* You should have received a copy of the GNU General Public License
* along with Lustre; if not, write to the Free Software
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
- *
- * derived in small part from linux/fs/ext2/namei.c
- *
- * Copyright (C) 1991, 1992 Linus Torvalds
- *
- * Big-endian to little-endian byte-swapping/bitmaps by
- * David S. Miller (davem@caip.rutgers.edu), 1995
- * Directory entry file type support and forward compatibility hooks
- * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
*/
#include <linux/fs.h>
#define DEBUG_SUBSYSTEM S_LLITE
-#include <linux/obd_support.h>
-#include <linux/lustre_lite.h>
-#include <linux/lustre_dlm.h>
+#include <obd_support.h>
+#include <lustre_fid.h>
+#include <lustre_lite.h>
+#include <lustre_dlm.h>
+#include <lustre_ver.h>
+#include <lustre_mdc.h>
#include "llite_internal.h"
/* methods */
+extern struct dentry_operations ll_d_ops;
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
-static int ll_test_inode(struct inode *inode, unsigned long ino, void *opaque)
-#else
-static int ll_test_inode(struct inode *inode, void *opaque)
-#endif
+/*
+ * Check if we have something mounted at the named dchild.
+ * In such a case there would always be dentry present.
+ */
+static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
+ struct qstr *name)
{
- struct lustre_md *md = opaque;
-
- if (!(md->body->valid & (OBD_MD_FLGENER | OBD_MD_FLID)))
- CERROR("invalid generation\n");
- CDEBUG(D_VFSTRACE, "comparing inode %p ino %lu/%u to body %u/%u\n",
- inode, inode->i_ino, inode->i_generation,
- md->body->ino, md->body->generation);
-
- if (inode->i_generation != md->body->generation)
- return 0;
-
- /* Apply the attributes in 'opaque' to this inode */
- ll_update_inode(inode, md->body, md->lsm);
- return 1;
+ int mounted = 0;
+
+ if (unlikely(dchild)) {
+ mounted = d_mountpoint(dchild);
+ } else if (dparent) {
+ dchild = d_lookup(dparent, name);
+ if (dchild) {
+ mounted = d_mountpoint(dchild);
+ dput(dchild);
+ }
+ }
+ return mounted;
}
-extern struct dentry_operations ll_d_ops;
-
int ll_unlock(__u32 mode, struct lustre_handle *lockh)
{
ENTRY;
RETURN(0);
}
-/* Get an inode by inode number (already instantiated by the intent lookup).
+/*
+ * Get an inode by inode number (already instantiated by the intent lookup).
* Returns inode or NULL
*/
-#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
-int ll_set_inode(struct inode *inode, void *opaque)
-{
- ll_read_inode2(inode, opaque);
- return 0;
-}
struct inode *ll_iget(struct super_block *sb, ino_t hash,
struct lustre_md *md)
{
+ struct ll_inode_info *lli;
struct inode *inode;
-
LASSERT(hash != 0);
- inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
- if (!inode)
- return (NULL); /* removed ERR_PTR(-ENOMEM) -eeb */
-
- if (inode->i_state & I_NEW)
- unlock_new_inode(inode);
+ inode = iget_locked(sb, hash);
+ if (inode) {
+ if (inode->i_state & I_NEW) {
+ lli = ll_i2info(inode);
+ ll_read_inode2(inode, md);
+ unlock_new_inode(inode);
+ } else {
+ if (!(inode->i_state & (I_FREEING | I_CLEAR)))
+ ll_update_inode(inode, md);
+ }
+ CDEBUG(D_VFSTRACE, "inode: %lu/%u(%p)\n",
+ inode->i_ino, inode->i_generation, inode);
+ }
- // XXX Coda always fills inodes, should Lustre?
- return inode;
-}
-#else
-struct inode *ll_iget(struct super_block *sb, ino_t hash,
- struct lustre_md *md)
-{
- struct inode *inode;
- LASSERT(hash != 0);
- inode = iget4(sb, hash, ll_test_inode, md);
- CDEBUG(D_VFSTRACE, "inode: %lu/%u(%p)\n", inode->i_ino,
- inode->i_generation, inode);
return inode;
}
-#endif
-static int ll_intent_to_lock_mode(struct lookup_intent *it)
-{
- /* CREAT needs to be tested before open (both could be set) */
- if (it->it_op & IT_CREAT)
- return LCK_PW;
- else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP))
- return LCK_PR;
-
- LBUG();
- RETURN(-EINVAL);
-}
-
-int ll_it_open_error(int phase, struct lookup_intent *it)
-{
- if (it_disposition(it, DISP_OPEN_OPEN)) {
- if (phase == DISP_OPEN_OPEN)
- return it->it_status;
- else
- return 0;
- }
+static void ll_drop_negative_dentry(struct inode *dir)
+{
+ struct dentry *dentry, *tmp_alias, *tmp_subdir;
- if (it_disposition(it, DISP_OPEN_CREATE)) {
- if (phase == DISP_OPEN_CREATE)
- return it->it_status;
- else
- return 0;
- }
-
- if (it_disposition(it, DISP_LOOKUP_EXECD)) {
- if (phase == DISP_LOOKUP_EXECD)
- return it->it_status;
- else
- return 0;
- }
-
- if (it_disposition(it, DISP_IT_EXECD)) {
- if (phase == DISP_IT_EXECD)
- return it->it_status;
- else
- return 0;
+ spin_lock(&dcache_lock);
+restart:
+ list_for_each_entry_safe(dentry, tmp_alias,
+ &dir->i_dentry,d_alias) {
+ if (!list_empty(&dentry->d_subdirs)) {
+ struct dentry *child;
+ list_for_each_entry_safe(child, tmp_subdir,
+ &dentry->d_subdirs,
+ d_child) {
+ /* XXX Print some debug here? */
+ if (!child->d_inode)
+ /* Negative dentry. If we were
+ dropping dcache lock, go
+ throught the list again */
+ if (ll_drop_dentry(child))
+ goto restart;
+ }
+ }
}
- CERROR("it disp: %X, status: %d\n", it->it_disposition, it->it_status);
- LBUG();
- return 0;
+ spin_unlock(&dcache_lock);
}
-int ll_mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
- void *data, int flag)
+
+int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
+ void *data, int flag)
{
int rc;
struct lustre_handle lockh;
- struct inode *inode = lock->l_data;
ENTRY;
switch (flag) {
}
break;
case LDLM_CB_CANCELING: {
+ struct inode *inode = ll_inode_from_lock(lock);
+ __u64 bits = lock->l_policy_data.l_inodebits.bits;
+ struct lu_fid *fid;
+
/* Invalidate all dentries associated with this inode */
if (inode == NULL)
break;
- if (lock->l_resource->lr_name.name[0] != inode->i_ino ||
- lock->l_resource->lr_name.name[1] != inode->i_generation) {
- LDLM_ERROR(lock, "data mismatch with ino %lu/%u",
- inode->i_ino, inode->i_generation);
+
+ LASSERT(lock->l_flags & LDLM_FL_CANCELING);
+ if ((bits & MDS_INODELOCK_LOOKUP) &&
+ ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
+ bits &= ~MDS_INODELOCK_LOOKUP;
+ if ((bits & MDS_INODELOCK_UPDATE) &&
+ ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
+ bits &= ~MDS_INODELOCK_UPDATE;
+ if ((bits & MDS_INODELOCK_OPEN) &&
+ ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
+ bits &= ~MDS_INODELOCK_OPEN;
+
+ fid = ll_inode2fid(inode);
+ if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
+ lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
+ lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
+ LDLM_ERROR(lock, "data mismatch with object "
+ DFID" (%p)", PFID(fid), inode);
}
- if (S_ISDIR(inode->i_mode)) {
+
+ if (bits & MDS_INODELOCK_OPEN) {
+ int flags = 0;
+ switch (lock->l_req_mode) {
+ case LCK_CW:
+ flags = FMODE_WRITE;
+ break;
+ case LCK_PR:
+ flags = FMODE_EXEC;
+ break;
+ case LCK_CR:
+ flags = FMODE_READ;
+ break;
+ default:
+ CERROR("Unexpected lock mode for OPEN lock "
+ "%d, inode %ld\n", lock->l_req_mode,
+ inode->i_ino);
+ }
+ ll_md_real_close(inode, flags);
+ }
+
+ if (bits & MDS_INODELOCK_UPDATE)
+ ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
+
+ if (S_ISDIR(inode->i_mode) &&
+ (bits & MDS_INODELOCK_UPDATE)) {
CDEBUG(D_INODE, "invalidating inode %lu\n",
inode->i_ino);
-
- ll_invalidate_inode_pages(inode);
+ truncate_inode_pages(inode->i_mapping, 0);
+ ll_drop_negative_dentry(inode);
}
-#warning FIXME: we should probably free this inode if there are no aliases
if (inode->i_sb->s_root &&
- inode != inode->i_sb->s_root->d_inode)
+ inode != inode->i_sb->s_root->d_inode &&
+ (bits & MDS_INODELOCK_LOOKUP))
ll_unhash_aliases(inode);
+ iput(inode);
break;
}
default:
RETURN(0);
}
-int ll_mdc_cancel_unused(struct lustre_handle *conn, struct inode *inode,
- int flags, void *opaque)
+/* Pack the required supplementary groups into the supplied groups array.
+ * If we don't need to use the groups from the target inode(s) then we
+ * instead pack one or more groups from the user's supplementary group
+ * array in case it might be useful. Not needed if doing an MDS-side upcall. */
+void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
{
- struct ldlm_res_id res_id =
- { .name = {inode->i_ino, inode->i_generation} };
- struct obd_device *obddev = class_conn2obd(conn);
- ENTRY;
- RETURN(ldlm_cli_cancel_unused(obddev->obd_namespace, &res_id, flags,
- opaque));
-}
+ int i;
-void ll_prepare_mdc_op_data(struct mdc_op_data *data,
- struct inode *i1,
- struct inode *i2,
- const char *name,
- int namelen,
- int mode)
-{
- LASSERT(i1);
+ LASSERT(i1 != NULL);
+ LASSERT(suppgids != NULL);
- data->ino1 = i1->i_ino;
- data->gen1 = i1->i_generation;
- data->typ1 = i1->i_mode & S_IFMT;
- data->gid1 = i1->i_gid;
+ if (in_group_p(i1->i_gid))
+ suppgids[0] = i1->i_gid;
+ else
+ suppgids[0] = -1;
if (i2) {
- data->ino2 = i2->i_ino;
- data->gen2 = i2->i_generation;
- data->typ2 = i2->i_mode & S_IFMT;
- data->gid2 = i2->i_gid;
+ if (in_group_p(i2->i_gid))
+ suppgids[1] = i2->i_gid;
+ else
+ suppgids[1] = -1;
} else {
- data->ino2 = 0;
+ suppgids[1] = -1;
}
- data->name = name;
- data->namelen = namelen;
- data->mode = mode;
+ for (i = 0; i < current_ngroups; i++) {
+ if (suppgids[0] == -1) {
+ if (current_groups[i] != suppgids[1])
+ suppgids[0] = current_groups[i];
+ continue;
+ }
+ if (suppgids[1] == -1) {
+ if (current_groups[i] != suppgids[0])
+ suppgids[1] = current_groups[i];
+ continue;
+ }
+ break;
+ }
}
-/*
- *This long block is all about fixing up the local state so that it is
- *correct as of the moment _before_ the operation was applied; that
- *way, the VFS will think that everything is normal and call Lustre's
- *regular VFS methods.
- *
- * If we're performing a creation, that means that unless the creation
- * failed with EEXIST, we should fake up a negative dentry.
- *
- * For everything else, we want to lookup to succeed.
- *
- * One additional note: if CREATE or OPEN succeeded, we add an extra
- * reference to the request because we need to keep it around until
- * ll_create/ll_open gets called.
- *
- * The server will return to us, in it_disposition, an indication of
- * exactly what it_status refers to.
- *
- * If DISP_OPEN_OPEN is set, then it_status refers to the open() call,
- * otherwise if DISP_OPEN_CREATE is set, then it status is the
- * creation failure mode. In either case, one of DISP_LOOKUP_NEG or
- * DISP_LOOKUP_POS will be set, indicating whether the child lookup
- * was successful.
- *
- * Else, if DISP_LOOKUP_EXECD then it_status is the rc of the child
- * lookup.
- */
-int ll_intent_lock(struct inode *parent, struct dentry **de,
- struct lookup_intent *it, int flags, intent_finish_cb intent_finish)
+static void ll_d_add(struct dentry *de, struct inode *inode)
{
- struct dentry *dentry = *de;
- struct inode *inode = dentry->d_inode;
- struct ll_sb_info *sbi = ll_i2sbi(parent);
- struct lustre_handle lockh;
- struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
- struct ptlrpc_request *request;
- int rc = 0;
- struct mds_body *mds_body;
- int mode;
- obd_id ino = 0;
- ENTRY;
-
-#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
- if (it && it->it_magic != INTENT_MAGIC) {
- CERROR("WARNING: uninitialized intent\n");
+ CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
+ /* d_instantiate */
+ if (!list_empty(&de->d_alias)) {
+ spin_unlock(&dcache_lock);
+ CERROR("dentry %.*s %p alias next %p, prev %p\n",
+ de->d_name.len, de->d_name.name, de,
+ de->d_alias.next, de->d_alias.prev);
LBUG();
- intent_init(it, IT_LOOKUP, 0);
}
- if (it->it_op == IT_GETATTR ||
- it->it_op == 0)
- it->it_op = IT_LOOKUP;
-
-#endif
- if (!it ||it->it_op == IT_GETXATTR)
- it = &lookup_it;
-
- it->it_op_release = ll_intent_release;
-
- CDEBUG(D_DLMTRACE, "name: %*s, intent: %s\n", dentry->d_name.len,
- dentry->d_name.name, ldlm_it2str(it->it_op));
-
- if (dentry->d_name.len > EXT2_NAME_LEN)
- RETURN(-ENAMETOOLONG);
-
- /* This function may be called twice, we only once want to
- execute the request associated with the intent. If it was
- done already, we skip past this and use the results. */
- if (!it_disposition(it, DISP_ENQ_COMPLETE)) {
- struct mdc_op_data op_data;
-
- ll_prepare_mdc_op_data(&op_data, parent, dentry->d_inode,
- dentry->d_name.name, dentry->d_name.len,
- 0);
-
- rc = mdc_enqueue(&sbi->ll_mdc_conn, LDLM_PLAIN, it,
- ll_intent_to_lock_mode(it), &op_data,
- &lockh, NULL, 0, ldlm_completion_ast,
- ll_mdc_blocking_ast, NULL);
- if (rc < 0)
- RETURN(rc);
- memcpy(it->it_lock_handle, &lockh, sizeof(lockh));
- }
- request = it->it_data;
- LASSERT(request != NULL);
-
- if (!it_disposition(it, DISP_IT_EXECD)) {
- /* The server failed before it even started executing the
- * intent, i.e. because it couldn't unpack the request. */
- LASSERT(it->it_status != 0);
- GOTO(drop_req, rc = it->it_status);
- }
- rc = ll_it_open_error(DISP_IT_EXECD, it);
- if (rc)
- GOTO(drop_req, rc);
-
- mds_body = lustre_msg_buf(request->rq_repmsg, 1, sizeof(*mds_body));
- LASSERT(mds_body != NULL); /* mdc_enqueue checked */
- LASSERT_REPSWABBED(request, 1); /* mdc_enqueue swabbed */
-
- /* XXX everything with fids please, no ino's inode's etc */
- ino = mds_body->fid1.id;
- mode = mds_body->mode;
-
- /*We were called from revalidate2: did we find the same inode?*/
- if (inode &&
- (ino != inode->i_ino ||
- mds_body->fid1.generation != inode->i_generation)) {
- it_set_disposition(it, DISP_ENQ_COMPLETE);
- RETURN(-ESTALE);
- }
-
- /* If we're doing an IT_OPEN which did not result in an actual
- * successful open, then we need to remove the bit which saves
- * this request for unconditional replay. */
- if (it->it_op & IT_OPEN) {
- if (!it_disposition(it, DISP_OPEN_OPEN) ||
- it->it_status != 0) {
- unsigned long flags;
-
- spin_lock_irqsave (&request->rq_lock, flags);
- request->rq_replay = 0;
- spin_unlock_irqrestore (&request->rq_lock, flags);
- }
- }
-
- rc = ll_it_open_error(DISP_LOOKUP_EXECD, it);
- if (rc)
- GOTO(drop_req, rc);
-
- /* keep requests around for the multiple phases of the call
- * this shows the DISP_XX must guarantee we make it into the call
- */
- if (it_disposition(it, DISP_OPEN_CREATE))
- ptlrpc_request_addref(request);
- if (it_disposition(it, DISP_OPEN_OPEN))
- ptlrpc_request_addref(request);
-
- if (it->it_op & IT_CREAT) {
- /* XXX this belongs in ll_create_iit */
- } else if (it->it_op == IT_OPEN) {
- LASSERT(!it_disposition(it, DISP_OPEN_CREATE));
- } else
- LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP));
-
- if (intent_finish != NULL) {
- struct lustre_handle old_lock;
- struct ldlm_lock *lock;
-
- rc = intent_finish(request, parent, de, it, 1, ino);
- dentry = *de; /* intent_finish may change *de */
- inode = dentry->d_inode;
- if (rc != 0)
- GOTO(drop_lock, rc);
-
- /* The intent processing may well have given us a lock different
- * from the one we requested. If we already have a matching
- * lock, then cancel the new one. (We have to do this here,
- * instead of in mdc_enqueue, because we need to use the child's
- * inode as the l_data to match, and that's not available until
- * intent_finish has performed the iget().) */
- lock = ldlm_handle2lock(&lockh);
- if (lock) {
- LDLM_DEBUG(lock, "matching against this");
- LDLM_LOCK_PUT(lock);
- memcpy(&old_lock, &lockh, sizeof(lockh));
- if (ldlm_lock_match(NULL,
- LDLM_FL_BLOCK_GRANTED |
- LDLM_FL_MATCH_DATA,
- NULL, LDLM_PLAIN, NULL, 0, LCK_NL,
- inode, &old_lock)) {
- ldlm_lock_decref_and_cancel(&lockh,
- it->it_lock_mode);
- memcpy(&lockh, &old_lock, sizeof(old_lock));
- memcpy(it->it_lock_handle, &lockh,
- sizeof(lockh));
- }
- }
+ if (inode)
+ list_add(&de->d_alias, &inode->i_dentry);
+ de->d_inode = inode;
+ /* d_rehash */
+ if (!d_unhashed(de)) {
+ spin_unlock(&dcache_lock);
+ CERROR("dentry %.*s %p hash next %p\n",
+ de->d_name.len, de->d_name.name, de, de->d_hash.next);
+ LBUG();
}
- ptlrpc_req_finished(request);
-
- CDEBUG(D_DENTRY, "D_IT dentry %p intent: %s status %d disp %x\n",
- dentry, ldlm_it2str(it->it_op), it->it_status, it->it_disposition);
-
- /* drop IT_LOOKUP locks */
- if (it->it_op == IT_LOOKUP)
- ll_intent_release(it);
- RETURN(rc);
-
- drop_lock:
- ll_intent_release(it);
- drop_req:
- ptlrpc_req_finished(request);
- RETURN(rc);
+ __d_rehash(de, 0);
}
-/* Search "inode"'s alias list for a dentry that has the same name and parent as
- * de. If found, return it. If not found, return de. */
+/* Search "inode"'s alias list for a dentry that has the same name and parent
+ * as de. If found, return it. If not found, return de.
+ * Lustre can't use d_add_unique because don't unhash aliases for directory
+ * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
+ * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
+ */
struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
{
struct list_head *tmp;
-
+ struct dentry *dentry;
+ struct dentry *last_discon = NULL;
+
spin_lock(&dcache_lock);
list_for_each(tmp, &inode->i_dentry) {
- struct dentry *dentry = list_entry(tmp, struct dentry, d_alias);
+ dentry = list_entry(tmp, struct dentry, d_alias);
/* We are called here with 'de' already on the aliases list. */
- if (dentry == de) {
+ if (unlikely(dentry == de)) {
CERROR("whoops\n");
continue;
}
+ if (dentry->d_flags & DCACHE_DISCONNECTED) {
+ LASSERT(last_discon == NULL);
+ last_discon = dentry;
+ continue;
+ }
+
if (dentry->d_parent != de->d_parent)
continue;
+ if (dentry->d_name.hash != de->d_name.hash)
+ continue;
+
if (dentry->d_name.len != de->d_name.len)
continue;
de->d_name.len) != 0)
continue;
- if (!list_empty(&dentry->d_lru))
- list_del_init(&dentry->d_lru);
-
- hlist_del_init(&dentry->d_hash);
+ dget_locked(dentry);
+ lock_dentry(dentry);
+ __d_drop(dentry);
+#ifdef LUSTRE_KERNEL_VERSION
+ dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
+#endif
+ unlock_dentry(dentry);
__d_rehash(dentry, 0); /* avoid taking dcache_lock inside */
spin_unlock(&dcache_lock);
- atomic_inc(&dentry->d_count);
iput(inode);
- dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
+ CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
+ "refc %d\n", de->d_name.len, de->d_name.name, de,
+ de->d_parent, de->d_inode, atomic_read(&de->d_count));
return dentry;
}
+
+ if (last_discon) {
+ CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
+ "refc %d\n", last_discon, last_discon->d_inode,
+ atomic_read(&last_discon->d_count));
+ __d_rehash(de, 0);
+ dget_locked(last_discon);
+ __d_move(last_discon, de);
+ spin_unlock(&dcache_lock);
+ d_rehash(de);
+ iput(inode);
+ return last_discon;
+ }
+
+ ll_d_add(de, inode);
spin_unlock(&dcache_lock);
return de;
}
-static int
-lookup2_finish(struct ptlrpc_request *request,
- struct inode *parent, struct dentry **de,
- struct lookup_intent *it, int offset, obd_id ino)
+static int lookup_it_finish(struct ptlrpc_request *request, int offset,
+ struct lookup_intent *it, void *data)
{
+ struct it_cb_data *icbd = data;
+ struct dentry **de = icbd->icbd_childp;
+ struct inode *parent = icbd->icbd_parent;
struct ll_sb_info *sbi = ll_i2sbi(parent);
- struct dentry *dentry = *de, *saved = *de;
struct inode *inode = NULL;
int rc;
/* NB 1 request reference will be taken away by ll_intent_lock()
* when I return */
if (!it_disposition(it, DISP_LOOKUP_NEG)) {
- struct lustre_md md;
ENTRY;
- rc =mdc_req2lustre_md(request, offset, &sbi->ll_osc_conn, &md);
- if (rc)
+ rc = ll_prep_inode(&inode, request, offset,
+ (*de)->d_sb);
+ if (rc)
RETURN(rc);
- inode = ll_iget(dentry->d_sb, ino, &md);
- if (!inode) {
- /* free the lsm if we allocated one above */
- if (md.lsm != NULL)
- obd_free_memmd(&sbi->ll_osc_conn, &md.lsm);
- RETURN(-ENOMEM);
- } else if (md.lsm != NULL &&
- ll_i2info(inode)->lli_smd != md.lsm) {
- obd_free_memmd(&sbi->ll_osc_conn, &md.lsm);
- }
-
- /* If this is a stat, get the authoritative file size */
- if (it->it_op == IT_GETATTR && S_ISREG(inode->i_mode) &&
- ll_i2info(inode)->lli_smd != NULL) {
- struct ldlm_extent extent = {0, OBD_OBJECT_EOF};
- struct lustre_handle lockh = {0};
- struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
- ldlm_error_t rc;
-
- LASSERT(lsm->lsm_object_id != 0);
-
- rc = ll_extent_lock(NULL, inode, lsm, LCK_PR, &extent,
- &lockh);
- if (rc != ELDLM_OK) {
- iput(inode);
- RETURN(-EIO);
- }
- ll_extent_unlock(NULL, inode, lsm, LCK_PR, &lockh);
- }
-
- dentry = *de = ll_find_alias(inode, dentry);
-
- /* We asked for a lock on the directory, and may have been
- * granted a lock on the inode. Just in case, fixup the data
- * pointer. */
CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
inode, inode->i_ino, inode->i_generation);
- ldlm_lock_set_data((struct lustre_handle*)it->it_lock_handle,
- inode);
+ md_set_lock_data(sbi->ll_md_exp,
+ &it->d.lustre.it_lock_handle, inode);
+
+ /* We used to query real size from OSTs here, but actually
+ this is not needed. For stat() calls size would be updated
+ from subsequent do_revalidate()->ll_inode_revalidate_it() in
+ 2.4 and
+ vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
+ Everybody else who needs correct file size would call
+ ll_glimpse_size or some equivalent themselves anyway.
+ Also see bug 7198. */
+
+ *de = ll_find_alias(inode, *de);
} else {
ENTRY;
+ /* Check that parent has UPDATE lock. If there is none, we
+ cannot afford to hash this dentry (done by ll_d_add) as it
+ might get picked up later when UPDATE lock will appear */
+ if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
+ spin_lock(&dcache_lock);
+ ll_d_add(*de, inode);
+ spin_unlock(&dcache_lock);
+ } else {
+ (*de)->d_inode = NULL;
+ /* We do not want to hash the dentry if don`t have a
+ * lock, but if this dentry is later used in d_move,
+ * we'd hit uninitialised list head d_hash, so we just
+ * do this to init d_hash field but leave dentry
+ * unhashed. (bug 10796). */
+ d_rehash(*de);
+ d_drop(*de);
+ }
}
- dentry->d_op = &ll_d_ops;
- ll_set_dd(dentry);
-
- if (dentry == saved)
- d_add(dentry, inode);
+ ll_set_dd(*de);
+ (*de)->d_op = &ll_d_ops;
RETURN(0);
}
static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
- struct lookup_intent *it, int flags)
+ struct lookup_intent *it, int lookup_flags)
{
+ struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
struct dentry *save = dentry, *retval;
+ struct ptlrpc_request *req = NULL;
+ struct md_op_data *op_data;
+ struct it_cb_data icbd;
+ __u32 opc;
int rc;
ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p),intent=%s\n",
- dentry->d_name.name, parent->i_ino, parent->i_generation,
- parent, LL_IT2STR(it));
+ if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
+ RETURN(ERR_PTR(-ENAMETOOLONG));
- if (d_mountpoint(dentry)) {
+ CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
+ dentry->d_name.len, dentry->d_name.name, parent->i_ino,
+ parent->i_generation, parent, LL_IT2STR(it));
+
+ if (d_mountpoint(dentry))
CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
+
+ ll_frob_intent(&it, &lookup_it);
+
+ /* As do_lookup is called before follow_mount, root dentry may be left
+ * not valid, revalidate it here. */
+ if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
+ (it->it_op & (IT_OPEN | IT_CREAT))) {
+ rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
+ if (rc)
+ RETURN(ERR_PTR(rc));
}
- rc = ll_intent_lock(parent, &dentry, it, flags, lookup2_finish);
- if (rc < 0) {
- CDEBUG(D_INFO, "ll_intent_lock: %d\n", rc);
+ icbd.icbd_childp = &dentry;
+ icbd.icbd_parent = parent;
+
+ if (it->it_op & IT_CREAT ||
+ (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
+ opc = LUSTRE_OPC_CREATE;
+ else
+ opc = LUSTRE_OPC_ANY;
+
+ op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
+ dentry->d_name.len, lookup_flags, opc,
+ NULL);
+ if (IS_ERR(op_data))
+ RETURN((void *)op_data);
+
+ it->it_create_mode &= ~current->fs->umask;
+
+ rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
+ lookup_flags, &req, ll_md_blocking_ast, 0);
+ ll_finish_md_op_data(op_data);
+ if (rc < 0)
+ GOTO(out, retval = ERR_PTR(rc));
+
+ rc = lookup_it_finish(req, DLM_REPLY_REC_OFF, it, &icbd);
+ if (rc != 0) {
+ ll_intent_release(it);
GOTO(out, retval = ERR_PTR(rc));
}
+ if ((it->it_op & IT_OPEN) && dentry->d_inode &&
+ !S_ISREG(dentry->d_inode->i_mode) &&
+ !S_ISDIR(dentry->d_inode->i_mode)) {
+ ll_release_openhandle(dentry, it);
+ }
+ ll_lookup_finish_locks(it, dentry);
+
if (dentry == save)
GOTO(out, retval = NULL);
else
GOTO(out, retval = dentry);
out:
+ if (req)
+ ptlrpc_req_finished(req);
return retval;
}
-#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
-static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
+#ifdef LUSTRE_KERNEL_VERSION
+static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
struct nameidata *nd)
{
struct dentry *de;
ENTRY;
- if (nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
- de = ll_lookup_it(parent, dentry, &nd->it, nd->flags);
- else
+ if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
+ de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
+ else
de = ll_lookup_it(parent, dentry, NULL, 0);
RETURN(de);
}
+#else
+struct lookup_intent *ll_convert_intent(struct open_intent *oit,
+ int lookup_flags)
+{
+ struct lookup_intent *it;
+
+ OBD_ALLOC(it, sizeof(*it));
+ if (!it)
+ return ERR_PTR(-ENOMEM);
+
+ if (lookup_flags & LOOKUP_OPEN) {
+ it->it_op = IT_OPEN;
+ if (lookup_flags & LOOKUP_CREATE)
+ it->it_op |= IT_CREAT;
+ it->it_create_mode = oit->create_mode;
+ it->it_flags = oit->flags;
+ } else {
+ it->it_op = IT_GETATTR;
+ }
+
+#ifndef HAVE_FILE_IN_STRUCT_INTENT
+ /* Since there is no way to pass our intent to ll_file_open,
+ * just check the file is there. Actual open will be done
+ * in ll_file_open */
+ if (it->it_op & IT_OPEN)
+ it->it_op = IT_LOOKUP;
#endif
-static int ll_mdc_unlink(struct inode *dir, struct inode *child, __u32 mode,
- const char *name, int len)
+ return it;
+}
+
+static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
+ struct nameidata *nd)
{
- struct ptlrpc_request *request = NULL;
- struct mds_body *body;
- struct lov_mds_md *eadata;
- struct lov_stripe_md *lsm = NULL;
- struct obd_trans_info oti = { 0 };
- struct mdc_op_data op_data;
- struct obdo *oa;
- int rc;
+ struct dentry *de;
ENTRY;
- ll_prepare_mdc_op_data(&op_data, dir, child, name, len, mode);
- rc = mdc_unlink(&ll_i2sbi(dir)->ll_mdc_conn, &op_data, &request);
- if (rc)
- GOTO(out, rc);
- /* req is swabbed so this is safe */
- body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
-
- if (!(body->valid & OBD_MD_FLEASIZE))
- GOTO(out, rc = 0);
-
- if (body->eadatasize == 0) {
- CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
- GOTO(out, rc = -EPROTO);
- }
+ if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
+ struct lookup_intent *it;
- /* The MDS sent back the EA because we unlinked the last reference
- * to this file. Use this EA to unlink the objects on the OST.
- * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
- * check it is complete and sensible. */
- eadata = lustre_swab_repbuf(request, 1, body->eadatasize, NULL);
- LASSERT(eadata != NULL);
- if (eadata == NULL) {
- CERROR("Can't unpack MDS EA data\n");
- GOTO(out, rc = -EPROTO);
- }
+#if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
+ /* Did we came here from failed revalidate just to propagate
+ * its error? */
+ if (nd->flags & LOOKUP_OPEN)
+ if (IS_ERR(nd->intent.open.file))
+ RETURN((struct dentry *)nd->intent.open.file);
+#endif
- rc = obd_unpackmd(ll_i2obdconn(dir), &lsm, eadata, body->eadatasize);
- if (rc < 0) {
- CERROR("obd_unpackmd: %d\n", rc);
- GOTO(out, rc);
- }
- LASSERT(rc >= sizeof(*lsm));
+ if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
+ it = ll_d2d(dentry)->lld_it;
+ ll_d2d(dentry)->lld_it = NULL;
+ } else {
+ it = ll_convert_intent(&nd->intent.open, nd->flags);
+ if (IS_ERR(it))
+ RETURN((struct dentry *)it);
+ }
- oa = obdo_alloc();
- if (oa == NULL)
- GOTO(out_free_memmd, rc = -ENOMEM);
+ de = ll_lookup_it(parent, dentry, it, nd->flags);
+ if (de)
+ dentry = de;
+ if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
+ if (dentry->d_inode &&
+ it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
+#ifdef HAVE_FILE_IN_STRUCT_INTENT
+ if (S_ISFIFO(dentry->d_inode->i_mode)) {
+ // We cannot call open here as it would
+ // deadlock.
+ ptlrpc_req_finished(
+ (struct ptlrpc_request *)
+ it->d.lustre.it_data);
+ } else {
+ struct file *filp;
+ nd->intent.open.file->private_data = it;
+ filp =lookup_instantiate_filp(nd,dentry,
+ NULL);
+#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
+/* 2.6.1[456] have a bug in open_namei() that forgets to check
+ * nd->intent.open.file for error, so we need to return it as lookup's result
+ * instead */
+ if (IS_ERR(filp)) {
+ if (de)
+ dput(de);
+ de = (struct dentry *) filp;
+ }
+#endif
- oa->o_id = lsm->lsm_object_id;
- oa->o_mode = body->mode & S_IFMT;
- oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
+ }
+#else /* HAVE_FILE_IN_STRUCT_INTENT */
+ /* Release open handle as we have no way to
+ * pass it to ll_file_open */
+ ll_release_openhandle(dentry, it);
+#endif /* HAVE_FILE_IN_STRUCT_INTENT */
+ } else if (it_disposition(it, DISP_OPEN_CREATE)) {
+ // XXX This can only reliably work on assumption
+ // that there are NO hashed negative dentries.
+ ll_d2d(dentry)->lld_it = it;
+ it = NULL; /* Will be freed in ll_create_nd */
+ /* We absolutely depend on ll_create_nd to be
+ * called to not leak this intent and possible
+ * data attached to it */
+ }
+ }
- if (body->valid & OBD_MD_FLCOOKIE) {
- oa->o_valid |= OBD_MD_FLCOOKIE;
- oti.oti_logcookies = lustre_msg_buf(request->rq_repmsg, 3,
- body->eadatasize);
+ if (it) {
+ ll_intent_release(it);
+ OBD_FREE(it, sizeof(*it));
+ }
+ } else {
+ de = ll_lookup_it(parent, dentry, NULL, 0);
}
- rc = obd_destroy(ll_i2obdconn(dir), oa, lsm, &oti);
- obdo_free(oa);
- if (rc)
- CERROR("obd destroy objid "LPX64": rc %d\n",
- lsm->lsm_object_id, rc);
- out_free_memmd:
- obd_free_memmd(ll_i2obdconn(dir), &lsm);
- out:
- ptlrpc_req_finished(request);
- return rc;
+ RETURN(de);
}
+#endif
/* We depend on "mode" being set with the proper file type/umask by now */
static struct inode *ll_create_node(struct inode *dir, const char *name,
int mode, __u64 extra,
struct lookup_intent *it)
{
- struct inode *inode;
+ struct inode *inode = NULL;
struct ptlrpc_request *request = NULL;
struct ll_sb_info *sbi = ll_i2sbi(dir);
- struct lustre_md md;
int rc;
ENTRY;
- LASSERT(it && it->it_disposition);
+ LASSERT(it && it->d.lustre.it_disposition);
- ll_invalidate_inode_pages(dir);
-
- request = it->it_data;
- rc = mdc_req2lustre_md(request, 1, &sbi->ll_osc_conn, &md);
- if (rc) {
+ LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
+ request = it->d.lustre.it_data;
+ it_clear_disposition(it, DISP_ENQ_CREATE_REF);
+ rc = ll_prep_inode(&inode, request, DLM_REPLY_REC_OFF, dir->i_sb);
+ if (rc)
GOTO(out, inode = ERR_PTR(rc));
- }
- inode = ll_iget(dir->i_sb, md.body->ino, &md);
- if (!inode || is_bad_inode(inode)) {
- /* XXX might need iput() for bad inode */
- int rc = -EIO;
- CERROR("new_inode -fatal: rc %d\n", rc);
- LBUG();
- GOTO(out, rc);
- }
LASSERT(list_empty(&inode->i_dentry));
- CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
+ /* We asked for a lock on the directory, but were granted a
+ * lock on the inode. Since we finally have an inode pointer,
+ * stuff it in the lock. */
+ CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
inode, inode->i_ino, inode->i_generation);
- ldlm_lock_set_data((struct lustre_handle*)it->it_lock_handle,
- inode);
-
+ md_set_lock_data(sbi->ll_md_exp,
+ &it->d.lustre.it_lock_handle, inode);
EXIT;
out:
ptlrpc_req_finished(request);
* If the create succeeds, we fill in the inode information
* with d_instantiate().
*/
-static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode, struct lookup_intent *it)
+static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
+ struct lookup_intent *it)
{
struct inode *inode;
- struct ptlrpc_request *request = it->it_data;
int rc = 0;
ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p),intent=%s\n",
- dentry->d_name.name, dir->i_ino, dir->i_generation, dir,
- LL_IT2STR(it));
+ CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
+ dentry->d_name.len, dentry->d_name.name, dir->i_ino,
+ dir->i_generation, dir, LL_IT2STR(it));
- rc = ll_it_open_error(DISP_OPEN_CREATE, it);
- if (rc) {
- ptlrpc_req_finished(request);
+ rc = it_open_error(DISP_OPEN_CREATE, it);
+ if (rc)
RETURN(rc);
- }
- mdc_store_inode_generation(request, 2, 1);
inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
NULL, 0, mode, 0, it);
if (IS_ERR(inode)) {
}
d_instantiate(dentry, inode);
+ /* Negative dentry may be unhashed if parent does not have UPDATE lock,
+ * but some callers, e.g. do_coredump, expect dentry to be hashed after
+ * successful create. Hash it here. */
+ spin_lock(&dcache_lock);
+ if (d_unhashed(dentry))
+ __d_rehash(dentry, 0);
+ spin_unlock(&dcache_lock);
RETURN(0);
}
-#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
-static int ll_create_nd(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
+static void ll_update_times(struct ptlrpc_request *request, int offset,
+ struct inode *inode)
{
- return ll_create_it(dir, dentry, mode, &nd->it);
+ struct mdt_body *body = lustre_msg_buf(request->rq_repmsg, offset,
+ sizeof(*body));
+ LASSERT(body);
+
+ /* mtime is always updated with ctime, but can be set in past.
+ As write and utime(2) may happen within 1 second, and utime's
+ mtime has a priority over write's one, so take mtime from mds
+ for the same ctimes. */
+ if (body->valid & OBD_MD_FLCTIME &&
+ body->ctime >= LTIME_S(inode->i_ctime)) {
+ LTIME_S(inode->i_ctime) = body->ctime;
+
+ if (body->valid & OBD_MD_FLMTIME) {
+ CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
+ "to "LPU64"\n", inode->i_ino,
+ LTIME_S(inode->i_mtime), body->mtime);
+ LTIME_S(inode->i_mtime) = body->mtime;
+ }
+ }
}
-#endif
-static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
+static int ll_new_node(struct inode *dir, struct qstr *name,
+ const char *tgt, int mode, int rdev,
+ struct dentry *dchild, __u32 opc)
{
- struct inode *dir = nd->dentry->d_inode;
- const char *name = nd->last.name;
- int len = nd->last.len;
struct ptlrpc_request *request = NULL;
- time_t time = LTIME_S(CURRENT_TIME);
+ struct md_op_data *op_data;
+ struct inode *inode = NULL;
struct ll_sb_info *sbi = ll_i2sbi(dir);
- struct mdc_op_data op_data;
- int err = -EMLINK;
+ int tgt_len = 0;
+ int err;
+
ENTRY;
+ if (unlikely(tgt != NULL))
+ tgt_len = strlen(tgt) + 1;
+
+ op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
+ name->len, 0, opc, NULL);
+ if (IS_ERR(op_data))
+ GOTO(err_exit, err = PTR_ERR(op_data));
+
+ err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
+ current->fsuid, current->fsgid,
+ current->cap_effective, rdev, &request);
+ ll_finish_md_op_data(op_data);
+ if (err)
+ GOTO(err_exit, err);
+
+ ll_update_times(request, REPLY_REC_OFF, dir);
+
+ if (dchild) {
+ err = ll_prep_inode(&inode, request, REPLY_REC_OFF,
+ dchild->d_sb);
+ if (err)
+ GOTO(err_exit, err);
+
+ d_drop(dchild);
+ d_instantiate(dchild, inode);
+ EXIT;
+ }
+err_exit:
+ ptlrpc_req_finished(request);
- CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p)\n",
- name, dir->i_ino, dir->i_generation, dir);
+ return err;
+}
- if (dir->i_nlink >= EXT2_LINK_MAX)
- RETURN(err);
+static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
+ unsigned rdev, struct dentry *dchild)
+{
+ int err;
+ ENTRY;
+
+ CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
+ name->len, name->name, dir->i_ino, dir->i_generation, dir,
+ mode, rdev);
mode &= ~current->fs->umask;
switch (mode & S_IFMT) {
- case 0:
- case S_IFREG:
+ case 0:
mode |= S_IFREG; /* for mode = 0 case, fallthrough */
- case S_IFCHR:
+ case S_IFREG:
+ case S_IFCHR:
case S_IFBLK:
- case S_IFIFO:
+ case S_IFIFO:
case S_IFSOCK:
- ll_prepare_mdc_op_data(&op_data, dir, NULL, name, len, 0);
- err = mdc_create(&sbi->ll_mdc_conn, &op_data, NULL, 0, mode,
- current->fsuid, current->fsgid, time,
- rdev, &request);
- ptlrpc_req_finished(request);
+ err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
+ LUSTRE_OPC_MKNOD);
break;
case S_IFDIR:
err = -EPERM;
RETURN(err);
}
-static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
+#ifndef LUSTRE_KERNEL_VERSION
+static int ll_create_nd(struct inode *dir, struct dentry *dentry,
+ int mode, struct nameidata *nd)
{
- struct inode *dir = nd->dentry->d_inode;
- const char *name = nd->last.name;
- int len = nd->last.len;
- struct ptlrpc_request *request = NULL;
- time_t time = LTIME_S(CURRENT_TIME);
- struct ll_sb_info *sbi = ll_i2sbi(dir);
- struct mdc_op_data op_data;
- int err = -EMLINK;
- ENTRY;
+ struct lookup_intent *it = ll_d2d(dentry)->lld_it;
+ int rc;
- CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p),target=%s\n",
- name, dir->i_ino, dir->i_generation, dir, tgt);
+ if (!it)
+ return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
- if (dir->i_nlink >= EXT2_LINK_MAX)
- RETURN(err);
+ ll_d2d(dentry)->lld_it = NULL;
- ll_prepare_mdc_op_data(&op_data, dir, NULL, name, len, 0);
- err = mdc_create(&sbi->ll_mdc_conn, &op_data,
- tgt, strlen(tgt) + 1, S_IFLNK | S_IRWXUGO,
- current->fsuid, current->fsgid, time, 0, &request);
- ptlrpc_req_finished(request);
+ /* Was there an error? Propagate it! */
+ if (it->d.lustre.it_status) {
+ rc = it->d.lustre.it_status;
+ goto out;
+ }
+
+ rc = ll_create_it(dir, dentry, mode, it);
+#ifdef HAVE_FILE_IN_STRUCT_INTENT
+ if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
+ nd->intent.open.file->private_data = it;
+ lookup_instantiate_filp(nd, dentry, NULL);
+ }
+#else
+ ll_release_openhandle(dentry,it);
+#endif
+
+out:
+ ll_intent_release(it);
+ OBD_FREE(it, sizeof(*it));
+
+ return rc;
+}
+#else
+static int ll_create_nd(struct inode *dir, struct dentry *dentry,
+ int mode, struct nameidata *nd)
+{
+ if (!nd || !nd->intent.d.lustre.it_disposition)
+ /* No saved request? Just mknod the file */
+ return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
+
+ return ll_create_it(dir, dentry, mode, &nd->intent);
+}
+#endif
+
+static int ll_symlink_generic(struct inode *dir, struct qstr *name,
+ const char *tgt, struct dentry *dchild)
+{
+ int err;
+ ENTRY;
+
+ CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
+ name->len, name->name, dir->i_ino, dir->i_generation,
+ dir, 3000, tgt);
+
+ err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
+ 0, dchild, LUSTRE_OPC_SYMLINK);
RETURN(err);
}
-static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
+static int ll_link_generic(struct inode *src, struct inode *dir,
+ struct qstr *name, struct dentry *dchild)
{
- struct inode *src = srcnd->dentry->d_inode;
- struct inode *dir = tgtnd->dentry->d_inode;
- const char *name = tgtnd->last.name;
- int len = tgtnd->last.len;
+ struct ll_sb_info *sbi = ll_i2sbi(dir);
struct ptlrpc_request *request = NULL;
- struct mdc_op_data op_data;
+ struct md_op_data *op_data;
int err;
- struct ll_sb_info *sbi = ll_i2sbi(dir);
ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),dir=%lu/%u(%p),target=%s\n",
- src->i_ino, src->i_generation, src,
- dir->i_ino, dir->i_generation, dir, name);
-
- ll_prepare_mdc_op_data(&op_data, src, dir, name, len, 0);
- err = mdc_link(&sbi->ll_mdc_conn, &op_data, &request);
+ CDEBUG(D_VFSTRACE,
+ "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
+ src->i_ino, src->i_generation, src, dir->i_ino,
+ dir->i_generation, dir, name->len, name->name);
+
+ op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
+ 0, LUSTRE_OPC_ANY, NULL);
+ if (IS_ERR(op_data))
+ RETURN(PTR_ERR(op_data));
+
+ err = md_link(sbi->ll_md_exp, op_data, &request);
+ ll_finish_md_op_data(op_data);
+ if (err)
+ GOTO(out, err);
+ if (dchild)
+ d_drop(dchild);
+
+ ll_update_times(request, REPLY_REC_OFF, dir);
+ EXIT;
+out:
ptlrpc_req_finished(request);
-
RETURN(err);
}
+static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
+ int mode, struct dentry *dchild)
-static int ll_mkdir_raw(struct nameidata *nd, int mode)
{
- struct inode *dir = nd->dentry->d_inode;
- const char *name = nd->last.name;
- int len = nd->last.len;
- struct ptlrpc_request *request = NULL;
- time_t time = LTIME_S(CURRENT_TIME);
- struct ll_sb_info *sbi = ll_i2sbi(dir);
- struct mdc_op_data op_data;
- int err = -EMLINK;
+ int err;
ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p)\n",
- name, dir->i_ino, dir->i_generation, dir);
- if (dir->i_nlink >= EXT2_LINK_MAX)
- RETURN(err);
+ CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
+ name->len, name->name, dir->i_ino, dir->i_generation, dir);
mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
- ll_prepare_mdc_op_data(&op_data, dir, NULL, name, len, 0);
- err = mdc_create(&sbi->ll_mdc_conn, &op_data, NULL, 0, mode,
- current->fsuid, current->fsgid, time, 0, &request);
- ptlrpc_req_finished(request);
+ err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
+
RETURN(err);
}
-static int ll_rmdir_raw(struct nameidata *nd)
+/* Try to find the child dentry by its name.
+ If found, put the result fid into @fid. */
+static void ll_get_child_fid(struct inode * dir, struct qstr *name,
+ struct lu_fid *fid)
+{
+ struct dentry *parent, *child;
+
+ parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
+ child = d_lookup(parent, name);
+ if (child) {
+ if (child->d_inode)
+ *fid = *ll_inode2fid(child->d_inode);
+ dput(child);
+ }
+}
+
+static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
+ struct dentry *dchild, struct qstr *name)
{
- struct inode *dir = nd->dentry->d_inode;
- const char *name = nd->last.name;
- int len = nd->last.len;
+ struct ptlrpc_request *request = NULL;
+ struct md_op_data *op_data;
int rc;
ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p)\n",
- name, dir->i_ino, dir->i_generation, dir);
-
- rc = ll_mdc_unlink(dir, NULL, S_IFDIR, name, len);
+
+ CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
+ name->len, name->name, dir->i_ino, dir->i_generation, dir);
+
+ if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
+ RETURN(-EBUSY);
+
+ op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
+ S_IFDIR, LUSTRE_OPC_ANY, NULL);
+ if (IS_ERR(op_data))
+ RETURN(PTR_ERR(op_data));
+
+ ll_get_child_fid(dir, name, &op_data->op_fid3);
+ rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
+ ll_finish_md_op_data(op_data);
+ if (rc == 0)
+ ll_update_times(request, REPLY_REC_OFF, dir);
+ ptlrpc_req_finished(request);
RETURN(rc);
}
-static int ll_unlink_raw(struct nameidata *nd)
+int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
{
- struct inode *dir = nd->dentry->d_inode;
- const char *name = nd->last.name;
- int len = nd->last.len;
+ struct mdt_body *body;
+ struct lov_mds_md *eadata;
+ struct lov_stripe_md *lsm = NULL;
+ struct obd_trans_info oti = { 0 };
+ struct obdo *oa;
int rc;
ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p)\n",
- name, dir->i_ino, dir->i_generation, dir);
- rc = ll_mdc_unlink(dir, NULL, S_IFREG, name, len);
+ /* req is swabbed so this is safe */
+ body = lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
+
+ if (!(body->valid & OBD_MD_FLEASIZE))
+ RETURN(0);
+
+ if (body->eadatasize == 0) {
+ CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
+ GOTO(out, rc = -EPROTO);
+ }
+
+ /* The MDS sent back the EA because we unlinked the last reference
+ * to this file. Use this EA to unlink the objects on the OST.
+ * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
+ * check it is complete and sensible. */
+ eadata = lustre_swab_repbuf(request, REPLY_REC_OFF + 1,
+ body->eadatasize, NULL);
+ LASSERT(eadata != NULL);
+ if (eadata == NULL) {
+ CERROR("Can't unpack MDS EA data\n");
+ GOTO(out, rc = -EPROTO);
+ }
+
+ rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
+ if (rc < 0) {
+ CERROR("obd_unpackmd: %d\n", rc);
+ GOTO(out, rc);
+ }
+ LASSERT(rc >= sizeof(*lsm));
+
+ rc = obd_checkmd(ll_i2dtexp(dir), ll_i2mdexp(dir), lsm);
+ if (rc)
+ GOTO(out_free_memmd, rc);
+
+ OBDO_ALLOC(oa);
+ if (oa == NULL)
+ GOTO(out_free_memmd, rc = -ENOMEM);
+
+ oa->o_id = lsm->lsm_object_id;
+ oa->o_gr = lsm->lsm_object_gr;
+ oa->o_mode = body->mode & S_IFMT;
+ oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
+
+ if (body->valid & OBD_MD_FLCOOKIE) {
+ oa->o_valid |= OBD_MD_FLCOOKIE;
+ oti.oti_logcookies =
+ lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF + 2,
+ sizeof(struct llog_cookie) *
+ lsm->lsm_stripe_count);
+ if (oti.oti_logcookies == NULL) {
+ oa->o_valid &= ~OBD_MD_FLCOOKIE;
+ body->valid &= ~OBD_MD_FLCOOKIE;
+ }
+ }
+
+ rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir));
+ OBDO_FREE(oa);
+ if (rc)
+ CERROR("obd destroy objid "LPX64" error %d\n",
+ lsm->lsm_object_id, rc);
+ out_free_memmd:
+ obd_free_memmd(ll_i2dtexp(dir), &lsm);
+ out:
+ return rc;
+}
+
+static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
+ struct dentry *dchild, struct qstr *name)
+{
+ struct ptlrpc_request *request = NULL;
+ struct md_op_data *op_data;
+ int rc;
+ ENTRY;
+ CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
+ name->len, name->name, dir->i_ino, dir->i_generation, dir);
+
+ /*
+ * XXX: unlink bind mountpoint maybe call to here,
+ * just check it as vfs_unlink does.
+ */
+ if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
+ RETURN(-EBUSY);
+
+ op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
+ name->len, 0, LUSTRE_OPC_ANY, NULL);
+ if (IS_ERR(op_data))
+ RETURN(PTR_ERR(op_data));
+
+ ll_get_child_fid(dir, name, &op_data->op_fid3);
+ rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
+ ll_finish_md_op_data(op_data);
+
+ if (rc)
+ GOTO(out, rc);
+
+ ll_update_times(request, REPLY_REC_OFF, dir);
+
+ rc = ll_objects_destroy(request, dir);
+ out:
+ ptlrpc_req_finished(request);
RETURN(rc);
}
-static int ll_rename_raw(struct nameidata *oldnd, struct nameidata *newnd)
+static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
+ struct dentry *src_dchild, struct qstr *src_name,
+ struct inode *tgt, struct dentry *tgt_dparent,
+ struct dentry *tgt_dchild, struct qstr *tgt_name)
{
- struct inode *src = oldnd->dentry->d_inode;
- struct inode *tgt = newnd->dentry->d_inode;
- const char *oldname = oldnd->last.name;
- int oldlen = oldnd->last.len;
- const char *newname = newnd->last.name;
- int newlen = newnd->last.len;
struct ptlrpc_request *request = NULL;
struct ll_sb_info *sbi = ll_i2sbi(src);
- struct mdc_op_data op_data;
+ struct md_op_data *op_data;
int err;
ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op:oldname=%s,src_dir=%lu/%u(%p),newname=%s,"
- "tgt_dir=%lu/%u(%p)\n", oldname, src->i_ino, src->i_generation,
- src, newname, tgt->i_ino, tgt->i_generation, tgt);
+ CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
+ "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
+ src->i_ino, src->i_generation, src, tgt_name->len,
+ tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
+
+ if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
+ ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
+ RETURN(-EBUSY);
+
+ op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
+ LUSTRE_OPC_ANY, NULL);
+ if (IS_ERR(op_data))
+ RETURN(PTR_ERR(op_data));
+
+ ll_get_child_fid(src, src_name, &op_data->op_fid3);
+ ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
+ err = md_rename(sbi->ll_md_exp, op_data,
+ src_name->name, src_name->len,
+ tgt_name->name, tgt_name->len, &request);
+ ll_finish_md_op_data(op_data);
+ if (!err) {
+ ll_update_times(request, REPLY_REC_OFF, src);
+ ll_update_times(request, REPLY_REC_OFF, tgt);
+ err = ll_objects_destroy(request, src);
+ }
- ll_prepare_mdc_op_data(&op_data, src, tgt, NULL, 0, 0);
- err = mdc_rename(&sbi->ll_mdc_conn, &op_data,
- oldname, oldlen, newname, newlen, &request);
ptlrpc_req_finished(request);
RETURN(err);
}
+#ifdef LUSTRE_KERNEL_VERSION
+static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
+{
+ return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
+}
+static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
+{
+ return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
+ NULL, &srcnd->last,
+ tgtnd->dentry->d_inode, tgtnd->dentry,
+ NULL, &tgtnd->last);
+}
+static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
+{
+ return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
+ &tgtnd->last, NULL);
+}
+static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
+{
+ return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
+}
+static int ll_rmdir_raw(struct nameidata *nd)
+{
+ return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
+ &nd->last);
+}
+static int ll_mkdir_raw(struct nameidata *nd, int mode)
+{
+ return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
+}
+static int ll_unlink_raw(struct nameidata *nd)
+{
+ return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
+ &nd->last);
+}
+#endif
+
+static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
+ ll_dev_t rdev)
+{
+ return ll_mknod_generic(dir, &dchild->d_name, mode,
+ old_encode_dev(rdev), dchild);
+}
+
+static int ll_unlink(struct inode * dir, struct dentry *dentry)
+{
+ return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
+}
+static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
+{
+ return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
+}
+static int ll_rmdir(struct inode *dir, struct dentry *dentry)
+{
+ return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
+}
+static int ll_symlink(struct inode *dir, struct dentry *dentry,
+ const char *oldname)
+{
+ return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
+}
+static int ll_link(struct dentry *old_dentry, struct inode *dir,
+ struct dentry *new_dentry)
+{
+ return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
+ new_dentry);
+}
+static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
+ struct inode *new_dir, struct dentry *new_dentry)
+{
+ return ll_rename_generic(old_dir, NULL,
+ old_dentry, &old_dentry->d_name,
+ new_dir, NULL, new_dentry,
+ &new_dentry->d_name);
+}
+
struct inode_operations ll_dir_inode_operations = {
- link_raw: ll_link_raw,
- unlink_raw: ll_unlink_raw,
- symlink_raw: ll_symlink_raw,
- mkdir_raw: ll_mkdir_raw,
- rmdir_raw: ll_rmdir_raw,
- mknod_raw: ll_mknod_raw,
- rename_raw: ll_rename_raw,
- setattr: ll_setattr,
- setattr_raw: ll_setattr_raw,
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
- create_it: ll_create_it,
- lookup_it: ll_lookup_it,
- revalidate_it: ll_inode_revalidate_it,
-#else
- lookup_it: ll_lookup_nd,
- create_nd: ll_create_nd,
- getattr_it: ll_getattr,
+#ifdef LUSTRE_KERNEL_VERSION
+ .link_raw = ll_link_raw,
+ .unlink_raw = ll_unlink_raw,
+ .symlink_raw = ll_symlink_raw,
+ .mkdir_raw = ll_mkdir_raw,
+ .rmdir_raw = ll_rmdir_raw,
+ .mknod_raw = ll_mknod_raw,
+ .rename_raw = ll_rename_raw,
+ .setattr = ll_setattr,
+ .setattr_raw = ll_setattr_raw,
+#endif
+ .mknod = ll_mknod,
+ .lookup = ll_lookup_nd,
+ .create = ll_create_nd,
+ /* We need all these non-raw things for NFSD, to not patch it. */
+ .unlink = ll_unlink,
+ .mkdir = ll_mkdir,
+ .rmdir = ll_rmdir,
+ .symlink = ll_symlink,
+ .link = ll_link,
+ .rename = ll_rename,
+ .setattr = ll_setattr,
+ .getattr = ll_getattr,
+ .permission = ll_inode_permission,
+ .setxattr = ll_setxattr,
+ .getxattr = ll_getxattr,
+ .listxattr = ll_listxattr,
+ .removexattr = ll_removexattr,
+};
+
+struct inode_operations ll_special_inode_operations = {
+#ifdef LUSTRE_KERNEL_VERSION
+ .setattr_raw = ll_setattr_raw,
#endif
+ .setattr = ll_setattr,
+ .getattr = ll_getattr,
+ .permission = ll_inode_permission,
+ .setxattr = ll_setxattr,
+ .getxattr = ll_getxattr,
+ .listxattr = ll_listxattr,
+ .removexattr = ll_removexattr,
};