/* Copyright (c) 2013 Red Hat, Inc. This file is part of GlusterFS. This file is licensed to you under your choice of the GNU Lesser General Public License, version 3 or any later version (LGPLv3 or later), or the GNU General Public License, version 2 (GPLv2), in all cases as published by the Free Software Foundation. */ #ifndef _CONFIG_H #define _CONFIG_H #include "config.h" #endif #include #include "call-stub.h" #include "defaults.h" #include "xlator.h" #include "glfs.h" #include "glfs-internal.h" #include "run.h" #include "common-utils.h" #include "syncop.h" #include "syscall.h" #include "compat-errno.h" #include "fdl.h" #include "jbr-internal.h" #include "jbr-messages.h" #define JBR_FLUSH_INTERVAL 5 enum { /* echo "cluster/jbr-server" | md5sum | cut -c 1-8 */ JBR_SERVER_IPC_BASE = 0x0e2d66a5, JBR_SERVER_TERM_RANGE, JBR_SERVER_OPEN_TERM, JBR_SERVER_NEXT_ENTRY }; /* * Need to declare jbr_lk_call_dispatch as jbr_lk_continue and * * jbr_lk_perform_local_op call it, before code is generated. * */ int32_t jbr_lk_call_dispatch (call_frame_t *frame, xlator_t *this, int *op_errno, fd_t *fd, int32_t cmd, struct gf_flock *lock, dict_t *xdata); int32_t jbr_lk_dispatch (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd, struct gf_flock *lock, dict_t *xdata); int32_t jbr_ipc_call_dispatch (call_frame_t *frame, xlator_t *this, int *op_errno, int32_t op, dict_t *xdata); int32_t jbr_ipc_complete (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata); /* Used to check the quorum of acks received after the fop * confirming the status of the fop on all the brick processes * for this particular subvolume */ gf_boolean_t fop_quorum_check (xlator_t *this, double n_children, double current_state) { jbr_private_t *priv = NULL; gf_boolean_t result = _gf_false; double required = 0; double current = 0; GF_VALIDATE_OR_GOTO ("jbr", this, out); priv = this->private; GF_VALIDATE_OR_GOTO (this->name, priv, out); required = n_children * priv->quorum_pct; /* * Before performing the fop on the leader, we need to check, * if there is any merit in performing the fop on the leader. * In a case, where even a successful write on the leader, will * not meet quorum, there is no point in trying the fop on the * leader. * When this function is called after the leader has tried * performing the fop, this check will calculate quorum taking into * account the status of the fop on the leader. If the leader's * op_ret was -1, the complete function would account that by * decrementing successful_acks by 1 */ current = current_state * 100.0; if (current < required) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_QUORUM_NOT_MET, "Quorum not met. quorum_pct = %f " "Current State = %f, Required State = %f", priv->quorum_pct, current, required); } else result = _gf_true; out: return result; } jbr_inode_ctx_t * jbr_get_inode_ctx (xlator_t *this, inode_t *inode) { uint64_t ctx_int = 0LL; jbr_inode_ctx_t *ctx_ptr; if (__inode_ctx_get(inode, this, &ctx_int) == 0) { ctx_ptr = (jbr_inode_ctx_t *)(long)ctx_int; } else { ctx_ptr = GF_CALLOC (1, sizeof(*ctx_ptr), gf_mt_jbr_inode_ctx_t); if (ctx_ptr) { ctx_int = (uint64_t)(long)ctx_ptr; if (__inode_ctx_set(inode, this, &ctx_int) == 0) { LOCK_INIT(&ctx_ptr->lock); INIT_LIST_HEAD(&ctx_ptr->aqueue); INIT_LIST_HEAD(&ctx_ptr->pqueue); } else { GF_FREE(ctx_ptr); ctx_ptr = NULL; } } } return ctx_ptr; } jbr_fd_ctx_t * jbr_get_fd_ctx (xlator_t *this, fd_t *fd) { uint64_t ctx_int = 0LL; jbr_fd_ctx_t *ctx_ptr; if (__fd_ctx_get(fd, this, &ctx_int) == 0) { ctx_ptr = (jbr_fd_ctx_t *)(long)ctx_int; } else { ctx_ptr = GF_CALLOC (1, sizeof(*ctx_ptr), gf_mt_jbr_fd_ctx_t); if (ctx_ptr) { if (__fd_ctx_set(fd, this, (uint64_t)ctx_ptr) == 0) { INIT_LIST_HEAD(&ctx_ptr->dirty_list); INIT_LIST_HEAD(&ctx_ptr->fd_list); } else { GF_FREE(ctx_ptr); ctx_ptr = NULL; } } } return ctx_ptr; } void jbr_mark_fd_dirty (xlator_t *this, jbr_local_t *local) { fd_t *fd = local->fd; jbr_fd_ctx_t *ctx_ptr; jbr_dirty_list_t *dirty; jbr_private_t *priv = this->private; /* * TBD: don't do any of this for O_SYNC/O_DIRECT writes. * Unfortunately, that optimization requires that we distinguish * between writev and other "write" calls, saving the original flags * and checking them in the callback. Too much work for too little * gain right now. */ LOCK(&fd->lock); ctx_ptr = jbr_get_fd_ctx(this, fd); dirty = GF_CALLOC(1, sizeof(*dirty), gf_mt_jbr_dirty_t); if (ctx_ptr && dirty) { gf_msg_trace (this->name, 0, "marking fd %p as dirty (%p)", fd, dirty); /* TBD: fill dirty->id from what changelog gave us */ list_add_tail(&dirty->links, &ctx_ptr->dirty_list); if (list_empty(&ctx_ptr->fd_list)) { /* Add a ref so _release doesn't get called. */ ctx_ptr->fd = fd_ref(fd); LOCK(&priv->dirty_lock); list_add_tail (&ctx_ptr->fd_list, &priv->dirty_fds); UNLOCK(&priv->dirty_lock); } } else { gf_msg (this->name, GF_LOG_ERROR, ENOMEM, J_MSG_MEM_ERR, "could not mark %p dirty", fd); if (ctx_ptr) { GF_FREE(ctx_ptr); } if (dirty) { GF_FREE(dirty); } } UNLOCK(&fd->lock); } #define JBR_TERM_XATTR "trusted.jbr.term" #define JBR_INDEX_XATTR "trusted.jbr.index" #define JBR_REP_COUNT_XATTR "trusted.jbr.rep-count" #define RECON_TERM_XATTR "trusted.jbr.recon-term" #define RECON_INDEX_XATTR "trusted.jbr.recon-index" int32_t jbr_leader_checks_and_init (call_frame_t *frame, xlator_t *this, int *op_errno, dict_t *xdata, fd_t *fd) { jbr_local_t *local = NULL; jbr_private_t *priv = NULL; int32_t ret = -1; gf_boolean_t result = _gf_false; int from_leader = _gf_false; int from_recon = _gf_false; GF_VALIDATE_OR_GOTO ("jbr", this, out); priv = this->private; GF_VALIDATE_OR_GOTO (this->name, priv, out); GF_VALIDATE_OR_GOTO (this->name, op_errno, out); GF_VALIDATE_OR_GOTO (this->name, frame, out); /* * Our first goal here is to avoid "split brain surprise" for users who * specify exactly 50% with two- or three-way replication. That means * either a more-than check against half the total replicas or an * at-least check against half of our peers (one less). Of the two, * only an at-least check supports the intuitive use of 100% to mean * all replicas must be present, because "more than 100%" will never * succeed regardless of which count we use. This leaves us with a * slightly non-traditional definition of quorum ("at least X% of peers * not including ourselves") but one that's useful enough to be worth * it. * * Note that n_children and up_children *do* include the local * subvolume, so we need to subtract one in each case. */ if (priv->leader) { result = fop_quorum_check (this, (double)(priv->n_children - 1), (double)(priv->up_children - 1)); if (result == _gf_false) { /* Emulate the AFR client-side-quorum behavior. */ gf_msg (this->name, GF_LOG_ERROR, EROFS, J_MSG_QUORUM_NOT_MET, "Sufficient number of " "subvolumes are not up to meet quorum."); *op_errno = EROFS; goto out; } } else { if (xdata) { from_leader = !!dict_get(xdata, JBR_TERM_XATTR); from_recon = !!dict_get(xdata, RECON_TERM_XATTR) && !!dict_get(xdata, RECON_INDEX_XATTR); } else { from_leader = from_recon = _gf_false; } /* follower/recon path * * just send it to local node * */ if (!from_leader && !from_recon) { *op_errno = EREMOTE; goto out; } } local = mem_get0(this->local_pool); if (!local) { goto out; } if (fd) local->fd = fd_ref(fd); else local->fd = NULL; INIT_LIST_HEAD(&local->qlinks); local->successful_acks = 0; frame->local = local; ret = 0; out: return ret; } int32_t jbr_initialize_xdata_set_attrs (xlator_t *this, dict_t **xdata) { jbr_private_t *priv = NULL; int32_t ret = -1; uint32_t ti = 0; GF_VALIDATE_OR_GOTO ("jbr", this, out); priv = this->private; GF_VALIDATE_OR_GOTO (this->name, priv, out); GF_VALIDATE_OR_GOTO (this->name, xdata, out); if (!*xdata) { *xdata = dict_new(); if (!*xdata) { gf_msg (this->name, GF_LOG_ERROR, ENOMEM, J_MSG_MEM_ERR, "failed to allocate xdata"); goto out; } } if (dict_set_int32(*xdata, JBR_TERM_XATTR, priv->current_term) != 0) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_DICT_FLR, "failed to set jbr-term"); goto out; } LOCK(&priv->index_lock); ti = ++(priv->index); UNLOCK(&priv->index_lock); if (dict_set_int32(*xdata, JBR_INDEX_XATTR, ti) != 0) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_DICT_FLR, "failed to set index"); goto out; } ret = 0; out: return ret; } int32_t jbr_remove_from_queue (call_frame_t *frame, xlator_t *this) { int32_t ret = -1; jbr_inode_ctx_t *ictx = NULL; jbr_local_t *local = NULL; jbr_local_t *next = NULL; GF_VALIDATE_OR_GOTO ("jbr", this, out); GF_VALIDATE_OR_GOTO (this->name, frame, out); local = frame->local; GF_VALIDATE_OR_GOTO (this->name, local, out); if (local->qlinks.next != &local->qlinks) { list_del(&local->qlinks); ictx = jbr_get_inode_ctx(this, local->fd->inode); if (ictx) { LOCK(&ictx->lock); if (ictx->pending) { /* * TBD: dequeue *all* non-conflicting * reqs * * With the stub implementation there * can only be one request active at a * time (zero here) so it's not an * issue. In a real implementation * there might still be other active * requests to check against, and * multiple pending requests that could * continue. */ gf_msg_debug (this->name, 0, "unblocking next request"); --(ictx->pending); next = list_entry (ictx->pqueue.next, jbr_local_t, qlinks); list_del(&next->qlinks); list_add_tail(&next->qlinks, &ictx->aqueue); call_resume(next->qstub); } else { --(ictx->active); } UNLOCK(&ictx->lock); } } ret = 0; out: return ret; } int32_t jbr_lk_complete (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct gf_flock *flock, dict_t *xdata) { int32_t ret = -1; jbr_private_t *priv = NULL; jbr_local_t *local = NULL; gf_boolean_t result = _gf_false; GF_VALIDATE_OR_GOTO ("jbr", this, err); priv = this->private; GF_VALIDATE_OR_GOTO (this->name, priv, err); GF_VALIDATE_OR_GOTO (this->name, frame, err); local = frame->local; GF_VALIDATE_OR_GOTO (this->name, local, err); GF_VALIDATE_OR_GOTO (this->name, flock, err); GF_VALIDATE_OR_GOTO (this->name, xdata, err); /* * Remove from queue for unlock operation only * * For lock operation, it will be done in fan-in * */ if (flock->l_type == F_UNLCK) { ret = jbr_remove_from_queue (frame, this); if (ret) goto err; } /* * On a follower, unwind with the op_ret and op_errno. On a * * leader, if the fop is a locking fop, and its a failure, * * send fail, else call stub which will dispatch the fop to * * the followers. * * * * If the fop is a unlocking fop, check quorum. If quorum * * is met, then send success. Else Rollback on leader, * * followed by followers, and then send -ve ack to client. * */ if (priv->leader) { /* Increase the successful acks if it's a success. */ LOCK(&frame->lock); if (op_ret != -1) (local->successful_acks)++; UNLOCK(&frame->lock); if (flock->l_type == F_UNLCK) { result = fop_quorum_check (this, (double)priv->n_children, (double)local->successful_acks); if (result == _gf_false) { op_ret = -1; op_errno = EROFS; gf_msg (this->name, GF_LOG_ERROR, EROFS, J_MSG_QUORUM_NOT_MET, "Quorum is not met. " "The operation has failed."); /* TODO: PERFORM UNLOCK ROLLBACK ON LEADER * * FOLLOWED BY FOLLOWERS. */ } else { op_ret = 0; op_errno = 0; } fd_unref(local->fd); STACK_UNWIND_STRICT (lk, frame, op_ret, op_errno, flock, xdata); } else { if (op_ret == -1) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_LOCK_FAILURE, "The lock operation failed on " "the leader."); fd_unref(local->fd); STACK_UNWIND_STRICT (lk, frame, op_ret, op_errno, flock, xdata); } else { if (!local->stub) { goto err; } call_resume(local->stub); } } } else { fd_unref(local->fd); STACK_UNWIND_STRICT (lk, frame, op_ret, op_errno, flock, xdata); } return 0; err: if (local) { if (local->stub) { call_stub_destroy(local->stub); } if (local->qstub) { call_stub_destroy(local->qstub); } if (local->fd) { fd_unref(local->fd); } mem_put(local); } STACK_UNWIND_STRICT (lk, frame, -1, op_errno, flock, xdata); return 0; } int32_t jbr_lk_fan_in (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct gf_flock *flock, dict_t *xdata) { uint8_t call_count = -1; int32_t ret = -1; gf_boolean_t result = _gf_false; jbr_local_t *local = NULL; jbr_private_t *priv = NULL; GF_VALIDATE_OR_GOTO ("jbr", this, out); GF_VALIDATE_OR_GOTO (this->name, frame, out); priv = this->private; local = frame->local; GF_VALIDATE_OR_GOTO (this->name, priv, out); GF_VALIDATE_OR_GOTO (this->name, local, out); gf_msg_trace (this->name, 0, "op_ret = %d, op_errno = %d\n", op_ret, op_errno); LOCK(&frame->lock); call_count = --(local->call_count); if (op_ret != -1) { /* Increment the number of successful acks * * received for the operation. * */ (local->successful_acks)++; local->successful_op_ret = op_ret; } gf_msg_debug (this->name, 0, "succ_acks = %d, op_ret = %d, op_errno = %d\n", op_ret, op_errno, local->successful_acks); UNLOCK(&frame->lock); if (call_count == 0) { /* * If the fop is a locking fop, then check quorum. If quorum * * is met, send successful ack to the client. If quorum is * * not met, then rollback locking on followers, followed by * * rollback of locking on leader, and then sending -ve ack * * to the client. * * * * If the fop is a unlocking fop, then call stub. * */ if (flock->l_type == F_UNLCK) { call_resume(local->stub); } else { /* * Remove from queue for locking fops, for unlocking * * fops, it is taken care of in jbr_lk_complete * */ ret = jbr_remove_from_queue (frame, this); if (ret) goto out; fd_unref(local->fd); result = fop_quorum_check (this, (double)priv->n_children, (double)local->successful_acks); if (result == _gf_false) { gf_msg (this->name, GF_LOG_ERROR, EROFS, J_MSG_QUORUM_NOT_MET, "Didn't receive enough acks to meet " "quorum. Failing the locking " "operation and initiating rollback on " "followers and the leader " "respectively."); /* TODO: PERFORM ROLLBACK OF LOCKING ON * FOLLOWERS, FOLLOWED BY ROLLBACK ON * LEADER. */ STACK_UNWIND_STRICT (lk, frame, -1, EROFS, flock, xdata); } else { STACK_UNWIND_STRICT (lk, frame, 0, 0, flock, xdata); } } } ret = 0; out: return ret; } /* * Called from leader for locking fop, being writen as a separate * * function so as to support queues. * */ int32_t jbr_perform_lk_on_leader (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata) { int32_t ret = -1; GF_VALIDATE_OR_GOTO ("jbr", this, out); GF_VALIDATE_OR_GOTO (this->name, frame, out); GF_VALIDATE_OR_GOTO (this->name, flock, out); GF_VALIDATE_OR_GOTO (this->name, fd, out); STACK_WIND (frame, jbr_lk_complete, FIRST_CHILD(this), FIRST_CHILD(this)->fops->lk, fd, cmd, flock, xdata); ret = 0; out: return ret; } int32_t jbr_lk_perform_local_op (call_frame_t *frame, xlator_t *this, int *op_errno, fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata) { int32_t ret = -1; jbr_local_t *local = NULL; GF_VALIDATE_OR_GOTO ("jbr", this, out); GF_VALIDATE_OR_GOTO (this->name, frame, out); local = frame->local; GF_VALIDATE_OR_GOTO (this->name, local, out); GF_VALIDATE_OR_GOTO (this->name, fd, out); GF_VALIDATE_OR_GOTO (this->name, op_errno, out); GF_VALIDATE_OR_GOTO (this->name, flock, out); /* * Check if the fop is a locking fop or unlocking fop, and * handle it accordingly. If it is a locking fop, take the * lock on leader first, and then send it to the followers. * If it is a unlocking fop, unlock the followers first, * and then on meeting quorum perform the unlock on the leader. */ if (flock->l_type == F_UNLCK) { ret = jbr_lk_call_dispatch (frame, this, op_errno, fd, cmd, flock, xdata); if (ret) goto out; } else { jbr_inode_ctx_t *ictx = jbr_get_inode_ctx(this, fd->inode); if (!ictx) { *op_errno = EIO; goto out; } LOCK(&ictx->lock); if (ictx->active) { gf_msg_debug (this->name, 0, "queuing request due to conflict"); local->qstub = fop_lk_stub (frame, jbr_perform_lk_on_leader, fd, cmd, flock, xdata); if (!local->qstub) { UNLOCK(&ictx->lock); goto out; } list_add_tail(&local->qlinks, &ictx->pqueue); ++(ictx->pending); UNLOCK(&ictx->lock); ret = 0; goto out; } else { list_add_tail(&local->qlinks, &ictx->aqueue); ++(ictx->active); } UNLOCK(&ictx->lock); ret = jbr_perform_lk_on_leader (frame, this, fd, cmd, flock, xdata); } ret = 0; out: return ret; } int32_t jbr_lk_continue (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata) { int32_t ret = -1; jbr_local_t *local = NULL; jbr_private_t *priv = NULL; GF_VALIDATE_OR_GOTO ("jbr", this, out); GF_VALIDATE_OR_GOTO (this->name, frame, out); priv = this->private; local = frame->local; GF_VALIDATE_OR_GOTO (this->name, priv, out); GF_VALIDATE_OR_GOTO (this->name, local, out); GF_VALIDATE_OR_GOTO (this->name, flock, out); GF_VALIDATE_OR_GOTO (this->name, fd, out); GF_VALIDATE_OR_GOTO (this->name, xdata, out); /* * If it's a locking fop, then call dispatch to followers * * If it's a unlock fop, then perform the unlock operation * */ if (flock->l_type == F_UNLCK) { STACK_WIND (frame, jbr_lk_complete, FIRST_CHILD(this), FIRST_CHILD(this)->fops->lk, fd, cmd, flock, xdata); } else { /* * Directly call jbr_lk_dispatch instead of appending * * in queue, which is done at jbr_lk_perform_local_op * * for locking fops * */ ret = jbr_lk_dispatch (frame, this, fd, cmd, flock, xdata); if (ret) { STACK_UNWIND_STRICT (lk, frame, -1, 0, flock, xdata); goto out; } } ret = 0; out: return ret; } uint8_t jbr_count_up_kids (jbr_private_t *priv) { uint8_t retval = 0; uint8_t i; for (i = 0; i < priv->n_children; ++i) { if (priv->kid_state & (1 << i)) { ++retval; } } return retval; } /* * The fsync machinery looks a lot like that for any write call, but there are * some important differences that are easy to miss. First, we don't care * about the xdata that shows whether the call came from a leader or * reconciliation process. If we're the leader we fan out; if we're not we * don't. Second, we don't wait for followers before we issue the local call. * The code generation system could be updated to handle this, and still might * if we need to implement other "almost identical" paths (e.g. for open), but * a copy is more readable as long as it's just one. */ int32_t jbr_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata) { jbr_local_t *local = frame->local; gf_boolean_t unwind; LOCK(&frame->lock); unwind = !--(local->call_count); UNLOCK(&frame->lock); if (unwind) { STACK_UNWIND_STRICT (fsync, frame, op_ret, op_errno, prebuf, postbuf, xdata); } return 0; } int32_t jbr_fsync_local_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata) { jbr_dirty_list_t *dirty; jbr_dirty_list_t *dtmp; jbr_local_t *local = frame->local; list_for_each_entry_safe (dirty, dtmp, &local->qlinks, links) { gf_msg_trace (this->name, 0, "sending post-op on %p (%p)", local->fd, dirty); GF_FREE(dirty); } return jbr_fsync_cbk (frame, cookie, this, op_ret, op_errno, prebuf, postbuf, xdata); } int32_t jbr_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags, dict_t *xdata) { jbr_private_t *priv = this->private; jbr_local_t *local; uint64_t ctx_int = 0LL; jbr_fd_ctx_t *ctx_ptr; xlator_list_t *trav; local = mem_get0(this->local_pool); if (!local) { STACK_UNWIND_STRICT(fsync, frame, -1, ENOMEM, NULL, NULL, xdata); return 0; } INIT_LIST_HEAD(&local->qlinks); frame->local = local; /* Move the dirty list from the fd to the fsync request. */ LOCK(&fd->lock); if (__fd_ctx_get(fd, this, &ctx_int) == 0) { ctx_ptr = (jbr_fd_ctx_t *)(long)ctx_int; list_splice_init (&ctx_ptr->dirty_list, &local->qlinks); } UNLOCK(&fd->lock); /* Issue the local call. */ local->call_count = priv->leader ? priv->n_children : 1; STACK_WIND (frame, jbr_fsync_local_cbk, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fsync, fd, flags, xdata); /* Issue remote calls if we're the leader. */ if (priv->leader) { for (trav = this->children->next; trav; trav = trav->next) { STACK_WIND (frame, jbr_fsync_cbk, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fsync, fd, flags, xdata); } } return 0; } int32_t jbr_getxattr_special (call_frame_t *frame, xlator_t *this, loc_t *loc, const char *name, dict_t *xdata) { dict_t *result; jbr_private_t *priv = this->private; if (!priv->leader) { STACK_UNWIND_STRICT (getxattr, frame, -1, EREMOTE, NULL, NULL); return 0; } if (!name || (strcmp(name, JBR_REP_COUNT_XATTR) != 0)) { STACK_WIND_TAIL (frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->getxattr, loc, name, xdata); return 0; } result = dict_new(); if (!result) { goto dn_failed; } priv->up_children = jbr_count_up_kids(this->private); if (dict_set_uint32(result, JBR_REP_COUNT_XATTR, priv->up_children) != 0) { goto dsu_failed; } STACK_UNWIND_STRICT (getxattr, frame, 0, 0, result, NULL); dict_unref(result); return 0; dsu_failed: dict_unref(result); dn_failed: STACK_UNWIND_STRICT (getxattr, frame, -1, ENOMEM, NULL, NULL); return 0; } void jbr_flush_fd (xlator_t *this, jbr_fd_ctx_t *fd_ctx) { jbr_dirty_list_t *dirty; jbr_dirty_list_t *dtmp; list_for_each_entry_safe (dirty, dtmp, &fd_ctx->dirty_list, links) { gf_msg_trace (this->name, 0, "sending post-op on %p (%p)", fd_ctx->fd, dirty); GF_FREE(dirty); } INIT_LIST_HEAD(&fd_ctx->dirty_list); } void * jbr_flush_thread (void *ctx) { xlator_t *this = ctx; jbr_private_t *priv = this->private; struct list_head dirty_fds; jbr_fd_ctx_t *fd_ctx; jbr_fd_ctx_t *fd_tmp; int ret; for (;;) { /* * We have to be very careful to avoid lock inversions here, so * we can't just hold priv->dirty_lock while we take and * release locks for each fd. Instead, we only hold dirty_lock * at the beginning of each iteration, as we (effectively) make * a copy of the current list head and then clear the original. * This leads to four scenarios for adding the first entry to * an fd and potentially putting it on the global list. * * (1) While we're asleep. No lock contention, it just gets * added and will be processed on the next iteration. * * (2) After we've made a local copy, but before we've started * processing that fd. The new entry will be added to the * fd (under its lock), and we'll process it on the current * iteration. * * (3) While we're processing the fd. They'll block on the fd * lock, then see that the list is empty and put it on the * global list. We'll process it here on the next * iteration. * * (4) While we're working, but after we've processed that fd. * Same as (1) as far as that fd is concerned. */ INIT_LIST_HEAD(&dirty_fds); LOCK(&priv->dirty_lock); list_splice_init(&priv->dirty_fds, &dirty_fds); UNLOCK(&priv->dirty_lock); list_for_each_entry_safe (fd_ctx, fd_tmp, &dirty_fds, fd_list) { ret = syncop_fsync(FIRST_CHILD(this), fd_ctx->fd, 0, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_WARNING, 0, J_MSG_SYS_CALL_FAILURE, "failed to fsync %p (%d)", fd_ctx->fd, -ret); } LOCK(&fd_ctx->fd->lock); jbr_flush_fd(this, fd_ctx); list_del_init(&fd_ctx->fd_list); UNLOCK(&fd_ctx->fd->lock); fd_unref(fd_ctx->fd); } sleep(JBR_FLUSH_INTERVAL); } return NULL; } int32_t jbr_get_changelog_dir (xlator_t *this, char **cl_dir_p) { xlator_t *cl_xl; /* Find our changelog translator. */ cl_xl = this; while (cl_xl) { if (strcmp(cl_xl->type, "features/changelog") == 0) { break; } cl_xl = cl_xl->children->xlator; } if (!cl_xl) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_INIT_FAIL, "failed to find changelog translator"); return ENOENT; } /* Find the actual changelog directory. */ if (dict_get_str(cl_xl->options, "changelog-dir", cl_dir_p) != 0) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_INIT_FAIL, "failed to find changelog-dir for %s", cl_xl->name); return ENODATA; } return 0; } void jbr_get_terms (call_frame_t *frame, xlator_t *this) { int32_t op_errno = 0; char *cl_dir = NULL; int32_t term_first = -1; int32_t term_contig = -1; int32_t term_last = -1; int term_num = 0; char *probe_str = NULL; dict_t *my_xdata = NULL; DIR *fp = NULL; struct dirent *entry = NULL; struct dirent scratch[2] = {{0,},}; op_errno = jbr_get_changelog_dir(this, &cl_dir); if (op_errno) { goto err; /* Error was already logged. */ } op_errno = ENODATA; /* Most common error after this. */ fp = sys_opendir (cl_dir); if (!fp) { op_errno = errno; goto err; } /* Find first and last terms. */ for (;;) { errno = 0; entry = sys_readdir (fp, scratch); if (!entry || errno != 0) { if (errno != 0) { op_errno = errno; goto err; } break; } if (fnmatch("TERM.*", entry->d_name, FNM_PATHNAME) != 0) { continue; } /* +5 points to the character after the period */ term_num = atoi(entry->d_name+5); gf_msg (this->name, GF_LOG_INFO, 0, J_MSG_GENERIC, "%s => %d", entry->d_name, term_num); if (term_num < 0) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_INVALID, "invalid term file name %s", entry->d_name); op_errno = EINVAL; goto err; } if ((term_first < 0) || (term_first > term_num)) { term_first = term_num; } if ((term_last < 0) || (term_last < term_num)) { term_last = term_num; } } if ((term_first < 0) || (term_last < 0)) { /* TBD: are we *sure* there should always be at least one? */ gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_NO_DATA, "no terms found"); op_errno = EINVAL; goto err; } (void) sys_closedir (fp); fp = NULL; /* * Find term_contig, which is the earliest term for which there are * no gaps between it and term_last. */ for (term_contig = term_last; term_contig > 0; --term_contig) { if (gf_asprintf(&probe_str, "%s/TERM.%d", cl_dir, term_contig-1) <= 0) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_MEM_ERR, "failed to format term %d", term_contig-1); goto err; } if (sys_access(probe_str, F_OK) != 0) { GF_FREE(probe_str); break; } GF_FREE(probe_str); } gf_msg (this->name, GF_LOG_INFO, 0, J_MSG_GENERIC, "found terms %d-%d (%d)", term_first, term_last, term_contig); /* Return what we've found */ my_xdata = dict_new(); if (!my_xdata) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_MEM_ERR, "failed to allocate reply dictionary"); goto err; } if (dict_set_int32(my_xdata, "term-first", term_first) != 0) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_DICT_FLR, "failed to set term-first"); goto err; } if (dict_set_int32(my_xdata, "term-contig", term_contig) != 0) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_DICT_FLR, "failed to set term-contig"); goto err; } if (dict_set_int32(my_xdata, "term-last", term_last) != 0) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_DICT_FLR, "failed to set term-last"); goto err; } /* Finally! */ STACK_UNWIND_STRICT (ipc, frame, 0, 0, my_xdata); dict_unref(my_xdata); return; err: if (fp) { (void) sys_closedir (fp); } if (my_xdata) { dict_unref(my_xdata); } STACK_UNWIND_STRICT (ipc, frame, -1, op_errno, NULL); } long get_entry_count (xlator_t *this, int fd) { struct stat buf; long min; /* last entry not known to be empty */ long max; /* first entry known to be empty */ long curr; char entry[CHANGELOG_ENTRY_SIZE]; if (sys_fstat (fd, &buf) < 0) { return -1; } min = 0; max = buf.st_size / CHANGELOG_ENTRY_SIZE; while ((min+1) < max) { curr = (min + max) / 2; if (sys_lseek(fd, curr*CHANGELOG_ENTRY_SIZE, SEEK_SET) < 0) { return -1; } if (sys_read(fd, entry, sizeof(entry)) != sizeof(entry)) { return -1; } if ((entry[0] == '_') && (entry[1] == 'P')) { min = curr; } else { max = curr; } } if (sys_lseek(fd, 0, SEEK_SET) < 0) { gf_msg (this->name, GF_LOG_WARNING, 0, J_MSG_SYS_CALL_FAILURE, "failed to reset offset"); } return max; } void jbr_open_term (call_frame_t *frame, xlator_t *this, dict_t *xdata) { int32_t op_errno; char *cl_dir; char *term; char *path = NULL; jbr_private_t *priv = this->private; op_errno = jbr_get_changelog_dir(this, &cl_dir); if (op_errno) { goto err; } if (dict_get_str(xdata, "term", &term) != 0) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_NO_DATA, "missing term"); op_errno = ENODATA; goto err; } if (gf_asprintf(&path, "%s/TERM.%s", cl_dir, term) < 0) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_MEM_ERR, "failed to construct path"); op_errno = ENOMEM; goto err; } if (priv->term_fd >= 0) { sys_close (priv->term_fd); } priv->term_fd = open(path, O_RDONLY); if (priv->term_fd < 0) { op_errno = errno; gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_SYS_CALL_FAILURE, "failed to open term file"); goto err; } priv->term_total = get_entry_count(this, priv->term_fd); if (priv->term_total < 0) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_NO_DATA, "failed to get entry count"); sys_close (priv->term_fd); priv->term_fd = -1; op_errno = EIO; goto err; } priv->term_read = 0; /* Success! */ STACK_UNWIND_STRICT (ipc, frame, 0, 0, NULL); GF_FREE (path); return; err: STACK_UNWIND_STRICT (ipc, frame, -1, op_errno, NULL); GF_FREE (path); } void jbr_next_entry (call_frame_t *frame, xlator_t *this) { int32_t op_errno = ENOMEM; jbr_private_t *priv = this->private; ssize_t nbytes; dict_t *my_xdata; if (priv->term_fd < 0) { op_errno = EBADFD; goto err; } if (priv->term_read >= priv->term_total) { op_errno = ENODATA; goto err; } nbytes = sys_read (priv->term_fd, priv->term_buf, CHANGELOG_ENTRY_SIZE); if (nbytes < CHANGELOG_ENTRY_SIZE) { if (nbytes < 0) { op_errno = errno; gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_SYS_CALL_FAILURE, "error reading next entry: %s", strerror(errno)); } else { op_errno = EIO; gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_SYS_CALL_FAILURE, "got %zd/%d bytes for next entry", nbytes, CHANGELOG_ENTRY_SIZE); } goto err; } ++(priv->term_read); my_xdata = dict_new(); if (!my_xdata) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_MEM_ERR, "failed to allocate reply xdata"); goto err; } if (dict_set_static_bin(my_xdata, "data", priv->term_buf, CHANGELOG_ENTRY_SIZE) != 0) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_DICT_FLR, "failed to assign reply xdata"); goto err; } STACK_UNWIND_STRICT (ipc, frame, 0, 0, my_xdata); dict_unref(my_xdata); return; err: STACK_UNWIND_STRICT (ipc, frame, -1, op_errno, NULL); } int32_t jbr_ipc_fan_in (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata) { jbr_local_t *local = NULL; int32_t ret = -1; uint8_t call_count; GF_VALIDATE_OR_GOTO ("jbr", this, out); GF_VALIDATE_OR_GOTO (this->name, frame, out); local = frame->local; GF_VALIDATE_OR_GOTO (this->name, local, out); gf_msg_trace (this->name, 0, "op_ret = %d, op_errno = %d\n", op_ret, op_errno); LOCK(&frame->lock); call_count = --(local->call_count); UNLOCK(&frame->lock); if (call_count == 0) { #if defined(JBR_CG_QUEUE) ret = jbr_remove_from_queue (frame, this); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_GENERIC, "Failed to remove from queue."); } #endif /* * Unrefing the reference taken in continue() or complete() * */ dict_unref (local->xdata); STACK_DESTROY (frame->root); } ret = 0; out: return ret; } int32_t jbr_ipc_complete (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata) { jbr_local_t *local = NULL; GF_VALIDATE_OR_GOTO ("jbr", this, out); GF_VALIDATE_OR_GOTO (this->name, frame, out); local = frame->local; GF_VALIDATE_OR_GOTO (this->name, local, out); jbr_ipc_call_dispatch (frame, this, &op_errno, FDL_IPC_JBR_SERVER_ROLLBACK, local->xdata); out: return 0; } int32_t jbr_ipc (call_frame_t *frame, xlator_t *this, int32_t op, dict_t *xdata) { switch (op) { case JBR_SERVER_TERM_RANGE: jbr_get_terms(frame, this); break; case JBR_SERVER_OPEN_TERM: jbr_open_term(frame, this, xdata); break; case JBR_SERVER_NEXT_ENTRY: jbr_next_entry(frame, this); break; case FDL_IPC_JBR_SERVER_ROLLBACK: /* * Just send the fop down to fdl. Need not * * dispatch it to other bricks in the sub- * * volume, as it will be done where the op * * has failed. * */ default: STACK_WIND_TAIL (frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->ipc, op, xdata); } return 0; } #pragma generate int32_t jbr_forget (xlator_t *this, inode_t *inode) { uint64_t ctx = 0LL; if ((inode_ctx_del(inode, this, &ctx) == 0) && ctx) { GF_FREE((void *)(long)ctx); } return 0; } int32_t jbr_release (xlator_t *this, fd_t *fd) { uint64_t ctx = 0LL; if ((fd_ctx_del(fd, this, &ctx) == 0) && ctx) { GF_FREE((void *)(long)ctx); } return 0; } struct xlator_cbks cbks = { .forget = jbr_forget, .release = jbr_release, }; int jbr_reconfigure (xlator_t *this, dict_t *options) { jbr_private_t *priv = this->private; GF_OPTION_RECONF ("leader", priv->config_leader, options, bool, err); GF_OPTION_RECONF ("quorum-percent", priv->quorum_pct, options, percent, err); gf_msg (this->name, GF_LOG_INFO, 0, J_MSG_GENERIC, "reconfigure called, config_leader = %d, quorum_pct = %.1f\n", priv->leader, priv->quorum_pct); priv->leader = priv->config_leader; return 0; err: return -1; } int jbr_get_child_index (xlator_t *this, xlator_t *kid) { xlator_list_t *trav; int retval = -1; for (trav = this->children; trav; trav = trav->next) { ++retval; if (trav->xlator == kid) { return retval; } } return -1; } /* * Child notify handling is unreasonably FUBAR. Sometimes we'll get a * CHILD_DOWN for a protocol/client child before we ever got a CHILD_UP for it. * Other times we won't. Because it's effectively random (probably racy), we * can't just maintain a count. We actually have to keep track of the state * for each child separately, to filter out the bogus CHILD_DOWN events, and * then generate counts on demand. */ int jbr_notify (xlator_t *this, int event, void *data, ...) { jbr_private_t *priv = this->private; int index = -1; int ret = -1; gf_boolean_t result = _gf_false; gf_boolean_t relevant = _gf_false; switch (event) { case GF_EVENT_CHILD_UP: index = jbr_get_child_index(this, data); if (index >= 0) { /* Check if the child was previously down * and it's not a false CHILD_UP */ if (!(priv->kid_state & (1 << index))) { relevant = _gf_true; } priv->kid_state |= (1 << index); priv->up_children = jbr_count_up_kids(priv); gf_msg (this->name, GF_LOG_INFO, 0, J_MSG_GENERIC, "got CHILD_UP for %s, now %u kids", ((xlator_t *)data)->name, priv->up_children); if (!priv->config_leader && (priv->up_children > 1)) { priv->leader = _gf_false; } /* If it's not relevant, or we have already * * sent CHILD_UP just break */ if (!relevant || priv->child_up) break; /* If it's not a leader, just send the notify up */ if (!priv->leader) { ret = default_notify(this, event, data); if (!ret) priv->child_up = _gf_true; break; } result = fop_quorum_check (this, (double)(priv->n_children - 1), (double)(priv->up_children - 1)); if (result == _gf_false) { gf_msg (this->name, GF_LOG_INFO, 0, J_MSG_GENERIC, "Not enough children " "are up to meet quorum. Waiting to " "send CHILD_UP from leader"); } else { gf_msg (this->name, GF_LOG_INFO, 0, J_MSG_GENERIC, "Enough children are up " "to meet quorum. Sending CHILD_UP " "from leader"); ret = default_notify(this, event, data); if (!ret) priv->child_up = _gf_true; } } break; case GF_EVENT_CHILD_DOWN: index = jbr_get_child_index(this, data); if (index >= 0) { /* Check if the child was previously up * and it's not a false CHILD_DOWN */ if (priv->kid_state & (1 << index)) { relevant = _gf_true; } priv->kid_state &= ~(1 << index); priv->up_children = jbr_count_up_kids(priv); gf_msg (this->name, GF_LOG_INFO, 0, J_MSG_GENERIC, "got CHILD_DOWN for %s, now %u kids", ((xlator_t *)data)->name, priv->up_children); if (!priv->config_leader && (priv->up_children < 2) && relevant) { priv->leader = _gf_true; } /* If it's not relevant, or we have already * * sent CHILD_DOWN just break */ if (!relevant || !priv->child_up) break; /* If it's not a leader, just break coz we shouldn't * * propagate the failure from the failure till it * * itself goes down * */ if (!priv->leader) { break; } result = fop_quorum_check (this, (double)(priv->n_children - 1), (double)(priv->up_children - 1)); if (result == _gf_false) { gf_msg (this->name, GF_LOG_INFO, 0, J_MSG_GENERIC, "Enough children are " "to down to fail quorum. " "Sending CHILD_DOWN from leader"); ret = default_notify(this, event, data); if (!ret) priv->child_up = _gf_false; } else { gf_msg (this->name, GF_LOG_INFO, 0, J_MSG_GENERIC, "Not enough children " "are down to fail quorum. Waiting to " "send CHILD_DOWN from leader"); } } break; default: ret = default_notify(this, event, data); } return ret; } int32_t mem_acct_init (xlator_t *this) { int ret = -1; GF_VALIDATE_OR_GOTO ("jbr", this, out); ret = xlator_mem_acct_init (this, gf_mt_jbr_end + 1); if (ret != 0) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_MEM_ERR, "Memory accounting init" "failed"); return ret; } out: return ret; } void jbr_deallocate_priv (jbr_private_t *priv) { if (!priv) { return; } GF_FREE(priv); } int32_t jbr_init (xlator_t *this) { xlator_list_t *remote; xlator_list_t *local; jbr_private_t *priv = NULL; xlator_list_t *trav; pthread_t kid; extern xlator_t global_xlator; glusterfs_ctx_t *oldctx = global_xlator.ctx; /* * Any fop that gets special treatment has to be patched in here, * because the compiled-in table is produced by the code generator and * only contains generated functions. Note that we have to go through * this->fops because of some dynamic-linking strangeness; modifying * the static table doesn't work. */ this->fops->getxattr = jbr_getxattr_special; this->fops->fsync = jbr_fsync; local = this->children; if (!local) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_NO_DATA, "no local subvolume"); goto err; } remote = local->next; if (!remote) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_NO_DATA, "no remote subvolumes"); goto err; } this->local_pool = mem_pool_new (jbr_local_t, 128); if (!this->local_pool) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_MEM_ERR, "failed to create jbr_local_t pool"); goto err; } priv = GF_CALLOC (1, sizeof(*priv), gf_mt_jbr_private_t); if (!priv) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_MEM_ERR, "could not allocate priv"); goto err; } for (trav = this->children; trav; trav = trav->next) { ++(priv->n_children); } LOCK_INIT(&priv->dirty_lock); LOCK_INIT(&priv->index_lock); INIT_LIST_HEAD(&priv->dirty_fds); priv->term_fd = -1; this->private = priv; GF_OPTION_INIT ("leader", priv->config_leader, bool, err); GF_OPTION_INIT ("quorum-percent", priv->quorum_pct, percent, err); priv->leader = priv->config_leader; priv->child_up = _gf_false; if (pthread_create(&kid, NULL, jbr_flush_thread, this) != 0) { gf_msg (this->name, GF_LOG_ERROR, 0, J_MSG_SYS_CALL_FAILURE, "could not start flush thread"); /* TBD: treat this as a fatal error? */ } /* * Calling glfs_new changes old->ctx, even if THIS still points * to global_xlator. That causes problems later in the main * thread, when gf_log_dump_graph tries to use the FILE after * we've mucked with it and gets a segfault in __fprintf_chk. * We can avoid all that by undoing the damage before we * continue. */ global_xlator.ctx = oldctx; return 0; err: jbr_deallocate_priv(priv); return -1; } void jbr_fini (xlator_t *this) { jbr_deallocate_priv(this->private); } class_methods_t class_methods = { .init = jbr_init, .fini = jbr_fini, .reconfigure = jbr_reconfigure, .notify = jbr_notify, }; struct volume_options options[] = { { .key = {"leader"}, .type = GF_OPTION_TYPE_BOOL, .default_value = "false", .description = "Start in the leader role. This is only for " "bootstrapping the code, and should go away when we " "have real leader election." }, { .key = {"vol-name"}, .type = GF_OPTION_TYPE_STR, .description = "volume name" }, { .key = {"my-name"}, .type = GF_OPTION_TYPE_STR, .description = "brick name in form of host:/path" }, { .key = {"etcd-servers"}, .type = GF_OPTION_TYPE_STR, .description = "list of comma separated etc servers" }, { .key = {"subvol-uuid"}, .type = GF_OPTION_TYPE_STR, .description = "UUID for this JBR (sub)volume" }, { .key = {"quorum-percent"}, .type = GF_OPTION_TYPE_PERCENT, .default_value = "50.0", .description = "percentage of rep_count-1 that must be up" }, { .key = {NULL} }, };