. Final dput() cannot happen on it
+ * since we have it pinned, so our reference is safe.
+ */
+ dl = alias->d_fsdata;
+ mlog_bug_on_msg(!dl, "parent %llu, ino %llu\n",
+ (unsigned long long)parent_blkno,
+ (unsigned long long)OCFS2_I(inode)->ip_blkno);
+
+ mlog_bug_on_msg(dl->dl_parent_blkno != parent_blkno,
+ " \"%.*s\": old parent: %llu, new: %llu\n",
+ dentry->d_name.len, dentry->d_name.name,
+ (unsigned long long)parent_blkno,
+ (unsigned long long)dl->dl_parent_blkno);
+
+ mlog(0, "Found: %s\n", dl->dl_lockres.l_name);
+
+ goto out_attach;
+ }
+
+ /*
+ * There are no other aliases
+ */
+ dl = kmalloc(sizeof(*dl), GFP_NOFS);
+ if (!dl) {
+ ret = -ENOMEM;
+ mlog_errno(ret);
+ return ret;
+ }
+
+ dl->dl_count = 0;
+ /*
+ * Does this have to happen below, for all attaches, in case
+ * the struct inode gets blown away by votes?
+ */
+ dl->dl_inode = igrab(inode);
+ dl->dl_parent_blkno = parent_blkno;
+ ocfs2_dentry_lock_res_init(dl, parent_blkno, inode);
+
+out_attach:
+ spin_lock(&dentry_attach_lock);
+ dentry->d_fsdata = dl;
+ dl->dl_count++;
+ spin_unlock(&dentry_attach_lock);
+
+ /*
+ * This actually gets us our PRMODE level lock. From now on,
+ * we'll have a notification if one of these names is
+ * destroyed on another node.
+ */
+ ret = ocfs2_dentry_lock(dentry, 0);
+ if (!ret)
+ ocfs2_dentry_unlock(dentry, 0);
+ else
+ mlog_errno(ret);
+
+ dput(alias);
+
+ return ret;
+}
+
+/*
+ * ocfs2_dentry_iput() and friends.
+ *
+ * At this point, our particular dentry is detached from the inodes
+ * alias list, so there's no way that the locking code can find it.
+ *
+ * The interesting stuff happens when we determine that our lock needs
+ * to go away because this is the last subdir alias in the
+ * system. This function needs to handle a couple things:
+ *
+ * 1) Synchronizing lock shutdown with the downconvert threads. This
+ * is already handled for us via the lockres release drop function
+ * called in ocfs2_release_dentry_lock()
+ *
+ * 2) A race may occur when we're doing our lock shutdown and
+ * another process wants to create a new dentry lock. Right now we
+ * let them race, which means that for a very short while, this
+ * node might have two locks on a lock resource. This should be a
+ * problem though because one of them is in the process of being
+ * thrown out.
+ */
+static void ocfs2_drop_dentry_lock(struct ocfs2_super *osb,
+ struct ocfs2_dentry_lock *dl)
+{
+ ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
+ ocfs2_lock_res_free(&dl->dl_lockres);
+ iput(dl->dl_inode);
+ kfree(dl);
+}
+
+void ocfs2_dentry_lock_put(struct ocfs2_super *osb,
+ struct ocfs2_dentry_lock *dl)
+{
+ int unlock = 0;
+
+ BUG_ON(dl->dl_count == 0);
+
+ spin_lock(&dentry_attach_lock);
+ dl->dl_count--;
+ unlock = !dl->dl_count;
+ spin_unlock(&dentry_attach_lock);
+
+ if (unlock)
+ ocfs2_drop_dentry_lock(osb, dl);
+}
+
+static void ocfs2_dentry_iput(struct dentry *dentry, struct inode *inode)
+{
+ struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
+
+ mlog_bug_on_msg(!dl && !(dentry->d_flags & DCACHE_DISCONNECTED),
+ "dentry: %.*s\n", dentry->d_name.len,
+ dentry->d_name.name);
+
+ if (!dl)
+ goto out;
+
+ mlog_bug_on_msg(dl->dl_count == 0, "dentry: %.*s, count: %u\n",
+ dentry->d_name.len, dentry->d_name.name,
+ dl->dl_count);
+
+ ocfs2_dentry_lock_put(OCFS2_SB(dentry->d_sb), dl);
+
+out:
+ iput(inode);
+}
+
+/*
+ * d_move(), but keep the locks in sync.
+ *
+ * When we are done, "dentry" will have the parent dir and name of
+ * "target", which will be thrown away.
+ *
+ * We manually update the lock of "dentry" if need be.
+ *
+ * "target" doesn't have it's dentry lock touched - we allow the later
+ * dput() to handle this for us.
+ *
+ * This is called during ocfs2_rename(), while holding parent
+ * directory locks. The dentries have already been deleted on other
+ * nodes via ocfs2_remote_dentry_delete().
+ *
+ * Normally, the VFS handles the d_move() for the file sytem, after
+ * the ->rename() callback. OCFS2 wants to handle this internally, so
+ * the new lock can be created atomically with respect to the cluster.
+ */
+void ocfs2_dentry_move(struct dentry *dentry, struct dentry *target,
+ struct inode *old_dir, struct inode *new_dir)
+{
+ int ret;
+ struct ocfs2_super *osb = OCFS2_SB(old_dir->i_sb);
+ struct inode *inode = dentry->d_inode;
+
+ /*
+ * Move within the same directory, so the actual lock info won't
+ * change.
+ *
+ * XXX: Is there any advantage to dropping the lock here?
+ */
+ if (old_dir == new_dir)
+ goto out_move;
+
+ ocfs2_dentry_lock_put(osb, dentry->d_fsdata);
+
+ dentry->d_fsdata = NULL;
+ ret = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(new_dir)->ip_blkno);
+ if (ret)
+ mlog_errno(ret);
+
+out_move:
+ d_move(dentry, target);
+}
+
struct dentry_operations ocfs2_dentry_ops = {
.d_revalidate = ocfs2_dentry_revalidate,
+ .d_iput = ocfs2_dentry_iput,
};
diff --git a/fs/ocfs2/dcache.h b/fs/ocfs2/dcache.h
index 90072771114..c091c34d988 100644
--- a/fs/ocfs2/dcache.h
+++ b/fs/ocfs2/dcache.h
@@ -28,4 +28,31 @@
extern struct dentry_operations ocfs2_dentry_ops;
+struct ocfs2_dentry_lock {
+ unsigned int dl_count;
+ u64 dl_parent_blkno;
+
+ /*
+ * The ocfs2_dentry_lock keeps an inode reference until
+ * dl_lockres has been destroyed. This is usually done in
+ * ->d_iput() anyway, so there should be minimal impact.
+ */
+ struct inode *dl_inode;
+ struct ocfs2_lock_res dl_lockres;
+};
+
+int ocfs2_dentry_attach_lock(struct dentry *dentry, struct inode *inode,
+ u64 parent_blkno);
+
+void ocfs2_dentry_lock_put(struct ocfs2_super *osb,
+ struct ocfs2_dentry_lock *dl);
+
+struct dentry *ocfs2_find_local_alias(struct inode *inode, u64 parent_blkno,
+ int skip_unhashed);
+
+void ocfs2_dentry_move(struct dentry *dentry, struct dentry *target,
+ struct inode *old_dir, struct inode *new_dir);
+
+extern spinlock_t dentry_attach_lock;
+
#endif /* OCFS2_DCACHE_H */
diff --git a/fs/ocfs2/dir.c b/fs/ocfs2/dir.c
index 3d494d1a5f3..04e01915b86 100644
--- a/fs/ocfs2/dir.c
+++ b/fs/ocfs2/dir.c
@@ -74,14 +74,14 @@ static int ocfs2_extend_dir(struct ocfs2_super *osb,
int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir)
{
int error = 0;
- unsigned long offset, blk;
- int i, num, stored;
+ unsigned long offset, blk, last_ra_blk = 0;
+ int i, stored;
struct buffer_head * bh, * tmp;
struct ocfs2_dir_entry * de;