diff options
author | Jeff Mahoney <jeffm@suse.com> | 2009-03-30 14:02:39 -0400 |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2009-03-30 12:16:38 -0700 |
commit | 0ab2621ebd9a28bf7a524ecd50d492a10579dfcc (patch) | |
tree | 62dda6de2fed116aff363190f95a58d56c690e3e /fs | |
parent | 48b32a3553a54740d236b79a90f20147a25875e3 (diff) | |
download | op-kernel-dev-0ab2621ebd9a28bf7a524ecd50d492a10579dfcc.zip op-kernel-dev-0ab2621ebd9a28bf7a524ecd50d492a10579dfcc.tar.gz |
reiserfs: journaled xattrs
Deadlocks are possible in the xattr code between the journal lock and the
xattr sems.
This patch implements journalling for xattr operations. The benefit is
twofold:
* It gets rid of the deadlock possibility by always ensuring that xattr
write operations are initiated inside a transaction.
* It corrects the problem where xattr backing files aren't considered any
differently than normal files, despite the fact they are metadata.
I discussed the added journal load with Chris Mason, and we decided that
since xattrs (versus other journal activity) is fairly rare, the introduction
of larger transactions to support journaled xattrs wouldn't be too big a deal.
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Diffstat (limited to 'fs')
-rw-r--r-- | fs/reiserfs/inode.c | 3 | ||||
-rw-r--r-- | fs/reiserfs/namei.c | 14 | ||||
-rw-r--r-- | fs/reiserfs/xattr.c | 39 | ||||
-rw-r--r-- | fs/reiserfs/xattr_acl.c | 105 |
4 files changed, 115 insertions, 46 deletions
diff --git a/fs/reiserfs/inode.c b/fs/reiserfs/inode.c index 50a73e7..995f697 100644 --- a/fs/reiserfs/inode.c +++ b/fs/reiserfs/inode.c @@ -1914,9 +1914,8 @@ int reiserfs_new_inode(struct reiserfs_transaction_handle *th, goto out_inserted_sd; } - /* XXX CHECK THIS */ if (reiserfs_posixacl(inode->i_sb)) { - retval = reiserfs_inherit_default_acl(dir, dentry, inode); + retval = reiserfs_inherit_default_acl(th, dir, dentry, inode); if (retval) { err = retval; reiserfs_check_path(&path_to_key); diff --git a/fs/reiserfs/namei.c b/fs/reiserfs/namei.c index ddf1bcd..d9c1c8b 100644 --- a/fs/reiserfs/namei.c +++ b/fs/reiserfs/namei.c @@ -598,15 +598,13 @@ static int reiserfs_create(struct inode *dir, struct dentry *dentry, int mode, 2 * (REISERFS_QUOTA_INIT_BLOCKS(dir->i_sb) + REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb)); struct reiserfs_transaction_handle th; - int locked; if (!(inode = new_inode(dir->i_sb))) { return -ENOMEM; } new_inode_init(inode, dir, mode); - locked = reiserfs_cache_default_acl(dir); - + jbegin_count += reiserfs_cache_default_acl(dir); reiserfs_write_lock(dir->i_sb); retval = journal_begin(&th, dir->i_sb, jbegin_count); @@ -662,7 +660,6 @@ static int reiserfs_mknod(struct inode *dir, struct dentry *dentry, int mode, JOURNAL_PER_BALANCE_CNT * 3 + 2 * (REISERFS_QUOTA_INIT_BLOCKS(dir->i_sb) + REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb)); - int locked; if (!new_valid_dev(rdev)) return -EINVAL; @@ -672,8 +669,7 @@ static int reiserfs_mknod(struct inode *dir, struct dentry *dentry, int mode, } new_inode_init(inode, dir, mode); - locked = reiserfs_cache_default_acl(dir); - + jbegin_count += reiserfs_cache_default_acl(dir); reiserfs_write_lock(dir->i_sb); retval = journal_begin(&th, dir->i_sb, jbegin_count); @@ -732,7 +728,6 @@ static int reiserfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) JOURNAL_PER_BALANCE_CNT * 3 + 2 * (REISERFS_QUOTA_INIT_BLOCKS(dir->i_sb) + REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb)); - int locked; #ifdef DISPLACE_NEW_PACKING_LOCALITIES /* set flag that new packing locality created and new blocks for the content * of that directory are not displaced yet */ @@ -744,8 +739,7 @@ static int reiserfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) } new_inode_init(inode, dir, mode); - locked = reiserfs_cache_default_acl(dir); - + jbegin_count += reiserfs_cache_default_acl(dir); reiserfs_write_lock(dir->i_sb); retval = journal_begin(&th, dir->i_sb, jbegin_count); @@ -1034,8 +1028,6 @@ static int reiserfs_symlink(struct inode *parent_dir, memcpy(name, symname, strlen(symname)); padd_item(name, item_len, strlen(symname)); - /* We would inherit the default ACL here, but symlinks don't get ACLs */ - retval = journal_begin(&th, parent_dir->i_sb, jbegin_count); if (retval) { drop_new_inode(inode); diff --git a/fs/reiserfs/xattr.c b/fs/reiserfs/xattr.c index d3ce274..c2e3a92 100644 --- a/fs/reiserfs/xattr.c +++ b/fs/reiserfs/xattr.c @@ -632,8 +632,9 @@ out_dput: * inode->i_mutex: down */ int -__reiserfs_xattr_set(struct inode *inode, const char *name, const void *buffer, - size_t buffer_size, int flags) +reiserfs_xattr_set_handle(struct reiserfs_transaction_handle *th, + struct inode *inode, const char *name, + const void *buffer, size_t buffer_size, int flags) { int err = 0; struct dentry *dentry; @@ -723,14 +724,34 @@ out_unlock: return err; } -int -reiserfs_xattr_set(struct inode *inode, const char *name, const void *buffer, - size_t buffer_size, int flags) +/* We need to start a transaction to maintain lock ordering */ +int reiserfs_xattr_set(struct inode *inode, const char *name, + const void *buffer, size_t buffer_size, int flags) { - int err = __reiserfs_xattr_set(inode, name, buffer, buffer_size, flags); - if (err == -ENODATA) - err = 0; - return err; + + struct reiserfs_transaction_handle th; + int error, error2; + size_t jbegin_count = reiserfs_xattr_nblocks(inode, buffer_size); + + if (!(flags & XATTR_REPLACE)) + jbegin_count += reiserfs_xattr_jcreate_nblocks(inode); + + reiserfs_write_lock(inode->i_sb); + error = journal_begin(&th, inode->i_sb, jbegin_count); + if (error) { + reiserfs_write_unlock(inode->i_sb); + return error; + } + + error = reiserfs_xattr_set_handle(&th, inode, name, + buffer, buffer_size, flags); + + error2 = journal_end(&th, inode->i_sb, jbegin_count); + if (error == 0) + error = error2; + reiserfs_write_unlock(inode->i_sb); + + return error; } /* diff --git a/fs/reiserfs/xattr_acl.c b/fs/reiserfs/xattr_acl.c index bfecf75..d423416 100644 --- a/fs/reiserfs/xattr_acl.c +++ b/fs/reiserfs/xattr_acl.c @@ -10,15 +10,17 @@ #include <linux/reiserfs_acl.h> #include <asm/uaccess.h> -static int reiserfs_set_acl(struct inode *inode, int type, +static int reiserfs_set_acl(struct reiserfs_transaction_handle *th, + struct inode *inode, int type, struct posix_acl *acl); static int xattr_set_acl(struct inode *inode, int type, const void *value, size_t size) { struct posix_acl *acl; - int error; - + int error, error2; + struct reiserfs_transaction_handle th; + size_t jcreate_blocks; if (!reiserfs_posixacl(inode->i_sb)) return -EOPNOTSUPP; if (!is_owner_or_cap(inode)) @@ -36,7 +38,21 @@ xattr_set_acl(struct inode *inode, int type, const void *value, size_t size) } else acl = NULL; - error = reiserfs_set_acl(inode, type, acl); + /* Pessimism: We can't assume that anything from the xattr root up + * has been created. */ + + jcreate_blocks = reiserfs_xattr_jcreate_nblocks(inode) + + reiserfs_xattr_nblocks(inode, size) * 2; + + reiserfs_write_lock(inode->i_sb); + error = journal_begin(&th, inode->i_sb, jcreate_blocks); + if (error == 0) { + error = reiserfs_set_acl(&th, inode, type, acl); + error2 = journal_end(&th, inode->i_sb, jcreate_blocks); + if (error2) + error = error2; + } + reiserfs_write_unlock(inode->i_sb); release_and_out: posix_acl_release(acl); @@ -266,7 +282,8 @@ struct posix_acl *reiserfs_get_acl(struct inode *inode, int type) * BKL held [before 2.5.x] */ static int -reiserfs_set_acl(struct inode *inode, int type, struct posix_acl *acl) +reiserfs_set_acl(struct reiserfs_transaction_handle *th, struct inode *inode, + int type, struct posix_acl *acl) { char *name; void *value = NULL; @@ -310,7 +327,7 @@ reiserfs_set_acl(struct inode *inode, int type, struct posix_acl *acl) return (int)PTR_ERR(value); } - error = __reiserfs_xattr_set(inode, name, value, size, 0); + error = reiserfs_xattr_set_handle(th, inode, name, value, size, 0); /* * Ensure that the inode gets dirtied if we're only using @@ -337,7 +354,8 @@ reiserfs_set_acl(struct inode *inode, int type, struct posix_acl *acl) /* dir->i_mutex: locked, * inode is new and not released into the wild yet */ int -reiserfs_inherit_default_acl(struct inode *dir, struct dentry *dentry, +reiserfs_inherit_default_acl(struct reiserfs_transaction_handle *th, + struct inode *dir, struct dentry *dentry, struct inode *inode) { struct posix_acl *acl; @@ -374,7 +392,8 @@ reiserfs_inherit_default_acl(struct inode *dir, struct dentry *dentry, /* Copy the default ACL to the default ACL of a new directory */ if (S_ISDIR(inode->i_mode)) { - err = reiserfs_set_acl(inode, ACL_TYPE_DEFAULT, acl); + err = reiserfs_set_acl(th, inode, ACL_TYPE_DEFAULT, + acl); if (err) goto cleanup; } @@ -395,9 +414,9 @@ reiserfs_inherit_default_acl(struct inode *dir, struct dentry *dentry, /* If we need an ACL.. */ if (need_acl > 0) { - err = - reiserfs_set_acl(inode, ACL_TYPE_ACCESS, - acl_copy); + err = reiserfs_set_acl(th, inode, + ACL_TYPE_ACCESS, + acl_copy); if (err) goto cleanup_copy; } @@ -415,21 +434,45 @@ reiserfs_inherit_default_acl(struct inode *dir, struct dentry *dentry, return err; } -/* Looks up and caches the result of the default ACL. - * We do this so that we don't need to carry the xattr_sem into - * reiserfs_new_inode if we don't need to */ +/* This is used to cache the default acl before a new object is created. + * The biggest reason for this is to get an idea of how many blocks will + * actually be required for the create operation if we must inherit an ACL. + * An ACL write can add up to 3 object creations and an additional file write + * so we'd prefer not to reserve that many blocks in the journal if we can. + * It also has the advantage of not loading the ACL with a transaction open, + * this may seem silly, but if the owner of the directory is doing the + * creation, the ACL may not be loaded since the permissions wouldn't require + * it. + * We return the number of blocks required for the transaction. + */ int reiserfs_cache_default_acl(struct inode *inode) { - int ret = 0; - if (reiserfs_posixacl(inode->i_sb) && !IS_PRIVATE(inode)) { - struct posix_acl *acl; - acl = reiserfs_get_acl(inode, ACL_TYPE_DEFAULT); - ret = (acl && !IS_ERR(acl)); - if (ret) - posix_acl_release(acl); + struct posix_acl *acl; + int nblocks = 0; + + if (IS_PRIVATE(inode)) + return 0; + + acl = reiserfs_get_acl(inode, ACL_TYPE_DEFAULT); + + if (acl && !IS_ERR(acl)) { + int size = reiserfs_acl_size(acl->a_count); + + /* Other xattrs can be created during inode creation. We don't + * want to claim too many blocks, so we check to see if we + * we need to create the tree to the xattrs, and then we + * just want two files. */ + nblocks = reiserfs_xattr_jcreate_nblocks(inode); + nblocks += JOURNAL_BLOCKS_PER_OBJECT(inode->i_sb); + + REISERFS_I(inode)->i_flags |= i_has_xattr_dir; + + /* We need to account for writes + bitmaps for two files */ + nblocks += reiserfs_xattr_nblocks(inode, size) * 4; + posix_acl_release(acl); } - return ret; + return nblocks; } int reiserfs_acl_chmod(struct inode *inode) @@ -455,8 +498,22 @@ int reiserfs_acl_chmod(struct inode *inode) if (!clone) return -ENOMEM; error = posix_acl_chmod_masq(clone, inode->i_mode); - if (!error) - error = reiserfs_set_acl(inode, ACL_TYPE_ACCESS, clone); + if (!error) { + struct reiserfs_transaction_handle th; + size_t size = reiserfs_xattr_nblocks(inode, + reiserfs_acl_size(clone->a_count)); + reiserfs_write_lock(inode->i_sb); + error = journal_begin(&th, inode->i_sb, size * 2); + if (!error) { + int error2; + error = reiserfs_set_acl(&th, inode, ACL_TYPE_ACCESS, + clone); + error2 = journal_end(&th, inode->i_sb, size * 2); + if (error2) + error = error2; + } + reiserfs_write_unlock(inode->i_sb); + } posix_acl_release(clone); return error; } |