Whamcloud - gitweb
LU-7117 osp: set ptlrpc_request::rq_allow_replay properly
[fs/lustre-release.git] / lustre / fid / fid_request.c
index 14de931..7f96185 100644 (file)
@@ -27,7 +27,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2013, Intel Corporation.
+ * Copyright (c) 2011, 2015, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
 
 #define DEBUG_SUBSYSTEM S_FID
 
-#ifdef __KERNEL__
-# include <libcfs/libcfs.h>
-# include <linux/module.h>
-#else /* __KERNEL__ */
-# include <liblustre.h>
-#endif
-
+#include <linux/module.h>
+#include <libcfs/libcfs.h>
 #include <obd.h>
 #include <obd_class.h>
 #include <obd_support.h>
@@ -81,7 +76,7 @@ static int seq_client_rpc(struct lu_client_seq *seq,
 
        /* Zero out input range, this is not recovery yet. */
        in = req_capsule_client_get(&req->rq_pill, &RMF_SEQ_RANGE);
-       range_init(in);
+       lu_seq_range_init(in);
 
        ptlrpc_request_set_replen(req);
 
@@ -115,28 +110,26 @@ static int seq_client_rpc(struct lu_client_seq *seq,
                debug_mask = D_INFO;
        }
 
-       ptlrpc_at_set_req_timeout(req);
+       /* Allow seq client RPC during recovery time. */
+       req->rq_allow_replay = 1;
 
-       if (opc != SEQ_ALLOC_SUPER && seq->lcs_type == LUSTRE_SEQ_METADATA)
-               mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
+       ptlrpc_at_set_req_timeout(req);
 
        rc = ptlrpc_queue_wait(req);
 
-       if (opc != SEQ_ALLOC_SUPER && seq->lcs_type == LUSTRE_SEQ_METADATA)
-               mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
        if (rc)
                GOTO(out_req, rc);
 
        out = req_capsule_server_get(&req->rq_pill, &RMF_SEQ_RANGE);
        *output = *out;
 
-       if (!range_is_sane(output)) {
+       if (!lu_seq_range_is_sane(output)) {
                CERROR("%s: Invalid range received from server: "
                       DRANGE"\n", seq->lcs_name, PRANGE(output));
                GOTO(out_req, rc = -EINVAL);
        }
 
-       if (range_is_exhausted(output)) {
+       if (lu_seq_range_is_exhausted(output)) {
                CERROR("%s: Range received from server is exhausted: "
                       DRANGE"]\n", seq->lcs_name, PRANGE(output));
                GOTO(out_req, rc = -EINVAL);
@@ -213,14 +206,14 @@ static int seq_client_alloc_meta(const struct lu_env *env,
 
 /* Allocate new sequence for client. */
 static int seq_client_alloc_seq(const struct lu_env *env,
-                                struct lu_client_seq *seq, seqno_t *seqnr)
+                               struct lu_client_seq *seq, u64 *seqnr)
 {
-        int rc;
-        ENTRY;
+       int rc;
+       ENTRY;
 
-        LASSERT(range_is_sane(&seq->lcs_space));
+       LASSERT(lu_seq_range_is_sane(&seq->lcs_space));
 
-        if (range_is_exhausted(&seq->lcs_space)) {
+       if (lu_seq_range_is_exhausted(&seq->lcs_space)) {
                 rc = seq_client_alloc_meta(env, seq);
                 if (rc) {
                         CERROR("%s: Can't allocate new meta-sequence,"
@@ -234,9 +227,9 @@ static int seq_client_alloc_seq(const struct lu_env *env,
                 rc = 0;
         }
 
-        LASSERT(!range_is_exhausted(&seq->lcs_space));
-        *seqnr = seq->lcs_space.lsr_start;
-        seq->lcs_space.lsr_start += 1;
+       LASSERT(!lu_seq_range_is_exhausted(&seq->lcs_space));
+       *seqnr = seq->lcs_space.lsr_start;
+       seq->lcs_space.lsr_start += 1;
 
         CDEBUG(D_INFO, "%s: Allocated sequence ["LPX64"]\n", seq->lcs_name,
                *seqnr);
@@ -252,78 +245,96 @@ static int seq_fid_alloc_prep(struct lu_client_seq *seq,
                set_current_state(TASK_UNINTERRUPTIBLE);
                mutex_unlock(&seq->lcs_mutex);
 
-               waitq_wait(link, TASK_UNINTERRUPTIBLE);
+               schedule();
 
                mutex_lock(&seq->lcs_mutex);
                remove_wait_queue(&seq->lcs_waitq, link);
                set_current_state(TASK_RUNNING);
                return -EAGAIN;
        }
+
        ++seq->lcs_update;
        mutex_unlock(&seq->lcs_mutex);
+
        return 0;
 }
 
-static void seq_fid_alloc_fini(struct lu_client_seq *seq)
+static void seq_fid_alloc_fini(struct lu_client_seq *seq, __u64 seqnr,
+                              bool whole)
 {
        LASSERT(seq->lcs_update == 1);
+
        mutex_lock(&seq->lcs_mutex);
+       if (seqnr != 0) {
+               CDEBUG(D_INFO, "%s: New sequence [0x%16.16"LPF64"x]\n",
+                      seq->lcs_name, seqnr);
+
+               seq->lcs_fid.f_seq = seqnr;
+               if (whole) {
+                       /* Since the caller require the whole seq,
+                        * so marked this seq to be used */
+                       if (seq->lcs_type == LUSTRE_SEQ_METADATA)
+                               seq->lcs_fid.f_oid =
+                                       LUSTRE_METADATA_SEQ_MAX_WIDTH;
+                       else
+                               seq->lcs_fid.f_oid = LUSTRE_DATA_SEQ_MAX_WIDTH;
+               } else {
+                       seq->lcs_fid.f_oid = LUSTRE_FID_INIT_OID;
+               }
+               seq->lcs_fid.f_ver = 0;
+       }
+
        --seq->lcs_update;
-       wake_up(&seq->lcs_waitq);
+       wake_up_all(&seq->lcs_waitq);
 }
 
 /**
- * Allocate the whole seq to the caller.
- **/
+ * Allocate the whole non-used seq to the caller.
+ *
+ * \param[in] env      pointer to the thread context
+ * \param[in,out] seq  pointer to the client sequence manager
+ * \param[out] seqnr   to hold the new allocated sequence
+ *
+ * \retval             0 for new sequence allocated.
+ * \retval             Negative error number on failure.
+ */
 int seq_client_get_seq(const struct lu_env *env,
-                      struct lu_client_seq *seq, seqno_t *seqnr)
+                      struct lu_client_seq *seq, u64 *seqnr)
 {
        wait_queue_t link;
        int rc;
 
        LASSERT(seqnr != NULL);
-       mutex_lock(&seq->lcs_mutex);
-       init_waitqueue_entry_current(&link);
 
-        while (1) {
-                rc = seq_fid_alloc_prep(seq, &link);
-                if (rc == 0)
-                        break;
-        }
-
-        rc = seq_client_alloc_seq(env, seq, seqnr);
-        if (rc) {
-                CERROR("%s: Can't allocate new sequence, "
-                       "rc %d\n", seq->lcs_name, rc);
-                seq_fid_alloc_fini(seq);
-               mutex_unlock(&seq->lcs_mutex);
-                return rc;
-        }
-
-        CDEBUG(D_INFO, "%s: allocate sequence "
-               "[0x%16.16"LPF64"x]\n", seq->lcs_name, *seqnr);
+       mutex_lock(&seq->lcs_mutex);
+       init_waitqueue_entry(&link, current);
 
-       /* Since the caller require the whole seq,
-        * so marked this seq to be used */
-       if (seq->lcs_type == LUSTRE_SEQ_METADATA)
-               seq->lcs_fid.f_oid = LUSTRE_METADATA_SEQ_MAX_WIDTH;
-       else
-               seq->lcs_fid.f_oid = LUSTRE_DATA_SEQ_MAX_WIDTH;
+       /* To guarantee that we can get a whole non-used sequence. */
+       while (seq_fid_alloc_prep(seq, &link) != 0);
 
-       seq->lcs_fid.f_seq = *seqnr;
-       seq->lcs_fid.f_ver = 0;
-        /*
-         * Inform caller that sequence switch is performed to allow it
-         * to setup FLD for it.
-         */
-        seq_fid_alloc_fini(seq);
+       rc = seq_client_alloc_seq(env, seq, seqnr);
+       seq_fid_alloc_fini(seq, rc ? 0 : *seqnr, true);
+       if (rc)
+               CERROR("%s: Can't allocate new sequence: rc = %d\n",
+                      seq->lcs_name, rc);
        mutex_unlock(&seq->lcs_mutex);
 
-        return rc;
+       return rc;
 }
 EXPORT_SYMBOL(seq_client_get_seq);
 
-/* Allocate new fid on passed client @seq and save it to @fid. */
+/**
+ * Allocate new fid on passed client @seq and save it to @fid.
+ *
+ * \param[in] env      pointer to the thread context
+ * \param[in,out] seq  pointer to the client sequence manager
+ * \param[out] fid     to hold the new allocated fid
+ *
+ * \retval             1 for notify the caller that sequence switch
+ *                     is performed to allow it to setup FLD for it.
+ * \retval             0 for new FID allocated in current sequence.
+ * \retval             Negative error number on failure.
+ */
 int seq_client_alloc_fid(const struct lu_env *env,
                         struct lu_client_seq *seq, struct lu_fid *fid)
 {
@@ -334,58 +345,50 @@ int seq_client_alloc_fid(const struct lu_env *env,
        LASSERT(seq != NULL);
        LASSERT(fid != NULL);
 
-       init_waitqueue_entry_current(&link);
+       init_waitqueue_entry(&link, current);
        mutex_lock(&seq->lcs_mutex);
 
        if (OBD_FAIL_CHECK(OBD_FAIL_SEQ_EXHAUST))
                seq->lcs_fid.f_oid = seq->lcs_width;
 
-        while (1) {
-                seqno_t seqnr;
+       while (1) {
+               u64 seqnr;
 
-                if (!fid_is_zero(&seq->lcs_fid) &&
-                    fid_oid(&seq->lcs_fid) < seq->lcs_width) {
-                        /* Just bump last allocated fid and return to caller. */
-                        seq->lcs_fid.f_oid += 1;
-                        rc = 0;
-                        break;
-                }
-
-                rc = seq_fid_alloc_prep(seq, &link);
-                if (rc)
-                        continue;
+               if (unlikely(!fid_is_zero(&seq->lcs_fid) &&
+                            fid_oid(&seq->lcs_fid) < seq->lcs_width)) {
+                       /* Just bump last allocated fid and return to caller. */
+                       seq->lcs_fid.f_oid++;
+                       rc = 0;
+                       break;
+               }
 
-                rc = seq_client_alloc_seq(env, seq, &seqnr);
-                if (rc) {
-                        CERROR("%s: Can't allocate new sequence, "
-                               "rc %d\n", seq->lcs_name, rc);
-                        seq_fid_alloc_fini(seq);
+               /* Release seq::lcs_mutex via seq_fid_alloc_prep() to avoid
+                * deadlock during seq_client_alloc_seq(). */
+               rc = seq_fid_alloc_prep(seq, &link);
+               if (rc)
+                       continue;
+
+               rc = seq_client_alloc_seq(env, seq, &seqnr);
+               /* Re-take seq::lcs_mutex via seq_fid_alloc_fini(). */
+               seq_fid_alloc_fini(seq, rc ? 0 : seqnr, false);
+               if (rc) {
+                       CERROR("%s: Can't allocate new sequence: rc = %d\n",
+                              seq->lcs_name, rc);
                        mutex_unlock(&seq->lcs_mutex);
-                        RETURN(rc);
-                }
 
-                CDEBUG(D_INFO, "%s: Switch to sequence "
-                       "[0x%16.16"LPF64"x]\n", seq->lcs_name, seqnr);
-
-                seq->lcs_fid.f_oid = LUSTRE_FID_INIT_OID;
-                seq->lcs_fid.f_seq = seqnr;
-                seq->lcs_fid.f_ver = 0;
-
-                /*
-                 * Inform caller that sequence switch is performed to allow it
-                 * to setup FLD for it.
-                 */
-                rc = 1;
+                       RETURN(rc);
+               }
 
-                seq_fid_alloc_fini(seq);
-                break;
-        }
+               rc = 1;
+               break;
+       }
 
-        *fid = seq->lcs_fid;
+       *fid = seq->lcs_fid;
        mutex_unlock(&seq->lcs_mutex);
 
-        CDEBUG(D_INFO, "%s: Allocated FID "DFID"\n", seq->lcs_name,  PFID(fid));
-        RETURN(rc);
+       CDEBUG(D_INFO, "%s: Allocated FID "DFID"\n", seq->lcs_name,  PFID(fid));
+
+       RETURN(rc);
 }
 EXPORT_SYMBOL(seq_client_alloc_fid);
 
@@ -398,7 +401,7 @@ void seq_client_flush(struct lu_client_seq *seq)
        wait_queue_t link;
 
        LASSERT(seq != NULL);
-       init_waitqueue_entry_current(&link);
+       init_waitqueue_entry(&link, current);
        mutex_lock(&seq->lcs_mutex);
 
        while (seq->lcs_update) {
@@ -406,7 +409,7 @@ void seq_client_flush(struct lu_client_seq *seq)
                set_current_state(TASK_UNINTERRUPTIBLE);
                mutex_unlock(&seq->lcs_mutex);
 
-               waitq_wait(&link, TASK_UNINTERRUPTIBLE);
+               schedule();
 
                mutex_lock(&seq->lcs_mutex);
                remove_wait_queue(&seq->lcs_waitq, &link);
@@ -421,14 +424,14 @@ void seq_client_flush(struct lu_client_seq *seq)
 
         seq->lcs_space.lsr_index = -1;
 
-        range_init(&seq->lcs_space);
+       lu_seq_range_init(&seq->lcs_space);
        mutex_unlock(&seq->lcs_mutex);
 }
 EXPORT_SYMBOL(seq_client_flush);
 
 static void seq_client_proc_fini(struct lu_client_seq *seq)
 {
-#ifdef LPROCFS
+#ifdef CONFIG_PROC_FS
        ENTRY;
        if (seq->lcs_proc_dir) {
                if (!IS_ERR(seq->lcs_proc_dir))
@@ -436,18 +439,17 @@ static void seq_client_proc_fini(struct lu_client_seq *seq)
                seq->lcs_proc_dir = NULL;
        }
        EXIT;
-#endif /* LPROCFS */
+#endif /* CONFIG_PROC_FS */
 }
 
 static int seq_client_proc_init(struct lu_client_seq *seq)
 {
-#ifdef LPROCFS
+#ifdef CONFIG_PROC_FS
         int rc;
         ENTRY;
 
-       seq->lcs_proc_dir = lprocfs_seq_register(seq->lcs_name,
-                                                seq_type_proc_dir,
-                                                NULL, NULL);
+       seq->lcs_proc_dir = lprocfs_register(seq->lcs_name, seq_type_proc_dir,
+                                            NULL, NULL);
         if (IS_ERR(seq->lcs_proc_dir)) {
                 CERROR("%s: LProcFS failed in seq-init\n",
                        seq->lcs_name);
@@ -455,8 +457,7 @@ static int seq_client_proc_init(struct lu_client_seq *seq)
                 RETURN(rc);
         }
 
-       rc = lprocfs_seq_add_vars(seq->lcs_proc_dir,
-                                 seq_client_proc_list, seq);
+       rc = lprocfs_add_vars(seq->lcs_proc_dir, seq_client_proc_list, seq);
         if (rc) {
                 CERROR("%s: Can't init sequence manager "
                        "proc, rc %d\n", seq->lcs_name, rc);
@@ -469,9 +470,9 @@ out_cleanup:
         seq_client_proc_fini(seq);
         return rc;
 
-#else /* LPROCFS */
+#else /* !CONFIG_PROC_FS */
        return 0;
-#endif
+#endif /* CONFIG_PROC_FS */
 }
 
 int seq_client_init(struct lu_client_seq *seq,
@@ -575,14 +576,13 @@ int client_fid_fini(struct obd_device *obd)
 }
 EXPORT_SYMBOL(client_fid_fini);
 
-#ifdef __KERNEL__
 struct proc_dir_entry *seq_type_proc_dir;
 
-static int __init fid_mod_init(void)
+static int __init fid_init(void)
 {
-       seq_type_proc_dir = lprocfs_seq_register(LUSTRE_SEQ_NAME,
-                                                proc_lustre_root,
-                                                NULL, NULL);
+       seq_type_proc_dir = lprocfs_register(LUSTRE_SEQ_NAME,
+                                            proc_lustre_root,
+                                            NULL, NULL);
        if (IS_ERR(seq_type_proc_dir))
                return PTR_ERR(seq_type_proc_dir);
 
@@ -593,7 +593,7 @@ static int __init fid_mod_init(void)
        return 0;
 }
 
-static void __exit fid_mod_exit(void)
+static void __exit fid_exit(void)
 {
 # ifdef HAVE_SERVER_SUPPORT
        fid_server_mod_exit();
@@ -605,9 +605,10 @@ static void __exit fid_mod_exit(void)
        }
 }
 
-MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
-MODULE_DESCRIPTION("Lustre FID Module");
+MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
+MODULE_DESCRIPTION("Lustre File IDentifier");
+MODULE_VERSION(LUSTRE_VERSION_STRING);
 MODULE_LICENSE("GPL");
 
-cfs_module(fid, "0.1.0", fid_mod_init, fid_mod_exit);
-#endif /* __KERNEL__ */
+module_init(fid_init);
+module_exit(fid_exit);