struct ll_dentry_data *lld;
ENTRY;
LASSERT(de != NULL);
+
+ CDEBUG(D_DENTRY, "releasing dentry %p\n", de);
+
lld = ll_d2d(de);
- LASSERT(lld != NULL);
- LASSERT(lld->lld_cwd_count == 0);
- LASSERT(lld->lld_mnt_count == 0);
- OBD_FREE(de->d_fsdata, sizeof(struct ll_dentry_data));
+ if (lld) { /* Root dentry does not have ll_dentry_data */
+ LASSERT(lld->lld_cwd_count == 0);
+ LASSERT(lld->lld_mnt_count == 0);
+ OBD_FREE(de->d_fsdata, sizeof(struct ll_dentry_data));
+ }
EXIT;
}
+/* Compare if two dentries are the same. Don't match if the existing dentry
+ * is marked DCACHE_LUSTRE_INVALID. Returns 1 if different, 0 if the same.
+ *
+ * This avoids a race where ll_lookup_it() instantiates a dentry, but we get
+ * an AST before calling d_revalidate_it(). The dentry still exists (marked
+ * INVALID) so d_lookup() matches it, but we have no lock on it (so
+ * lock_match() fails) and we spin around real_lookup(). */
+static int ll_dcompare(struct dentry *parent, struct qstr *d_name,
+ struct qstr *name){
+ struct dentry *dchild;
+ ENTRY;
+
+ if (d_name->len != name->len)
+ RETURN(1);
+
+ if (memcmp(d_name->name, name->name, name->len))
+ RETURN(1);
+
+ dchild = container_of(d_name, struct dentry, d_name); /* ugh */
+ if (dchild->d_flags & DCACHE_LUSTRE_INVALID) {
+ CDEBUG(D_DENTRY,"INVALID dentry %p not matched, was bug 3784\n",
+ dchild);
+ RETURN(1);
+ }
+
+ RETURN(0);
+}
+
/* should NOT be called with the dcache lock, see fs/dcache.c */
static int ll_ddelete(struct dentry *de)
{
ENTRY;
LASSERT(de);
CDEBUG(D_DENTRY, "%s dentry %*s (%p, parent %p, inode %p) %s%s\n",
- (de->d_flags & DCACHE_LUSTRE_INVALID ? "keeping" : "deleting"),
+ (de->d_flags & DCACHE_LUSTRE_INVALID ? "deleting" : "keeping"),
de->d_name.len, de->d_name.name, de, de->d_parent, de->d_inode,
d_unhashed(de) ? "" : "hashed,",
list_empty(&de->d_subdirs) ? "" : "subdirs");
ENTRY;
LASSERT(de != NULL);
- CDEBUG(D_DENTRY, "ldd on dentry %*s (%p) parent %p inode %p refc %d\n",
+ CDEBUG(D_DENTRY, "ldd on 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));
lock_kernel();
void ll_intent_drop_lock(struct lookup_intent *it)
{
struct lustre_handle *handle;
+ struct lustre_intent_data *itdata = LUSTRE_IT(it);
- if (it->it_op && it->d.lustre.it_lock_mode) {
- handle = (struct lustre_handle *)&it->d.lustre.it_lock_handle;
+ if (it->it_op && itdata && itdata->it_lock_mode) {
+ handle = (struct lustre_handle *)&itdata->it_lock_handle;
CDEBUG(D_DLMTRACE, "releasing lock with cookie "LPX64
" from it %p\n", handle->cookie, it);
- ldlm_lock_decref(handle, it->d.lustre.it_lock_mode);
+ ldlm_lock_decref(handle, itdata->it_lock_mode);
/* bug 494: intent_release may be called multiple times, from
* this thread and we don't want to double-decref this lock */
- it->d.lustre.it_lock_mode = 0;
+ itdata->it_lock_mode = 0;
}
}
ll_intent_drop_lock(it);
it->it_magic = 0;
it->it_op_release = 0;
- it->d.lustre.it_disposition = 0;
- it->d.lustre.it_data = NULL;
+ ll_intent_free(it);
EXIT;
}
if (inode == NULL) {
CERROR("unexpected NULL inode, tell phil\n");
+ EXIT;
return;
}
while ((tmp = tmp->next) != head) {
struct dentry *dentry = list_entry(tmp, struct dentry, d_alias);
if (atomic_read(&dentry->d_count) == 0) {
- CDEBUG(D_DENTRY, "deleting dentry %*s (%p) parent %p "
+ CDEBUG(D_DENTRY, "deleting dentry %.*s (%p) parent %p "
"inode %p\n", dentry->d_name.len,
dentry->d_name.name, dentry, dentry->d_parent,
dentry->d_inode);
dput(dentry);
goto restart;
} else if (!(dentry->d_flags & DCACHE_LUSTRE_INVALID)) {
- CDEBUG(D_DENTRY, "unhashing dentry %*s (%p) parent %p "
+ CDEBUG(D_DENTRY, "unhashing dentry %.*s (%p) parent %p "
"inode %p refc %d\n", dentry->d_name.len,
dentry->d_name.name, dentry, dentry->d_parent,
dentry->d_inode, atomic_read(&dentry->d_count));
extern struct dentry *ll_find_alias(struct inode *, struct dentry *);
-static int revalidate_it_finish(struct ptlrpc_request *request,
- int offset, struct lookup_intent *it,
- struct dentry *de)
+int revalidate_it_finish(struct ptlrpc_request *request, int offset,
+ struct lookup_intent *it, struct dentry *de)
{
struct ll_sb_info *sbi;
int rc = 0;
RETURN(-ENOENT);
sbi = ll_i2sbi(de->d_inode);
- rc = ll_prep_inode(sbi->ll_osc_exp, sbi->ll_mdc_exp,
+ rc = ll_prep_inode(sbi->ll_dt_exp, sbi->ll_md_exp,
&de->d_inode, request, offset, NULL);
RETURN(rc);
LASSERT(it != NULL);
LASSERT(dentry != NULL);
- if (it->d.lustre.it_lock_mode && dentry->d_inode != NULL) {
+ if (LUSTRE_IT(it)->it_lock_mode && dentry->d_inode != NULL) {
struct inode *inode = dentry->d_inode;
CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
inode, inode->i_ino, inode->i_generation);
- mdc_set_lock_data(NULL, &it->d.lustre.it_lock_handle, inode);
+ mdc_set_lock_data(NULL, &LUSTRE_IT(it)->it_lock_handle, inode);
}
/* drop lookup or getattr locks immediately */
- if (it->it_op == IT_LOOKUP || it->it_op == IT_GETATTR) {
+ if (it->it_op == IT_LOOKUP || it->it_op == IT_GETATTR ||
+ it->it_op == IT_CHDIR) {
#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
- /* on 2.6 there are situation when several lookups and
+ /*
+ * on 2.6 there are situations when several lookups and
* revalidations may be requested during single operation.
- * therefore, we don't release intent here -bzzz */
+ * Therefore, we don't release intent here -bzzz
+ */
ll_intent_drop_lock(it);
#else
ll_intent_release(it);
void ll_frob_intent(struct lookup_intent **itp, struct lookup_intent *deft)
{
struct lookup_intent *it = *itp;
+
#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
- if (it && it->it_magic != INTENT_MAGIC) {
- CERROR("WARNING: uninitialized intent\n");
- LBUG();
+ if (it) {
+ LASSERTF(it->it_magic == INTENT_MAGIC, "bad intent magic: %x\n",
+ it->it_magic);
}
- if (it && (it->it_op == IT_GETATTR || it->it_op == 0))
- it->it_op = IT_LOOKUP;
#endif
if (!it || it->it_op == IT_GETXATTR)
it = *itp = deft;
+ if (it->d.fs_data)
+ return;
+
+ if (ll_intent_alloc(it)) {
+ CERROR("Failed to allocate memory for lustre specific intent "
+ "data\n");
+ /* XXX: we cannot return status just yet */
+ LBUG();
+ }
+}
+
+int ll_intent_alloc(struct lookup_intent *it)
+{
+ if (it->d.fs_data) {
+ CERROR("Intent alloc on already allocated intent\n");
+ return 0;
+ }
+ OBD_SLAB_ALLOC(it->d.fs_data, ll_intent_slab, SLAB_KERNEL,
+ sizeof(struct lustre_intent_data));
+ if (!it->d.fs_data) {
+ CERROR("Failed to allocate memory for lustre specific intent "
+ "data\n");
+ return -ENOMEM;
+ }
+
it->it_op_release = ll_intent_release;
+ return 0;
+}
+
+void ll_intent_free(struct lookup_intent *it)
+{
+ if (it->d.fs_data) {
+ OBD_SLAB_FREE(it->d.fs_data, ll_intent_slab,
+ sizeof(struct lustre_intent_data));
+ it->d.fs_data = NULL;
+ }
+}
+
+static inline int
+ll_special_name(struct dentry *de)
+{
+ if (de->d_name.name[0] == '.') switch (de->d_name.len) {
+ case 2:
+ if (de->d_name.name[1] == '.')
+ return 1;
+ case 1:
+ return 1;
+ default:
+ return 0;
+ }
+ return 0;
}
int ll_revalidate_it(struct dentry *de, int flags, struct nameidata *nd,
struct lookup_intent *it)
{
- int rc;
- struct ll_fid pfid, cfid;
- struct it_cb_data icbd;
- struct ptlrpc_request *req = NULL;
struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
+ struct ptlrpc_request *req = NULL;
+ int gns_it, gns_flags, rc = 0;
struct obd_export *exp;
-
+ struct it_cb_data icbd;
+ struct lustre_id pid;
+ struct lustre_id cid;
ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op:name=%s,intent=%s\n", de->d_name.name,
- LL_IT2STR(it));
+
+ CDEBUG(D_VFSTRACE, "VFS Op:name=%s (%p), intent=%s\n", de->d_name.name,
+ de, LL_IT2STR(it));
/* Cached negative dentries are unsafe for now - look them up again */
if (de->d_inode == NULL)
RETURN(0);
+ /*
+ * root of the tree is always valid, attributes would be fixed in
+ * ll_inode_revalidate_it()
+ */
+ if (de->d_sb->s_root == de)
+ RETURN(1);
+
CDEBUG(D_INODE, "revalidate 0x%p: %*s -> %lu/%lu\n",
- de, de->d_name.len, de->d_name.name,
- (unsigned long) de->d_inode->i_ino,
- (unsigned long) de->d_inode->i_generation);
+ de, de->d_name.len, de->d_name.name,
+ (unsigned long) de->d_inode->i_ino,
+ (unsigned long) de->d_inode->i_generation);
+
+ exp = ll_i2mdexp(de->d_inode);
+ ll_inode2id(&pid, de->d_parent->d_inode);
+ ll_inode2id(&cid, de->d_inode);
+ LASSERT(id_fid(&cid) != 0);
- exp = ll_i2mdcexp(de->d_inode);
- ll_inode2fid(&pfid, de->d_parent->d_inode);
- ll_inode2fid(&cfid, de->d_inode);
icbd.icbd_parent = de->d_parent->d_inode;
icbd.icbd_childp = &de;
- /* Never execute intents for mount points.
- * Attributes will be fixed up in ll_inode_revalidate_it */
+ /*
+ * never execute intents for mount points. Attributes will be fixed up
+ * in ll_inode_revalidate_it().
+ */
if (d_mountpoint(de))
RETURN(1);
if (nd != NULL)
nd->mnt->mnt_last_used = jiffies;
- ll_frob_intent(&it, &lookup_it);
- LASSERT(it);
-
- if (it->it_op == IT_GETATTR) { /* We need to check for LOOKUP lock
- as well */
- rc = md_intent_lock(exp, &pfid, de->d_name.name,
- de->d_name.len, NULL, 0, &cfid, &lookup_it,
- flags, &req, ll_mdc_blocking_ast);
- /* If there was no lookup lock, no point in even checking for
- UPDATE lock */
- if (!rc) {
- it = &lookup_it;
- GOTO(out, rc);
- }
- if (it_disposition(&lookup_it, DISP_LOOKUP_NEG)) {
- ll_intent_release(&lookup_it);
- it = &lookup_it;
- GOTO(out, rc = 0);
+ OBD_FAIL_TIMEOUT(OBD_FAIL_MDC_REVALIDATE_PAUSE, 5);
+ gns_it = nd ? nd->intent.open.it_op : IT_OPEN;
+ gns_flags = nd ? nd->flags : LOOKUP_CONTINUE;
+
+ if (it && it->it_op == IT_GETATTR)
+ it = NULL; /* will use it_lookup */
+ else if (it && (it->it_op == IT_OPEN) && de->d_inode) {
+ /* open lock stuff */
+ struct inode *inode = de->d_inode;
+ struct ll_inode_info *lli = ll_i2info(inode);
+ struct obd_client_handle **och_p;
+ __u64 *och_usecount;
+ struct obd_device *obddev;
+ struct lustre_handle lockh;
+ int flags = LDLM_FL_BLOCK_GRANTED;
+ ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_OPEN}};
+ struct ldlm_res_id file_res_id = {.name = {id_fid(&lli->lli_id),
+ id_group(&lli->lli_id)}};
+ int lockmode;
+
+ if (it->it_flags & FMODE_WRITE) {
+ och_p = &lli->lli_mds_write_och;
+ och_usecount = &lli->lli_open_fd_write_count;
+ lockmode = LCK_CW;
+ } else if (it->it_flags & FMODE_EXEC) {
+ och_p = &lli->lli_mds_exec_och;
+ och_usecount = &lli->lli_open_fd_exec_count;
+ lockmode = LCK_PR;
+ } else {
+ och_p = &lli->lli_mds_read_och;
+ och_usecount = &lli->lli_open_fd_read_count;
+ lockmode = LCK_CR;
}
- if (req)
- ptlrpc_req_finished(req);
- req = NULL;
- ll_lookup_finish_locks(&lookup_it, de);
+ /* Check for the proper lock */
+ obddev = md_get_real_obd(exp, &lli->lli_id);
+ if (!ldlm_lock_match(obddev->obd_namespace, flags, &file_res_id,
+ LDLM_IBITS, &policy, lockmode, &lockh))
+ goto do_lock;
+ down(&lli->lli_och_sem);
+ if (*och_p) { /* Everything is open already, do nothing */
+ /*(*och_usecount)++; Do not let them steal our open
+ handle from under us */
+ /* XXX The code above was my original idea, but in case
+ we have the handle, but we cannot use it due to later
+ checks (e.g. O_CREAT|O_EXCL flags set), nobody
+ would decrement counter increased here. So we just
+ hope the lock won't be invalidated in between. But
+ if it would be, we'll reopen the open request to
+ MDS later during file open path */
+ up(&lli->lli_och_sem);
+ if (ll_intent_alloc(it))
+ LBUG();
+ memcpy(&LUSTRE_IT(it)->it_lock_handle, &lockh,
+ sizeof(lockh));
+ LUSTRE_IT(it)->it_lock_mode = lockmode;
+
+ /*
+ * we do not check here for possible GNS dentry as if
+ * file is opened on it, it is mounted already and we do
+ * not need do anything. --umka
+ */
+ RETURN(1);
+ } else {
+ /* Hm, interesting. Lock is present, but no open
+ handle? */
+ up(&lli->lli_och_sem);
+ ldlm_lock_decref(&lockh, lockmode);
+ }
}
- rc = md_intent_lock(exp, &pfid, de->d_name.name, de->d_name.len,
- NULL, 0, &cfid, it, flags, &req,
- ll_mdc_blocking_ast);
- /* If req is NULL, then mdc_intent_lock only tried to do a lock match;
+do_lock:
+ ll_frob_intent(&it, &lookup_it);
+ LASSERT(it != NULL);
+
+ rc = md_intent_lock(exp, &pid, (char *)de->d_name.name, de->d_name.len,
+ NULL, 0, &cid, it, flags, &req, ll_mdc_blocking_ast);
+ /* If req is NULL, then md_intent_lock() only tried to do a lock match;
* if all was well, it will return 1 if it found locks, 0 otherwise. */
- if (req == NULL && rc >= 0)
+ if (req == NULL && rc >= 0) {
+ if (!rc)
+ goto do_lookup;
GOTO(out, rc);
+ }
if (rc < 0) {
if (rc != -ESTALE) {
- CDEBUG(D_INFO, "ll_intent_lock: rc %d : it->it_status "
- "%d\n", rc, it->d.lustre.it_status);
+ CDEBUG(D_INFO, "ll_intent_lock(): rc %d : it->it_status "
+ "%d\n", rc, LUSTRE_IT(it)->it_status);
}
GOTO(out, rc = 0);
}
-
+revalidate_finish:
rc = revalidate_it_finish(req, 1, it, de);
if (rc != 0) {
ll_intent_release(it);
dentry */
spin_lock(&dcache_lock);
hlist_del_init(&de->d_hash);
- __d_rehash(de, 0);
+ __d_rehash(de);
spin_unlock(&dcache_lock);
GOTO(out, rc);
- out:
- if (req != NULL && rc == 1)
+out:
+ /* If we had succesful it lookup on mds, but it happened to be negative,
+ we do not free request as it will be reused during lookup (see
+ comment in mdc/mdc_locks.c::mdc_intent_lock(). But if
+ request was not completed, we need to free it. (bug 5154) */
+ if (req != NULL && (rc == 1 || !it_disposition(it, DISP_ENQ_COMPLETE))) {
ptlrpc_req_finished(req);
+ req = NULL;
+ }
+
if (rc == 0) {
+ if (it == &lookup_it)
+ ll_intent_release(it);
+
ll_unhash_aliases(de->d_inode);
- return rc;
+ RETURN(0);
+ }
+
+ /*
+ * if we found that this is possible GNS mount and dentry is still valid
+ * and may be used by system, we drop the lock and return 0, that means
+ * that re-lookup is needed. Such a way we cause real mounting only in
+ * lookup control path, which is always made with parent's i_sem taken.
+ * --umka
+ */
+ if (nd && atomic_read(&ll_i2sbi(de->d_inode)->ll_gns_enabled) &&
+ (de->d_inode->i_mode & S_ISUID) && S_ISDIR(de->d_inode->i_mode) &&
+ (gns_flags & LOOKUP_CONTINUE || (gns_it & (IT_CHDIR | IT_OPEN)))) {
+ /*
+ * special "." and ".." has to be always revalidated because
+ * they never should be passed to lookup()
+ */
+ if (!ll_special_name(de)) {
+ CDEBUG(D_DENTRY, "possible GNS dentry %*s %p found, "
+ "causing mounting\n", (int)de->d_name.len,
+ de->d_name.name, de);
+
+ LASSERT(req == NULL);
+ if (it == &lookup_it) {
+ ll_intent_release(it);
+ } else {
+ ll_intent_drop_lock(it);
+ }
+ ll_unhash_aliases(de->d_inode);
+ RETURN(0);
+ }
}
CDEBUG(D_DENTRY, "revalidated 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));
- ll_lookup_finish_locks(it, de);
- de->d_flags &= ~DCACHE_LUSTRE_INVALID;
- if (!((de->d_inode->i_mode & S_ISUID) &&S_ISDIR(de->d_inode->i_mode)) ||
- !(flags & LOOKUP_CONTINUE || (it->it_op & (IT_CHDIR | IT_OPEN))))
- return rc;
- if (nd)
- (void)ll_dir_process_mount_object(de, nd->mnt);
+ if (it == &lookup_it)
+ ll_intent_release(it);
+ else
+ ll_lookup_finish_locks(it, de);
+
+ de->d_flags &= ~DCACHE_LUSTRE_INVALID;
return rc;
+
+do_lookup:
+ if (it != &lookup_it) {
+ ll_intent_release(it);
+ it = &lookup_it;
+ if (ll_intent_alloc(it))
+ LBUG();
+ }
+
+ rc = md_intent_lock(exp, &pid, (char *)de->d_name.name, de->d_name.len,
+ NULL, 0, NULL, it, 0, &req, ll_mdc_blocking_ast);
+ if (rc >= 0) {
+ struct mds_body *mds_body = lustre_msg_buf(req->rq_repmsg, 1,
+ sizeof(*mds_body));
+
+ /* see if we got same inode, if not - return error */
+ if (id_equal_stc(&cid, &mds_body->id1))
+ goto revalidate_finish;
+ }
+
+ GOTO(out, rc = 0);
}
/*static*/ void ll_pin(struct dentry *de, struct vfsmount *mnt, int flag)
unlock_kernel();
handle = (flag) ? &ldd->lld_mnt_och : &ldd->lld_cwd_och;
- rc = obd_pin(sbi->ll_mdc_exp, inode->i_ino, inode->i_generation,
+ rc = obd_pin(sbi->ll_md_exp, inode->i_ino, inode->i_generation,
inode->i_mode & S_IFMT, handle, flag);
if (rc) {
return;
}
- rc = obd_unpin(sbi->ll_mdc_exp, &handle, flag);
+ rc = obd_unpin(sbi->ll_md_exp, &handle, flag);
EXIT;
- return;
}
#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
ENTRY;
if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
- rc = ll_revalidate_it(dentry, nd->flags, nd, &nd->intent);
+ rc = ll_revalidate_it(dentry, nd->flags, nd, &nd->intent.open);
else
rc = ll_revalidate_it(dentry, 0, nd, NULL);
static void ll_dentry_iput(struct dentry *dentry, struct inode *inode)
{
struct ll_sb_info *sbi = ll_i2sbi(inode);
- struct ll_fid parent, child;
+ struct lustre_id parent, child;
LASSERT(dentry->d_parent && dentry->d_parent->d_inode);
- ll_inode2fid(&parent, dentry->d_parent->d_inode);
- ll_inode2fid(&child, inode);
- md_change_cbdata_name(sbi->ll_mdc_exp, &parent,
+ ll_inode2id(&parent, dentry->d_parent->d_inode);
+ ll_inode2id(&child, inode);
+ md_change_cbdata_name(sbi->ll_md_exp, &parent,
(char *)dentry->d_name.name,
dentry->d_name.len, &child,
null_if_equal, inode);
iput(inode);
}
#else
-#error "implement ->d_iput() for 2.6"
+static void ll_dentry_iput(struct dentry *dentry, struct inode *inode)
+{
+ struct ll_sb_info *sbi = ll_i2sbi(inode);
+ struct lustre_id parent, child;
+
+ if (dentry->d_parent != dentry) {
+ /* Do not do this for root of the tree */
+ LASSERT(dentry->d_parent && dentry->d_parent->d_inode);
+ ll_inode2id(&parent, dentry->d_parent->d_inode);
+ ll_inode2id(&child, inode);
+ md_change_cbdata_name(sbi->ll_md_exp, &parent,
+ (char *)dentry->d_name.name,
+ dentry->d_name.len, &child,
+ null_if_equal, inode);
+ }
+ iput(inode);
+
+}
#endif
struct dentry_operations ll_d_ops = {
.d_revalidate_it = ll_revalidate_it,
#endif
.d_release = ll_release,
- .d_iput = ll_dentry_iput,
+ /*.d_iput = ll_dentry_iput,*/
.d_delete = ll_ddelete,
+ .d_compare = ll_dcompare,
#if 0
.d_pin = ll_pin,
.d_unpin = ll_unpin,
git://git.whamcloud.com / fs / lustre-release.git / blobdiff