Lustre fixes

index a21b426..e52e33a 100644
@@ -46,16 +46,31 @@
 int
 libcfs_sock_ioctl(int cmd, unsigned long arg)
 {
+	mm_segment_t	oldmm = get_fs();
 	struct socket  *sock;
-	int	     rc;
+	int		rc;
+	struct file    *sock_filp;
 
 	rc = sock_create (PF_INET, SOCK_STREAM, 0, &sock);
 	if (rc != 0) {
 		CERROR ("Can't create socket: %d\n", rc);
 		return rc;
 	}
-	rc = kernel_sock_ioctl(sock, cmd, arg);
-	sock_release(sock);
+
+	sock_filp = sock_alloc_file(sock, 0, NULL);
+	if (IS_ERR(sock_filp)) {
+		sock_release(sock);
+		rc = PTR_ERR(sock_filp);
+		goto out;
+	}
+
+	set_fs(KERNEL_DS);
+	if (sock_filp->f_op->unlocked_ioctl)
+		rc = sock_filp->f_op->unlocked_ioctl(sock_filp, cmd, arg);
+	set_fs(oldmm);
+
+	fput(sock_filp);
+out:
 	return rc;
 }
 

index 383601c..ef44e09 100644
@@ -1687,12 +1687,24 @@ int gss_enlarge_reqbuf_intg(struct ptlrpc_sec *sec,
 		if (newbuf == NULL)
 			return -ENOMEM;
 
+		/* Must lock this, so that otherwise unprotected change of
+		 * rq_reqmsg is not racing with parallel processing of
+		 * imp_replay_list traversing threads. See LU-3333
+		 * This is a bandaid at best, we really need to deal with this
+		 * in request enlarging code before unpacking that's already
+		 * there */
+		if (req->rq_import)
+			spin_lock(&req->rq_import->imp_lock);
+
 		memcpy(newbuf, req->rq_reqbuf, req->rq_reqbuf_len);
 
 		OBD_FREE_LARGE(req->rq_reqbuf, req->rq_reqbuf_len);
 		req->rq_reqbuf = newbuf;
 		req->rq_reqbuf_len = newbuf_size;
 		req->rq_reqmsg = lustre_msg_buf(req->rq_reqbuf, 1, 0);
+
+		if (req->rq_import)
+			spin_unlock(&req->rq_import->imp_lock);
 	}
 
 	/* do enlargement, from wrapper to embedded, from end to begin */

@@ -1753,6 +1765,8 @@ int gss_enlarge_reqbuf_priv(struct ptlrpc_sec *sec,
 		if (newclrbuf_size + newcipbuf_size <= req->rq_reqbuf_len) {
 			void *src, *dst;
 
+			if (req->rq_import)
+				spin_lock(&req->rq_import->imp_lock);
 			/* move clear text backward. */
 			src = req->rq_clrbuf;
 			dst = (char *) req->rq_reqbuf + newcipbuf_size;

@@ -1762,6 +1776,9 @@ int gss_enlarge_reqbuf_priv(struct ptlrpc_sec *sec,
 			req->rq_clrbuf = (struct lustre_msg *) dst;
 			req->rq_clrbuf_len = newclrbuf_size;
 			req->rq_reqmsg = lustre_msg_buf(req->rq_clrbuf, 0, 0);
+
+			if (req->rq_import)
+				spin_unlock(&req->rq_import->imp_lock);
 		} else {
 			/* sadly we have to split out the clear buffer */
 			LASSERT(req->rq_reqbuf_len >= newcipbuf_size);

@@ -1776,6 +1793,15 @@ int gss_enlarge_reqbuf_priv(struct ptlrpc_sec *sec,
 		if (newclrbuf == NULL)
 			return -ENOMEM;
 
+		/* Must lock this, so that otherwise unprotected change of
+		 * rq_reqmsg is not racing with parallel processing of
+		 * imp_replay_list traversing threads. See LU-3333
+		 * This is a bandaid at best, we really need to deal with this
+		 * in request enlarging code before unpacking that's already
+		 * there */
+		if (req->rq_import)
+			spin_lock(&req->rq_import->imp_lock);
+
 		memcpy(newclrbuf, req->rq_clrbuf, req->rq_clrbuf_len);
 
 		if (req->rq_reqbuf == NULL ||

@@ -1788,6 +1814,9 @@ int gss_enlarge_reqbuf_priv(struct ptlrpc_sec *sec,
 		req->rq_clrbuf = newclrbuf;
 		req->rq_clrbuf_len = newclrbuf_size;
 		req->rq_reqmsg = lustre_msg_buf(req->rq_clrbuf, 0, 0);
+
+		if (req->rq_import)
+			spin_unlock(&req->rq_import->imp_lock);
 	}
 
 	_sptlrpc_enlarge_msg_inplace(req->rq_clrbuf, 0, newmsg_size);

index ff1137f..ac967cb 100644
@@ -260,11 +260,22 @@ int null_enlarge_reqbuf(struct ptlrpc_sec *sec,
 		if (newbuf == NULL)
 			return -ENOMEM;
 
+		/* Must lock this, so that otherwise unprotected change of
+		 * rq_reqmsg is not racing with parallel processing of
+		 * imp_replay_list traversing threads. See LU-3333
+		 * This is a bandaid at best, we really need to deal with this
+		 * in request enlarging code before unpacking that's already
+		 * there */
+		if (req->rq_import)
+			spin_lock(&req->rq_import->imp_lock);
 		memcpy(newbuf, req->rq_reqbuf, req->rq_reqlen);
 
 		OBD_FREE_LARGE(req->rq_reqbuf, req->rq_reqbuf_len);
 		req->rq_reqbuf = req->rq_reqmsg = newbuf;
 		req->rq_reqbuf_len = alloc_size;
+
+		if (req->rq_import)
+			spin_unlock(&req->rq_import->imp_lock);
 	}
 
 	_sptlrpc_enlarge_msg_inplace(req->rq_reqmsg, segment, newsize);

index 416401b..12c6cef 100644
@@ -669,6 +669,15 @@ int plain_enlarge_reqbuf(struct ptlrpc_sec *sec,
 		if (newbuf == NULL)
 			return -ENOMEM;
 
+		/* Must lock this, so that otherwise unprotected change of
+		 * rq_reqmsg is not racing with parallel processing of
+		 * imp_replay_list traversing threads. See LU-3333
+		 * This is a bandaid at best, we really need to deal with this
+		 * in request enlarging code before unpacking that's already
+		 * there */
+		if (req->rq_import)
+			spin_lock(&req->rq_import->imp_lock);
+
 		memcpy(newbuf, req->rq_reqbuf, req->rq_reqbuf_len);
 
 		OBD_FREE_LARGE(req->rq_reqbuf, req->rq_reqbuf_len);

@@ -676,6 +685,9 @@ int plain_enlarge_reqbuf(struct ptlrpc_sec *sec,
 		req->rq_reqbuf_len = newbuf_size;
 		req->rq_reqmsg = lustre_msg_buf(req->rq_reqbuf,
 						PLAIN_PACK_MSG_OFF, 0);
+
+		if (req->rq_import)
+			spin_unlock(&req->rq_import->imp_lock);
 	}
 
 	_sptlrpc_enlarge_msg_inplace(req->rq_reqbuf, PLAIN_PACK_MSG_OFF,

index deca27e..7eadd60 100644
@@ -1537,12 +1537,14 @@ int ll_setattr(struct dentry *de, struct iattr *attr)
 	      !(attr->ia_mode & S_ISGID))))
 		attr->ia_valid |= ATTR_FORCE;
 
-	if ((mode & S_ISUID) &&
+	if ((attr->ia_valid & ATTR_MODE) &&
+	    (mode & S_ISUID) &&
 	    !(attr->ia_mode & S_ISUID) &&
 	    !(attr->ia_valid & ATTR_KILL_SUID))
 		attr->ia_valid |= ATTR_KILL_SUID;
 
-	if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
+	if ((attr->ia_valid & ATTR_MODE) &&
+	    ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
 	    !(attr->ia_mode & S_ISGID) &&
 	    !(attr->ia_valid & ATTR_KILL_SGID))
 		attr->ia_valid |= ATTR_KILL_SGID;

index 0c6b784..bf5b2cb 100644
@@ -1390,7 +1390,7 @@ int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
 int ldlm_cancel_resource_local(struct ldlm_resource *res,
 			       struct list_head *cancels,
 			       ldlm_policy_data_t *policy,
-			       ldlm_mode_t mode, int lock_flags,
+			       ldlm_mode_t mode, __u64 lock_flags,
 			       ldlm_cancel_flags_t cancel_flags, void *opaque);
 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
 			       ldlm_cancel_flags_t flags);

index e265820..f8a9d7c 100644
@@ -1818,7 +1818,7 @@ static inline int md_enqueue(struct obd_export *exp,
 			     struct lustre_handle *lockh,
 			     void *lmm, int lmmsize,
 			     struct ptlrpc_request **req,
-			     int extra_lock_flags)
+			     __u64 extra_lock_flags)
 {
 	int rc;
 

index af89843..54ef540 100644
@@ -87,6 +87,10 @@ struct osc_enqueue_args {
 	unsigned int	      oa_agl:1;
 };
 
+extern void osc_update_enqueue(struct lustre_handle *lov_lockhp,
+			       struct lov_oinfo *loi, __u64 flags,
+			       struct ost_lvb *lvb, __u32 mode, int rc);
+
 #if 0
 int osc_extent_blocking_cb(struct ldlm_lock *lock,
 			   struct ldlm_lock_desc *new, void *data,

index fcc7a99..3accbce 100644
@@ -1768,7 +1768,7 @@ int ldlm_cancel_lru(struct ldlm_namespace *ns, int nr,
 int ldlm_cancel_resource_local(struct ldlm_resource *res,
 			       struct list_head *cancels,
 			       ldlm_policy_data_t *policy,
-			       ldlm_mode_t mode, int lock_flags,
+			       ldlm_mode_t mode, __u64 lock_flags,
 			       ldlm_cancel_flags_t cancel_flags, void *opaque)
 {
 	struct ldlm_lock *lock;

index 716e1ee..660fd4d 100644
@@ -290,7 +290,7 @@ static int ll_md_close(struct obd_export *md_exp, struct inode *inode,
 	   we can skip talking to MDS */
 	if (file->f_dentry->d_inode) { /* Can this ever be false? */
 		int lockmode;
-		int flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
+		__u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
 		struct lustre_handle lockh;
 		struct inode *inode = file->f_dentry->d_inode;
 		ldlm_policy_data_t policy = {.l_inodebits={MDS_INODELOCK_OPEN}};

@@ -2623,7 +2623,7 @@ ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
 	struct md_op_data *op_data;
 	struct lustre_handle lockh = {0};
 	ldlm_policy_data_t flock = {{0}};
-	int flags = 0;
+	__u64 flags = 0;
 	int rc;
 	int rc2 = 0;
 

@@ -2708,7 +2708,7 @@ ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
 	if (IS_ERR(op_data))
 		return PTR_ERR(op_data);
 
-	CDEBUG(D_DLMTRACE, "inode=%lu, pid=%u, flags=%#x, mode=%u, "
+	CDEBUG(D_DLMTRACE, "inode=%lu, pid=%u, flags=%#llx, mode=%u, "
 	       "start="LPU64", end="LPU64"\n", inode->i_ino, flock.l_flock.pid,
 	       flags, einfo.ei_mode, flock.l_flock.start, flock.l_flock.end);
 

index 4edf8a3..c17a49e 100644
@@ -1715,7 +1715,7 @@ static int
 lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
 		   struct lookup_intent *it, struct md_op_data *op_data,
 		   struct lustre_handle *lockh, void *lmm, int lmmsize,
-		   int extra_lock_flags)
+		   __u64 extra_lock_flags)
 {
 	struct ptlrpc_request      *req = it->d.lustre.it_data;
 	struct obd_device	  *obd = exp->exp_obd;

index 38508a5..2232643 100644
@@ -252,7 +252,7 @@ int lov_prep_match_set(struct obd_export *exp, struct obd_info *oinfo,
 		       ldlm_policy_data_t *policy, __u32 mode,
 		       struct lustre_handle *lockh,
 		       struct lov_request_set **reqset);
-int lov_fini_match_set(struct lov_request_set *set, __u32 mode, int flags);
+int lov_fini_match_set(struct lov_request_set *set, __u32 mode, __u64 flags);
 int lov_prep_cancel_set(struct obd_export *exp, struct obd_info *oinfo,
 			struct lov_stripe_md *lsm,
 			__u32 mode, struct lustre_handle *lockh,

index bd6490d..984f4c3 100644
@@ -39,8 +39,8 @@
 #include <linux/libcfs/libcfs.h>
 
 #include <obd_class.h>
+#include <obd_ost.h>
 #include <lustre/lustre_idl.h>
-
 #include "lov_internal.h"
 
 static void lov_init_set(struct lov_request_set *set)

@@ -194,13 +194,9 @@ out:
 	return rc;
 }
 
-extern void osc_update_enqueue(struct lustre_handle *lov_lockhp,
-			       struct lov_oinfo *loi, int flags,
-			       struct ost_lvb *lvb, __u32 mode, int rc);
-
 static int lov_update_enqueue_lov(struct obd_export *exp,
 				  struct lustre_handle *lov_lockhp,
-				  struct lov_oinfo *loi, int flags, int idx,
+				  struct lov_oinfo *loi, __u64 flags, int idx,
 				  struct ost_id *oi, int rc)
 {
 	struct lov_obd *lov = &exp->exp_obd->u.lov;

@@ -443,7 +439,7 @@ out_set:
 	return rc;
 }
 
-int lov_fini_match_set(struct lov_request_set *set, __u32 mode, int flags)
+int lov_fini_match_set(struct lov_request_set *set, __u32 mode, __u64 flags)
 {
 	int rc = 0;
 

index efc5db4..9c4a189 100644
@@ -112,7 +112,7 @@ int osc_cancel_base(struct lustre_handle *lockh, __u32 mode);
 
 int osc_match_base(struct obd_export *exp, struct ldlm_res_id *res_id,
 		   __u32 type, ldlm_policy_data_t *policy, __u32 mode,
-		   int *flags, void *data, struct lustre_handle *lockh,
+		   __u64 *flags, void *data, struct lustre_handle *lockh,
 		   int unref);
 
 int osc_setattr_async_base(struct obd_export *exp, struct obd_info *oinfo,

index 96cb6e2..71a2447 100644
@@ -70,7 +70,7 @@ static int osc_page_is_dlocked(const struct lu_env *env,
 	struct lustre_handle   *lockh;
 	ldlm_policy_data_t     *policy;
 	ldlm_mode_t	     dlmmode;
-	int		     flags;
+	__u64                   flags;
 
 	might_sleep();
 

index 294db84..5804104 100644
@@ -635,7 +635,7 @@ static int osc_sync(const struct lu_env *env, struct obd_export *exp,
  * locks added to @cancels list. */
 static int osc_resource_get_unused(struct obd_export *exp, struct obdo *oa,
 				   struct list_head *cancels,
-				   ldlm_mode_t mode, int lock_flags)
+				   ldlm_mode_t mode, __u64 lock_flags)
 {
 	struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
 	struct ldlm_res_id res_id;

@@ -2398,7 +2398,7 @@ static int osc_enqueue_interpret(const struct lu_env *env,
 }
 
 void osc_update_enqueue(struct lustre_handle *lov_lockhp,
-			struct lov_oinfo *loi, int flags,
+			struct lov_oinfo *loi, __u64 flags,
 			struct ost_lvb *lvb, __u32 mode, int rc)
 {
 	struct ldlm_lock *lock = ldlm_handle2lock(lov_lockhp);

@@ -2462,7 +2462,7 @@ int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id,
 	struct obd_device *obd = exp->exp_obd;
 	struct ptlrpc_request *req = NULL;
 	int intent = *flags & LDLM_FL_HAS_INTENT;
-	int match_lvb = (agl != 0 ? 0 : LDLM_FL_LVB_READY);
+	__u64 match_lvb = (agl != 0 ? 0 : LDLM_FL_LVB_READY);
 	ldlm_mode_t mode;
 	int rc;
 

@@ -2613,11 +2613,11 @@ static int osc_enqueue(struct obd_export *exp, struct obd_info *oinfo,
 
 int osc_match_base(struct obd_export *exp, struct ldlm_res_id *res_id,
 		   __u32 type, ldlm_policy_data_t *policy, __u32 mode,
-		   int *flags, void *data, struct lustre_handle *lockh,
+		   __u64 *flags, void *data, struct lustre_handle *lockh,
 		   int unref)
 {
 	struct obd_device *obd = exp->exp_obd;
-	int lflags = *flags;
+	__u64 lflags = *flags;
 	ldlm_mode_t rc;
 
 	if (OBD_FAIL_CHECK(OBD_FAIL_OSC_MATCH))

index a47a8d8..ef18639 100644
@@ -506,10 +506,9 @@ int ptl_send_rpc(struct ptlrpc_request *request, int noreply)
 	 * cleanly from the previous attempt */
 	LASSERT(!request->rq_receiving_reply);
 
-	if (request->rq_import->imp_obd &&
-	    request->rq_import->imp_obd->obd_fail) {
+	if (unlikely(obd != NULL && obd->obd_fail)) {
 		CDEBUG(D_HA, "muting rpc for failed imp obd %s\n",
-		       request->rq_import->imp_obd->obd_name);
+			obd->obd_name);
 		/* this prevents us from waiting in ptlrpc_queue_wait */
 		spin_lock(&request->rq_lock);
 		request->rq_err = 1;

@@ -625,7 +624,7 @@ int ptl_send_rpc(struct ptlrpc_request *request, int noreply)
 
 	/* add references on request for request_out_callback */
 	ptlrpc_request_addref(request);
-	if (obd->obd_svc_stats != NULL)
+	if (obd != NULL && obd->obd_svc_stats != NULL)
 		lprocfs_counter_add(obd->obd_svc_stats, PTLRPC_REQACTIVE_CNTR,
 			atomic_read(&request->rq_import->imp_inflight));
 

index d278f2e..214daa2 100644
@@ -1100,6 +1100,7 @@ static void ptlrpc_update_export_timer(struct obd_export *exp, long extra_delay)
  */
 static int ptlrpc_check_req(struct ptlrpc_request *req)
 {
+	struct obd_device *obd = req->rq_export->exp_obd;
 	int rc = 0;
 
 	if (unlikely(lustre_msg_get_conn_cnt(req->rq_reqmsg) <

@@ -1110,24 +1111,23 @@ static int ptlrpc_check_req(struct ptlrpc_request *req)
 			  req->rq_export->exp_conn_cnt);
 		return -EEXIST;
 	}
-	if (unlikely(req->rq_export->exp_obd &&
-		     req->rq_export->exp_obd->obd_fail)) {
+	if (unlikely(obd == NULL || obd->obd_fail)) {
 		/*
 		 * Failing over, don't handle any more reqs, send
 		 * error response instead.
 		 */
 		CDEBUG(D_RPCTRACE, "Dropping req %p for failed obd %s\n",
-		       req, req->rq_export->exp_obd->obd_name);
+		       req, (obd != NULL) ? obd->obd_name : "unknown");
 		rc = -ENODEV;
 	} else if (lustre_msg_get_flags(req->rq_reqmsg) &
 		   (MSG_REPLAY | MSG_REQ_REPLAY_DONE) &&
-		   !(req->rq_export->exp_obd->obd_recovering)) {
+		   !obd->obd_recovering) {
 			DEBUG_REQ(D_ERROR, req,
 				  "Invalid replay without recovery");
 			class_fail_export(req->rq_export);
 			rc = -ENODEV;
 	} else if (lustre_msg_get_transno(req->rq_reqmsg) != 0 &&
-		   !(req->rq_export->exp_obd->obd_recovering)) {
+		   !obd->obd_recovering) {
 			DEBUG_REQ(D_ERROR, req, "Invalid req with transno "
 				  LPU64" without recovery",
 				  lustre_msg_get_transno(req->rq_reqmsg));

index a034aee..03d9a6a 100644
@@ -219,7 +219,6 @@ int lustre_start_mgc(struct super_block *sb)
 	lnet_nid_t nid;
 	char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
 	char *ptr;
-	int recov_bk;
 	int rc = 0, i = 0, j, len;
 
 	LASSERT(lsi->lsi_lmd);

@@ -269,6 +268,8 @@ int lustre_start_mgc(struct super_block *sb)
 
 	obd = class_name2obd(mgcname);
 	if (obd && !obd->obd_stopping) {
+		int recov_bk;
+
 		rc = obd_set_info_async(NULL, obd->obd_self_export,
 					strlen(KEY_MGSSEC), KEY_MGSSEC,
 					strlen(mgssec), mgssec, NULL);

@@ -429,16 +430,6 @@ int lustre_start_mgc(struct super_block *sb)
 	   so we know when we can get rid of the mgc. */
 	atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
 
-	/* Try all connections, but only once. */
-	recov_bk = 1;
-	rc = obd_set_info_async(NULL, obd->obd_self_export,
-				sizeof(KEY_INIT_RECOV_BACKUP),
-				KEY_INIT_RECOV_BACKUP,
-				sizeof(recov_bk), &recov_bk, NULL);
-	if (rc)
-		/* nonfatal */
-		CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
-
 	/* We connect to the MGS at setup, and don't disconnect until cleanup */
 	data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
 				  OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV |

71 lines not on this page · raw message

index 38099d9..2898087 100644
@@ -224,6 +224,11 @@ static void ptlrpc_pinger_process_import(struct obd_import *imp,
 		       "or recovery disabled: %s)\n",
 		       imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd),
 		       ptlrpc_import_state_name(level));
+		if (force) {
+			spin_lock(&imp->imp_lock);
+			imp->imp_force_verify = 1;
+			spin_unlock(&imp->imp_lock);
+		}
 	} else if ((imp->imp_pingable && !suppress) || force_next || force) {
 		ptlrpc_ping(imp);
 	}

index 5804104..90e8912 100644
@@ -1570,12 +1570,7 @@ static int osc_brw_fini_request(struct ptlrpc_request *req, int rc)
 			router = libcfs_nid2str(req->rq_bulk->bd_sender);
 		}
 
-		if (server_cksum == ~0 && rc > 0) {
-			CERROR("Protocol error: server %s set the 'checksum' "
-			       "bit, but didn't send a checksum.  Not fatal, "
-			       "but please notify on http://bugs.whamcloud.com/\n",
-			       libcfs_nid2str(peer->nid));
-		} else if (server_cksum != client_cksum) {
+		if (server_cksum != client_cksum) {
 			LCONSOLE_ERROR_MSG(0x133, "%s: BAD READ CHECKSUM: from "
 					   "%s%s%s inode "DFID" object "DOSTID
 					   " extent ["LPU64"-"LPU64"]\n",

index 7246e8c..d806257 100644
@@ -2530,10 +2530,19 @@ EXPORT_SYMBOL(ptlrpc_cleanup_client);
 void ptlrpc_resend_req(struct ptlrpc_request *req)
 {
 	DEBUG_REQ(D_HA, req, "going to resend");
+	spin_lock(&req->rq_lock);
+
+	/* Request got reply but linked to the import list still.
+	   Let ptlrpc_check_set() to process it. */
+	if (ptlrpc_client_replied(req)) {
+		spin_unlock(&req->rq_lock);
+		DEBUG_REQ(D_HA, req, "it has reply, so skip it");
+		return;
+	}
+
 	lustre_msg_set_handle(req->rq_reqmsg, &(struct lustre_handle){ 0 });
 	req->rq_status = -EAGAIN;
 
-	spin_lock(&req->rq_lock);
 	req->rq_resend = 1;
 	req->rq_net_err = 0;
 	req->rq_timedout = 0;

index aa85239..9f9b8d1 100644
@@ -145,6 +145,8 @@ void reply_in_callback(lnet_event_t *ev)
 		/* Real reply */
 		req->rq_rep_swab_mask = 0;
 		req->rq_replied = 1;
+		/* Got reply, no resend required */
+		req->rq_resend = 0;
 		req->rq_reply_off = ev->offset;
 		req->rq_nob_received = ev->mlength;
 		/* LNetMDUnlink can't be called under the LNET_LOCK,

index ef18639..f760504 100644
@@ -505,6 +505,8 @@ int ptl_send_rpc(struct ptlrpc_request *request, int noreply)
 	/* If this is a re-transmit, we're required to have disengaged
 	 * cleanly from the previous attempt */
 	LASSERT(!request->rq_receiving_reply);
+	LASSERT(!((lustre_msg_get_flags(request->rq_reqmsg) & MSG_REPLAY) &&
+		(request->rq_import->imp_state == LUSTRE_IMP_FULL)));
 
 	if (unlikely(obd != NULL && obd->obd_fail)) {
 		CDEBUG(D_HA, "muting rpc for failed imp obd %s\n",

index a806aef..28960f9 100644
@@ -950,7 +950,10 @@ static int mgc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 }
 
 /* Not sure where this should go... */
-#define  MGC_ENQUEUE_LIMIT 50
+/* This is the timeout value for MGS_CONNECT request plus a ping interval, such
+ * that we can have a chance to try the secondary MGS if any. */
+#define  MGC_ENQUEUE_LIMIT (INITIAL_CONNECT_TIMEOUT + (AT_OFF ? 0 : at_min) \
+				+ PING_INTERVAL)
 #define  MGC_TARGET_REG_LIMIT 10
 #define  MGC_SEND_PARAM_LIMIT 10
 

index d806257..1890482 100644
@@ -1496,6 +1496,8 @@ static inline int ptlrpc_set_producer(struct ptlrpc_request_set *set)
  * and no more replies are expected.
  * (it is possible to get less replies than requests sent e.g. due to timed out
  * requests or requests that we had trouble to send out)
+ *
+ * NOTE: This function contains a potential schedule point (cond_resched()).
  */
 int ptlrpc_check_set(const struct lu_env *env, struct ptlrpc_request_set *set)
 {

@@ -1513,6 +1515,14 @@ int ptlrpc_check_set(const struct lu_env *env, struct ptlrpc_request_set *set)
 		int unregistered = 0;
 		int rc = 0;
 
+		/* This schedule point is mainly for the ptlrpcd caller of this
+		 * function.  Most ptlrpc sets are not long-lived and unbounded
+		 * in length, but at the least the set used by the ptlrpcd is.
+		 * Since the processing time is unbounded, we need to insert an
+		 * explicit schedule point to make the thread well-behaved.
+		 */
+		cond_resched();
+
 		if (req->rq_phase == RQ_PHASE_NEW &&
 		    ptlrpc_send_new_req(req)) {
 			force_timer_recalc = 1;

index be1c613..cf1c497 100644
@@ -279,6 +279,7 @@ int capa_hmac(__u8 *hmac, struct lustre_capa *capa, __u8 *key)
 	}
 	keylen = alg->ha_keylen;
 
+	sg_init_table(&sl, 1);
 	sg_set_page(&sl, virt_to_page(capa),
 		    offsetof(struct lustre_capa, lc_hmac),
 		    (unsigned long)(capa) % PAGE_CACHE_SIZE);

@@ -320,9 +321,11 @@ int capa_encrypt_id(__u32 *d, __u32 *s, __u8 *key, int keylen)
 		GOTO(out, rc);
 	}
 
+	sg_init_table(&sd, 1);
 	sg_set_page(&sd, virt_to_page(d), 16,
 		    (unsigned long)(d) % PAGE_CACHE_SIZE);
 
+	sg_init_table(&ss, 1);
 	sg_set_page(&ss, virt_to_page(s), 16,
 		    (unsigned long)(s) % PAGE_CACHE_SIZE);
 	desc.tfm   = tfm;

@@ -370,9 +373,11 @@ int capa_decrypt_id(__u32 *d, __u32 *s, __u8 *key, int keylen)
 		GOTO(out, rc);
 	}
 
+	sg_init_table(&sd, 1);
 	sg_set_page(&sd, virt_to_page(d), 16,
 		    (unsigned long)(d) % PAGE_CACHE_SIZE);
 
+	sg_init_table(&ss, 1);
 	sg_set_page(&ss, virt_to_page(s), 16,
 		    (unsigned long)(s) % PAGE_CACHE_SIZE);
 

index 00f38ee..f075b69 100644
@@ -871,7 +871,8 @@ static int osc_extent_wait(const struct lu_env *env, struct osc_extent *ext,
 	LASSERT(sanity_check_nolock(ext) == 0);
 	/* `Kick' this extent only if the caller is waiting for it to be
 	 * written out. */
-	if (state == OES_INV && !ext->oe_urgent && !ext->oe_hp) {
+	if (state == OES_INV && !ext->oe_urgent && !ext->oe_hp &&
+	    !ext->oe_trunc_pending) {
 		if (ext->oe_state == OES_ACTIVE) {
 			ext->oe_urgent = 1;
 		} else if (ext->oe_state == OES_CACHE) {

@@ -922,8 +923,8 @@ static int osc_extent_truncate(struct osc_extent *ext, pgoff_t trunc_index,
 	int		    rc       = 0;
 
 	LASSERT(sanity_check(ext) == 0);
-	LASSERT(ext->oe_state == OES_TRUNC);
-	LASSERT(!ext->oe_urgent);
+	EASSERT(ext->oe_state == OES_TRUNC, ext);
+	EASSERT(!ext->oe_urgent, ext);
 
 	/* Request new lu_env.
 	 * We can't use that env from osc_cache_truncate_start() because

index c6c27bb..c1eff65 100644
@@ -246,6 +246,7 @@ int ll_setxattr(struct dentry *dentry, const char *name,
 			int lum_size = (lump->lmm_magic == LOV_USER_MAGIC_V1) ?
 				sizeof(*lump) : sizeof(struct lov_user_md_v3);
 
+			memset(&f, 0, sizeof(f)); /* f.f_flags is used below */
 			f.f_dentry = dentry;
 			rc = ll_lov_setstripe_ea_info(inode, &f, flags, lump,
 						      lum_size);

index f6b7d10..b837d34 100644
@@ -1591,7 +1591,8 @@ struct ptlrpc_request {
 		rq_replay:1,
 		rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
 		rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1,
-		rq_early:1, rq_must_unlink:1,
+		rq_early:1,
+		rq_req_unlink:1, rq_reply_unlink:1,
 		rq_memalloc:1,      /* req originated from "kswapd" */
 		/* server-side flags */
 		rq_packed_final:1,  /* packed final reply */

@@ -3039,7 +3040,8 @@ ptlrpc_client_recv_or_unlink(struct ptlrpc_request *req)
 		spin_unlock(&req->rq_lock);
 		return 1;
 	}
-	rc = req->rq_receiving_reply || req->rq_must_unlink;
+	rc = req->rq_receiving_reply;
+	rc = rc || req->rq_req_unlink || req->rq_reply_unlink;
 	spin_unlock(&req->rq_lock);
 	return rc;
 }

index 1890482..0e0ea5c 100644
@@ -1202,7 +1202,7 @@ static int after_reply(struct ptlrpc_request *req)
 
 	LASSERT(obd != NULL);
 	/* repbuf must be unlinked */
-	LASSERT(!req->rq_receiving_reply && !req->rq_must_unlink);
+	LASSERT(!req->rq_receiving_reply && !req->rq_reply_unlink);
 
 	if (req->rq_reply_truncate) {
 		if (ptlrpc_no_resend(req)) {

@@ -2406,9 +2406,10 @@ int ptlrpc_unregister_reply(struct ptlrpc_request *request, int async)
 		}
 
 		LASSERT(rc == -ETIMEDOUT);
-		DEBUG_REQ(D_WARNING, request, "Unexpectedly long timeout "
-			  "rvcng=%d unlnk=%d", request->rq_receiving_reply,
-			  request->rq_must_unlink);
+		DEBUG_REQ(D_WARNING, request,
+			  "Unexpectedly long timeout rvcng=%d unlnk=%d/%d",
+			  request->rq_receiving_reply,
+			  request->rq_req_unlink, request->rq_reply_unlink);
 	}
 	return 0;
 }

@@ -3081,7 +3082,7 @@ void *ptlrpcd_alloc_work(struct obd_import *imp,
 	req->rq_interpret_reply = work_interpreter;
 	/* don't want reply */
 	req->rq_receiving_reply = 0;
-	req->rq_must_unlink = 0;
+	req->rq_req_unlink = req->rq_reply_unlink = 0;
 	req->rq_no_delay = req->rq_no_resend = 1;
 	req->rq_pill.rc_fmt = (void *)&worker_format;
 

index 9f9b8d1..209fcc1 100644
@@ -63,19 +63,20 @@ void request_out_callback(lnet_event_t *ev)
 	DEBUG_REQ(D_NET, req, "type %d, status %d", ev->type, ev->status);
 
 	sptlrpc_request_out_callback(req);
+	spin_lock(&req->rq_lock);
 	req->rq_real_sent = cfs_time_current_sec();
+	if (ev->unlinked)
+		req->rq_req_unlink = 0;
 
 	if (ev->type == LNET_EVENT_UNLINK || ev->status != 0) {
 
 		/* Failed send: make it seem like the reply timed out, just
 		 * like failing sends in client.c does currently...  */
 
-		spin_lock(&req->rq_lock);
 		req->rq_net_err = 1;
-		spin_unlock(&req->rq_lock);
-
 		ptlrpc_client_wake_req(req);
 	}
+	spin_unlock(&req->rq_lock);
 
 	ptlrpc_req_finished(req);
 }

@@ -102,7 +103,7 @@ void reply_in_callback(lnet_event_t *ev)
 	req->rq_receiving_reply = 0;
 	req->rq_early = 0;
 	if (ev->unlinked)
-		req->rq_must_unlink = 0;
+		req->rq_reply_unlink = 0;
 
 	if (ev->status)
 		goto out_wake;

index f760504..3f0ca23 100644
@@ -580,8 +580,9 @@ int ptl_send_rpc(struct ptlrpc_request *request, int noreply)
 	spin_lock(&request->rq_lock);
 	/* If the MD attach succeeds, there _will_ be a reply_in callback */
 	request->rq_receiving_reply = !noreply;
+	request->rq_req_unlink = 1;
 	/* We are responsible for unlinking the reply buffer */
-	request->rq_must_unlink = !noreply;
+	request->rq_reply_unlink = !noreply;
 	/* Clear any flags that may be present from previous sends. */
 	request->rq_replied = 0;
 	request->rq_err = 0;

@@ -604,7 +605,7 @@ int ptl_send_rpc(struct ptlrpc_request *request, int noreply)
 		reply_md.user_ptr  = &request->rq_reply_cbid;
 		reply_md.eq_handle = ptlrpc_eq_h;
 
-		/* We must see the unlink callback to unset rq_must_unlink,
+		/* We must see the unlink callback to unset rq_reply_unlink,
 		   so we can't auto-unlink */
 		rc = LNetMDAttach(reply_me_h, reply_md, LNET_RETAIN,
 				  &request->rq_reply_md_h);

index 03d9a6a..4e77f07 100644
@@ -1216,7 +1216,9 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent)
 
 	if (lmd_is_client(lmd)) {
 		CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
-		if (!client_fill_super) {
+		if (client_fill_super == NULL)
+			request_module("lustre");
+		if (client_fill_super == NULL) {
 			LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
 					   "client mount! Is the 'lustre' "
 					   "module loaded?\n");

@@ -1299,6 +1301,7 @@ struct file_system_type lustre_fs_type = {
 	.fs_flags     = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
 			FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE,
 };
+MODULE_ALIAS_FS("lustre");
 
 int lustre_register_fs(void)
 {

index 0e0b404..04230ed 100644
@@ -269,8 +269,10 @@ static int vvp_mmap_locks(const struct lu_env *env,
 			       descr->cld_mode, descr->cld_start,
 			       descr->cld_end);
 
-			if (result < 0)
+			if (result < 0) {
+				up_read(&mm->mmap_sem);
 				return result;
+			}
 
 			if (vma->vm_end - addr >= count)
 				break;

index 04230ed..2539a89 100644
@@ -624,7 +624,7 @@ static int vvp_io_kernel_fault(struct vvp_fault_io *cfio)
 		       page_private(vmf->page), vmf->virtual_address);
 		if (unlikely(!(cfio->fault.ft_flags & VM_FAULT_LOCKED))) {
 			lock_page(vmf->page);
-			cfio->fault.ft_flags &= VM_FAULT_LOCKED;
+			cfio->fault.ft_flags |= VM_FAULT_LOCKED;
 		}
 
 		cfio->ft_vmpage = vmf->page;

index a63654b..6816aa0 100644
@@ -280,6 +280,7 @@ typedef struct lnet_libmd {
 
 #define LNET_MD_FLAG_ZOMBIE	   (1 << 0)
 #define LNET_MD_FLAG_AUTO_UNLINK      (1 << 1)
+#define LNET_MD_FLAG_ABORTED	 (1 << 2)
 
 #ifdef LNET_USE_LIB_FREELIST
 typedef struct {

index ae643f2..d68c6e0 100644
@@ -387,7 +387,8 @@ EXPORT_SYMBOL(LNetMDBind);
 
 /**
  * Unlink the memory descriptor from any ME it may be linked to and release
- * the internal resources associated with it.
+ * the internal resources associated with it. As a result, active messages
+ * associated with the MD may get aborted.
  *
  * This function does not free the memory region associated with the MD;
  * i.e., the memory the user allocated for this MD. If the ME associated with

@@ -433,12 +434,11 @@ LNetMDUnlink (lnet_handle_md_t mdh)
 		return -ENOENT;
 	}
 
+	md->md_flags |= LNET_MD_FLAG_ABORTED;
 	/* If the MD is busy, lnet_md_unlink just marks it for deletion, and
-	 * when the NAL is done, the completion event flags that the MD was
+	 * when the LND is done, the completion event flags that the MD was
 	 * unlinked.  Otherwise, we enqueue an event now... */
-
-	if (md->md_eq != NULL &&
-	    md->md_refcount == 0) {
+	if (md->md_eq != NULL && md->md_refcount == 0) {
 		lnet_build_unlink_event(md, &ev);
 		lnet_eq_enqueue_event(md->md_eq, &ev);
 	}

index 0081075..0e42209 100644
@@ -246,11 +246,12 @@ LNetMEUnlink(lnet_handle_me_t meh)
 	}
 
 	md = me->me_md;
-	if (md != NULL &&
-	    md->md_eq != NULL &&
-	    md->md_refcount == 0) {
-		lnet_build_unlink_event(md, &ev);
-		lnet_eq_enqueue_event(md->md_eq, &ev);
+	if (md != NULL) {
+		md->md_flags |= LNET_MD_FLAG_ABORTED;
+		if (md->md_eq != NULL && md->md_refcount == 0) {
+			lnet_build_unlink_event(md, &ev);
+			lnet_eq_enqueue_event(md->md_eq, &ev);
+		}
 	}
 
 	lnet_me_unlink(me);

85 lines not on this page · raw message