[PATCH] r/o bind mounts: unlink: monitor i_nlink
[safe/jmp/linux-2.6] / fs / ocfs2 / namei.c
index d8161a7..40f83f5 100644 (file)
@@ -56,6 +56,7 @@
 #include "journal.h"
 #include "namei.h"
 #include "suballoc.h"
+#include "super.h"
 #include "symlink.h"
 #include "sysfile.h"
 #include "uptodate.h"
@@ -178,7 +179,7 @@ static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
        if (status < 0)
                goto bail_add;
 
-       inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno);
+       inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0);
        if (IS_ERR(inode)) {
                mlog(ML_ERROR, "Unable to create inode %llu\n",
                     (unsigned long long)blkno);
@@ -198,10 +199,32 @@ static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
        spin_unlock(&oi->ip_lock);
 
 bail_add:
-
        dentry->d_op = &ocfs2_dentry_ops;
        ret = d_splice_alias(inode, dentry);
 
+       if (inode) {
+               /*
+                * If d_splice_alias() finds a DCACHE_DISCONNECTED
+                * dentry, it will d_move() it on top of ourse. The
+                * return value will indicate this however, so in
+                * those cases, we switch them around for the locking
+                * code.
+                *
+                * NOTE: This dentry already has ->d_op set from
+                * ocfs2_get_parent() and ocfs2_get_dentry()
+                */
+               if (ret)
+                       dentry = ret;
+
+               status = ocfs2_dentry_attach_lock(dentry, inode,
+                                                 OCFS2_I(dir)->ip_blkno);
+               if (status) {
+                       mlog_errno(status);
+                       ret = ERR_PTR(status);
+                       goto bail_unlock;
+               }
+       }
+
 bail_unlock:
        /* Don't drop the cluster lock until *after* the d_add --
         * unlink on another node will message us to remove that
@@ -417,6 +440,13 @@ static int ocfs2_mknod(struct inode *dir,
                goto leave;
        }
 
+       status = ocfs2_dentry_attach_lock(dentry, inode,
+                                         OCFS2_I(dir)->ip_blkno);
+       if (status) {
+               mlog_errno(status);
+               goto leave;
+       }
+
        insert_inode_hash(inode);
        dentry->d_op = &ocfs2_dentry_ops;
        d_instantiate(dentry, inode);
@@ -641,11 +671,6 @@ static int ocfs2_link(struct dentry *old_dentry,
                goto bail;
        }
 
-       if (inode->i_nlink >= OCFS2_LINK_MAX) {
-               err = -EMLINK;
-               goto bail;
-       }
-
        handle = ocfs2_alloc_handle(osb);
        if (handle == NULL) {
                err = -ENOMEM;
@@ -659,6 +684,11 @@ static int ocfs2_link(struct dentry *old_dentry,
                goto bail;
        }
 
+       if (!dir->i_nlink) {
+               err = -ENOENT;
+               goto bail;
+       }
+
        err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
                                        dentry->d_name.len);
        if (err)
@@ -709,7 +739,7 @@ static int ocfs2_link(struct dentry *old_dentry,
        err = ocfs2_journal_dirty(handle, fe_bh);
        if (err < 0) {
                le16_add_cpu(&fe->i_links_count, -1);
-               inode->i_nlink--;
+               drop_nlink(inode);
                mlog_errno(err);
                goto bail;
        }
@@ -719,7 +749,13 @@ static int ocfs2_link(struct dentry *old_dentry,
                              parent_fe_bh, de_bh);
        if (err) {
                le16_add_cpu(&fe->i_links_count, -1);
-               inode->i_nlink--;
+               drop_nlink(inode);
+               mlog_errno(err);
+               goto bail;
+       }
+
+       err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
+       if (err) {
                mlog_errno(err);
                goto bail;
        }
@@ -742,6 +778,23 @@ bail:
        return err;
 }
 
+/*
+ * Takes and drops an exclusive lock on the given dentry. This will
+ * force other nodes to drop it.
+ */
+static int ocfs2_remote_dentry_delete(struct dentry *dentry)
+{
+       int ret;
+
+       ret = ocfs2_dentry_lock(dentry, 1);
+       if (ret)
+               mlog_errno(ret);
+       else
+               ocfs2_dentry_unlock(dentry, 1);
+
+       return ret;
+}
+
 static int ocfs2_unlink(struct inode *dir,
                        struct dentry *dentry)
 {
@@ -831,8 +884,7 @@ static int ocfs2_unlink(struct inode *dir,
        else
                inode->i_nlink--;
 
-       status = ocfs2_request_unlink_vote(inode, dentry,
-                                          (unsigned int) inode->i_nlink);
+       status = ocfs2_remote_dentry_delete(dentry);
        if (status < 0) {
                /* This vote should succeed under all normal
                 * circumstances. */
@@ -1018,7 +1070,6 @@ static int ocfs2_rename(struct inode *old_dir,
        struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
                                                    // this is the 1st dirent bh
        nlink_t old_dir_nlink = old_dir->i_nlink, new_dir_nlink = new_dir->i_nlink;
-       unsigned int links_count;
 
        /* At some point it might be nice to break this function up a
         * bit. */
@@ -1092,23 +1143,26 @@ static int ocfs2_rename(struct inode *old_dir,
                }
        }
 
-       if (S_ISDIR(old_inode->i_mode)) {
-               /* Directories actually require metadata updates to
-                * the directory info so we can't get away with not
-                * doing node locking on it. */
-               status = ocfs2_meta_lock(old_inode, handle, NULL, 1);
-               if (status < 0) {
-                       if (status != -ENOENT)
-                               mlog_errno(status);
-                       goto bail;
-               }
-
-               status = ocfs2_request_rename_vote(old_inode, old_dentry);
-               if (status < 0) {
+       /*
+        * Though we don't require an inode meta data update if
+        * old_inode is not a directory, we lock anyway here to ensure
+        * the vote thread on other nodes won't have to concurrently
+        * downconvert the inode and the dentry locks.
+        */
+       status = ocfs2_meta_lock(old_inode, handle, NULL, 1);
+       if (status < 0) {
+               if (status != -ENOENT)
                        mlog_errno(status);
-                       goto bail;
-               }
+               goto bail;
+       }
 
+       status = ocfs2_remote_dentry_delete(old_dentry);
+       if (status < 0) {
+               mlog_errno(status);
+               goto bail;
+       }
+
+       if (S_ISDIR(old_inode->i_mode)) {
                status = -EIO;
                old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
                if (!old_inode_de_bh)
@@ -1122,14 +1176,6 @@ static int ocfs2_rename(struct inode *old_dir,
                if (!new_inode && new_dir!=old_dir &&
                    new_dir->i_nlink >= OCFS2_LINK_MAX)
                        goto bail;
-       } else {
-               /* Ah, the simple case - we're a file so just send a
-                * message. */
-               status = ocfs2_request_rename_vote(old_inode, old_dentry);
-               if (status < 0) {
-                       mlog_errno(status);
-                       goto bail;
-               }
        }
 
        status = -ENOENT;
@@ -1201,13 +1247,7 @@ static int ocfs2_rename(struct inode *old_dir,
                        goto bail;
                }
 
-               if (S_ISDIR(new_inode->i_mode))
-                       links_count = 0;
-               else
-                       links_count = (unsigned int) (new_inode->i_nlink - 1);
-
-               status = ocfs2_request_unlink_vote(new_inode, new_dentry,
-                                                  links_count);
+               status = ocfs2_remote_dentry_delete(new_dentry);
                if (status < 0) {
                        mlog_errno(status);
                        goto bail;
@@ -1386,6 +1426,7 @@ static int ocfs2_rename(struct inode *old_dir,
                }
        }
 
+       ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
        status = 0;
 bail:
        if (rename_lock)
@@ -1674,6 +1715,12 @@ static int ocfs2_symlink(struct inode *dir,
                goto bail;
        }
 
+       status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
+       if (status) {
+               mlog_errno(status);
+               goto bail;
+       }
+
        insert_inode_hash(inode);
        dentry->d_op = &ocfs2_dentry_ops;
        d_instantiate(dentry, inode);
@@ -1962,13 +2009,8 @@ restart:
                                }
                                num++;
 
-                               /* XXX: questionable readahead stuff here */
                                bh = ocfs2_bread(dir, b++, &err, 1);
                                bh_use[ra_max] = bh;
-#if 0          // ???
-                               if (bh)
-                                       ll_rw_block(READ, 1, &bh);
-#endif
                        }
                }
                if ((bh = bh_use[ra_ptr++]) == NULL)
@@ -1976,6 +2018,10 @@ restart:
                wait_on_buffer(bh);
                if (!buffer_uptodate(bh)) {
                        /* read error, skip block & hope for the best */
+                       ocfs2_error(dir->i_sb, "reading directory %llu, "
+                                   "offset %lu\n",
+                                   (unsigned long long)OCFS2_I(dir)->ip_blkno,
+                                   block);
                        brelse(bh);
                        goto next;
                }