diff options
Diffstat (limited to 'xlators/cluster/afr/src/afr-open.c')
| -rw-r--r-- | xlators/cluster/afr/src/afr-open.c | 667 |
1 files changed, 287 insertions, 380 deletions
diff --git a/xlators/cluster/afr/src/afr-open.c b/xlators/cluster/afr/src/afr-open.c index c0be197f212..64856042b65 100644 --- a/xlators/cluster/afr/src/afr-open.c +++ b/xlators/cluster/afr/src/afr-open.c @@ -8,439 +8,346 @@ cases as published by the Free Software Foundation. */ -#include <libgen.h> #include <unistd.h> -#include <fnmatch.h> #include <sys/time.h> #include <stdlib.h> #include <signal.h> -#ifndef _CONFIG_H -#define _CONFIG_H -#include "config.h" -#endif - -#include "glusterfs.h" +#include <glusterfs/glusterfs.h> #include "afr.h" -#include "dict.h" -#include "xlator.h" -#include "hashfn.h" -#include "logging.h" -#include "stack.h" -#include "list.h" -#include "call-stub.h" -#include "defaults.h" -#include "common-utils.h" -#include "compat-errno.h" -#include "compat.h" -#include "byte-order.h" -#include "statedump.h" - -#include "fd.h" - -#include "afr-inode-read.h" -#include "afr-inode-write.h" -#include "afr-dir-read.h" -#include "afr-dir-write.h" -#include "afr-transaction.h" -#include "afr-self-heal.h" -#include "afr-self-heal-common.h" - -int -afr_stale_child_up (afr_local_t *local, xlator_t *this) -{ - int i = 0; - afr_private_t *priv = NULL; - int up = -1; +#include <glusterfs/dict.h> +#include <glusterfs/logging.h> +#include <glusterfs/defaults.h> +#include <glusterfs/common-utils.h> +#include <glusterfs/compat-errno.h> +#include <glusterfs/compat.h> +#include <glusterfs/byte-order.h> +#include <glusterfs/statedump.h> - priv = this->private; - - if (!local->fresh_children) - local->fresh_children = afr_children_create (priv->child_count); - if (!local->fresh_children) - goto out; - - afr_inode_get_read_ctx (this, local->fd->inode, local->fresh_children); - if (priv->child_count == afr_get_children_count (local->fresh_children, - priv->child_count)) - goto out; - - for (i = 0; i < priv->child_count; i++) { - if (!local->child_up[i]) - continue; - if (afr_is_child_present (local->fresh_children, - priv->child_count, i)) - continue; - up = i; - break; - } -out: - return up; -} +#include "afr-transaction.h" -void -afr_perform_data_self_heal (call_frame_t *frame, xlator_t *this) +gf_boolean_t +afr_is_fd_fixable(fd_t *fd) { - afr_local_t *local = NULL; - afr_self_heal_t *sh = NULL; - inode_t *inode = NULL; - int st_child = -1; - char reason[64] = {0}; - - local = frame->local; - sh = &local->self_heal; - inode = local->fd->inode; - - if (!IA_ISREG (inode->ia_type)) - goto out; - - st_child = afr_stale_child_up (local, this); - if (st_child < 0) - goto out; - - sh->do_data_self_heal = _gf_true; - sh->do_metadata_self_heal = _gf_true; - sh->do_gfid_self_heal = _gf_true; - sh->do_missing_entry_self_heal = _gf_true; - - snprintf (reason, sizeof (reason), "stale subvolume %d detected", - st_child); - afr_launch_self_heal (frame, this, inode, _gf_true, inode->ia_type, - reason, NULL, NULL); -out: - return; + if (!fd || !fd->inode) + return _gf_false; + else if (fd_is_anonymous(fd)) + return _gf_false; + else if (gf_uuid_is_null(fd->inode->gfid)) + return _gf_false; + + return _gf_true; } int -afr_open_ftruncate_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) +afr_open_ftruncate_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) { - afr_local_t * local = frame->local; - afr_private_t *priv = NULL; - - priv = this->private; - if (afr_open_only_data_self_heal (priv->data_self_heal)) - afr_perform_data_self_heal (frame, this); - AFR_STACK_UNWIND (open, frame, local->op_ret, local->op_errno, - local->fd, xdata); - return 0; -} + afr_local_t *local = frame->local; + AFR_STACK_UNWIND(open, frame, local->op_ret, local->op_errno, + local->cont.open.fd, xdata); + return 0; +} int -afr_open_cbk (call_frame_t *frame, void *cookie, - xlator_t *this, int32_t op_ret, int32_t op_errno, - fd_t *fd, dict_t *xdata) +afr_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, + int32_t op_errno, fd_t *fd, dict_t *xdata) { - afr_local_t * local = NULL; - int ret = 0; - int call_count = -1; - int child_index = (long) cookie; - afr_private_t *priv = NULL; - - priv = this->private; - local = frame->local; - - LOCK (&frame->lock); - { - if (op_ret == -1) { - local->op_errno = op_errno; - } - - if (op_ret >= 0) { - local->op_ret = op_ret; - local->success_count++; - - ret = afr_child_fd_ctx_set (this, fd, child_index, - local->cont.open.flags); - if (ret) { - local->op_ret = -1; - local->op_errno = -ret; - goto unlock; - } - } + afr_local_t *local = NULL; + int call_count = -1; + int child_index = (long)cookie; + afr_fd_ctx_t *fd_ctx = NULL; + + local = frame->local; + fd_ctx = local->fd_ctx; + + local->replies[child_index].valid = 1; + local->replies[child_index].op_ret = op_ret; + local->replies[child_index].op_errno = op_errno; + + LOCK(&frame->lock); + { + if (op_ret == -1) { + local->op_errno = op_errno; + fd_ctx->opened_on[child_index] = AFR_FD_NOT_OPENED; + } else { + local->op_ret = op_ret; + fd_ctx->opened_on[child_index] = AFR_FD_OPENED; + if (!local->xdata_rsp && xdata) + local->xdata_rsp = dict_ref(xdata); } -unlock: - UNLOCK (&frame->lock); - - call_count = afr_frame_return (frame); - - if (call_count == 0) { - if ((local->cont.open.flags & O_TRUNC) - && (local->op_ret >= 0)) { - STACK_WIND (frame, afr_open_ftruncate_cbk, - this, this->fops->ftruncate, - fd, 0, NULL); - } else { - if (afr_open_only_data_self_heal (priv->data_self_heal)) - afr_perform_data_self_heal (frame, this); - AFR_STACK_UNWIND (open, frame, local->op_ret, - local->op_errno, local->fd, xdata); - } + call_count = --local->call_count; + } + UNLOCK(&frame->lock); + + if (call_count == 0) { + afr_handle_replies_quorum(frame, this); + if (local->op_ret == -1) { + AFR_STACK_UNWIND(open, frame, local->op_ret, local->op_errno, NULL, + NULL); + } else if (fd_ctx->flags & O_TRUNC) { + STACK_WIND(frame, afr_open_ftruncate_cbk, this, + this->fops->ftruncate, fd, 0, NULL); + } else { + AFR_STACK_UNWIND(open, frame, local->op_ret, local->op_errno, + local->cont.open.fd, local->xdata_rsp); } + } - return 0; + return 0; } int -afr_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, - fd_t *fd, dict_t *xdata) +afr_open_continue(call_frame_t *frame, xlator_t *this, int err) { - afr_private_t * priv = NULL; - afr_local_t * local = NULL; - int i = 0; - int ret = -1; - int32_t call_count = 0; - int32_t op_errno = 0; - int32_t wind_flags = flags & (~O_TRUNC); - //We can't let truncation to happen outside transaction. - - VALIDATE_OR_GOTO (frame, out); - VALIDATE_OR_GOTO (this, out); - VALIDATE_OR_GOTO (this->private, out); - VALIDATE_OR_GOTO (loc, out); - - priv = this->private; - - if (flags & (O_CREAT|O_TRUNC)) { - QUORUM_CHECK(open,out); - } - - if (afr_is_split_brain (this, loc->inode)) { - /* self-heal failed */ - gf_log (this->name, GF_LOG_WARNING, - "failed to open as split brain seen, returning EIO"); - op_errno = EIO; - goto out; - } - - AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); - local = frame->local; + afr_local_t *local = NULL; + afr_private_t *priv = NULL; + int call_count = 0; + int i = 0; - ret = afr_local_init (local, priv, &op_errno); - if (ret < 0) - goto out; + local = frame->local; + priv = this->private; - call_count = local->call_count; - loc_copy (&local->loc, loc); - - local->cont.open.flags = flags; - - local->fd = fd_ref (fd); + if (err) { + AFR_STACK_UNWIND(open, frame, -1, err, NULL, NULL); + } else { + local->call_count = AFR_COUNT(local->child_up, priv->child_count); + call_count = local->call_count; for (i = 0; i < priv->child_count; i++) { - if (local->child_up[i]) { - STACK_WIND_COOKIE (frame, afr_open_cbk, (void *) (long) i, - priv->children[i], - priv->children[i]->fops->open, - loc, wind_flags, fd, xdata); - - if (!--call_count) - break; - } + if (local->child_up[i]) { + STACK_WIND_COOKIE(frame, afr_open_cbk, (void *)(long)i, + priv->children[i], + priv->children[i]->fops->open, &local->loc, + (local->cont.open.flags & ~O_TRUNC), + local->cont.open.fd, local->xdata_req); + if (!--call_count) + break; + } } + } + return 0; +} - ret = 0; +int +afr_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, + fd_t *fd, dict_t *xdata) +{ + afr_private_t *priv = NULL; + afr_local_t *local = NULL; + int spb_subvol = 0; + int event_generation = 0; + int ret = 0; + int32_t op_errno = 0; + afr_fd_ctx_t *fd_ctx = NULL; + + // We can't let truncation to happen outside transaction. + + priv = this->private; + + local = AFR_FRAME_INIT(frame, op_errno); + if (!local) + goto out; + + local->op = GF_FOP_OPEN; + fd_ctx = afr_fd_ctx_get(fd, this); + if (!fd_ctx) { + op_errno = ENOMEM; + goto out; + } + + if (priv->quorum_count && !afr_has_quorum(local->child_up, this, NULL)) { + op_errno = afr_quorum_errno(priv); + goto out; + } + + if (!afr_is_consistent_io_possible(local, priv, &op_errno)) + goto out; + + local->inode = inode_ref(loc->inode); + loc_copy(&local->loc, loc); + local->fd_ctx = fd_ctx; + fd_ctx->flags = flags; + if (xdata) + local->xdata_req = dict_ref(xdata); + + local->cont.open.flags = flags; + local->cont.open.fd = fd_ref(fd); + + ret = afr_inode_get_readable(frame, local->inode, this, NULL, + &event_generation, AFR_DATA_TRANSACTION); + if ((ret < 0) && + (afr_split_brain_read_subvol_get(local->inode, this, NULL, + &spb_subvol) == 0) && + spb_subvol < 0) { + afr_inode_refresh(frame, this, local->inode, local->inode->gfid, + afr_open_continue); + } else { + afr_open_continue(frame, this, 0); + } + + return 0; out: - if (ret < 0) - AFR_STACK_UNWIND (open, frame, -1, op_errno, fd, xdata); + AFR_STACK_UNWIND(open, frame, -1, op_errno, fd, NULL); - return 0; + return 0; } -//NOTE: this function should be called with holding the lock on -//fd to which fd_ctx belongs -void -afr_get_resumable_calls (xlator_t *this, afr_fd_ctx_t *fd_ctx, - struct list_head *list) +int +afr_openfd_fix_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, fd_t *fd, + dict_t *xdata) { - afr_fd_paused_call_t *paused_call = NULL; - afr_fd_paused_call_t *tmp = NULL; - afr_local_t *call_local = NULL; - afr_private_t *priv = NULL; - int i = 0; - gf_boolean_t call = _gf_false; - - priv = this->private; - list_for_each_entry_safe (paused_call, tmp, &fd_ctx->paused_calls, - call_list) { - call = _gf_true; - call_local = paused_call->frame->local; - for (i = 0; i < priv->child_count; i++) { - if (call_local->child_up[i] && - (fd_ctx->opened_on[i] == AFR_FD_OPENING)) - call = _gf_false; - } - - if (call) { - list_del_init (&paused_call->call_list); - list_add (&paused_call->call_list, list); - } + afr_local_t *local = NULL; + afr_private_t *priv = NULL; + afr_fd_ctx_t *fd_ctx = NULL; + int call_count = 0; + int child_index = (long)cookie; + + priv = this->private; + local = frame->local; + + if (op_ret >= 0) { + gf_msg_debug(this->name, 0, + "fd for %s opened " + "successfully on subvolume %s", + local->loc.path, priv->children[child_index]->name); + } else { + gf_smsg(this->name, fop_log_level(GF_FOP_OPEN, op_errno), op_errno, + AFR_MSG_OPEN_FAIL, "path=%s", local->loc.path, "subvolume=%s", + priv->children[child_index]->name, NULL); + } + + fd_ctx = local->fd_ctx; + + LOCK(&local->fd->lock); + { + if (op_ret >= 0) { + fd_ctx->opened_on[child_index] = AFR_FD_OPENED; + } else { + fd_ctx->opened_on[child_index] = AFR_FD_NOT_OPENED; } + } + UNLOCK(&local->fd->lock); + + call_count = afr_frame_return(frame); + if (call_count == 0) + AFR_STACK_DESTROY(frame); + + return 0; } -void -afr_resume_calls (xlator_t *this, struct list_head *list) +static int +afr_fd_ctx_need_open(fd_t *fd, xlator_t *this, unsigned char *need_open) { - afr_fd_paused_call_t *paused_call = NULL; - afr_fd_paused_call_t *tmp = NULL; - afr_local_t *call_local = NULL; - - list_for_each_entry_safe (paused_call, tmp, list, call_list) { - list_del_init (&paused_call->call_list); - call_local = paused_call->frame->local; - call_local->fop_call_continue (paused_call->frame, this); - GF_FREE (paused_call); + afr_fd_ctx_t *fd_ctx = NULL; + afr_private_t *priv = NULL; + int i = 0; + int count = 0; + + priv = this->private; + + fd_ctx = afr_fd_ctx_get(fd, this); + if (!fd_ctx) + return 0; + + LOCK(&fd->lock); + { + for (i = 0; i < priv->child_count; i++) { + if (fd_ctx->opened_on[i] == AFR_FD_NOT_OPENED && + priv->child_up[i]) { + fd_ctx->opened_on[i] = AFR_FD_OPENING; + need_open[i] = 1; + count++; + } else { + need_open[i] = 0; + } } + } + UNLOCK(&fd->lock); + + return count; } -int -afr_openfd_fix_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata) +void +afr_fix_open(fd_t *fd, xlator_t *this) { - afr_local_t *local = NULL; - afr_private_t *priv = NULL; - afr_fd_ctx_t *fd_ctx = NULL; - int call_count = 0; - int child_index = (long) cookie; - struct list_head paused_calls = {0}; - gf_boolean_t fop_paused = _gf_false; - fd_t *local_fd = NULL; - - priv = this->private; - local = frame->local; - fop_paused = local->fop_paused; + afr_private_t *priv = NULL; + int i = 0; + call_frame_t *frame = NULL; + afr_local_t *local = NULL; + int ret = -1; + int32_t op_errno = 0; + afr_fd_ctx_t *fd_ctx = NULL; + unsigned char *need_open = NULL; + int call_count = 0; - if (op_ret >= 0) { - gf_log (this->name, GF_LOG_INFO, "fd for %s opened " - "successfully on subvolume %s", local->loc.path, - priv->children[child_index]->name); - } + priv = this->private; - local_fd = fd_ref (local->fd); - call_count = afr_frame_return (frame); - //Note: Do not access any thing using the frame outside call_count 0 + if (!afr_is_fd_fixable(fd)) + goto out; - //Note: No frame locking needed for this block of code - fd_ctx = afr_fd_ctx_get (local_fd, this); - if (!fd_ctx) { - gf_log (this->name, GF_LOG_WARNING, - "failed to get fd context, %p", local_fd); - goto out; - } + fd_ctx = afr_fd_ctx_get(fd, this); + if (!fd_ctx) + goto out; - LOCK (&local_fd->lock); - { - if (op_ret >= 0) { - fd_ctx->opened_on[child_index] = AFR_FD_OPENED; - } else { - fd_ctx->opened_on[child_index] = AFR_FD_NOT_OPENED; - } - if (call_count == 0) { - INIT_LIST_HEAD (&paused_calls); - afr_get_resumable_calls (this, fd_ctx, &paused_calls); - } - } - UNLOCK (&local_fd->lock); -out: - if (call_count == 0) { - afr_resume_calls (this, &paused_calls); - //If the fop is paused then resume_calls will continue the fop - if (fop_paused) - goto done; - - if (local->fop_call_continue) - local->fop_call_continue (frame, this); - else - AFR_STACK_DESTROY (frame); - } + need_open = alloca0(priv->child_count); -done: - fd_unref (local_fd); - return 0; -} + call_count = afr_fd_ctx_need_open(fd, this, need_open); + if (!call_count) + goto out; -int -afr_fix_open (call_frame_t *frame, xlator_t *this, afr_fd_ctx_t *fd_ctx, - int need_open_count, int *need_open) -{ - afr_local_t *local = NULL; - afr_private_t *priv = NULL; - int i = 0; - call_frame_t *open_frame = NULL; - afr_local_t *open_local = NULL; - int ret = -1; - ia_type_t ia_type = IA_INVAL; - int32_t op_errno = 0; - - GF_ASSERT (fd_ctx); - GF_ASSERT (need_open_count > 0); - GF_ASSERT (need_open); - - local = frame->local; - priv = this->private; - if (!local->fop_call_continue) { - open_frame = copy_frame (frame); - if (!open_frame) { - ret = -ENOMEM; - goto out; - } - AFR_LOCAL_ALLOC_OR_GOTO (open_frame->local, out); - open_local = open_frame->local; - ret = afr_local_init (open_local, priv, &op_errno); - if (ret < 0) - goto out; - loc_copy (&open_local->loc, &local->loc); - open_local->fd = fd_ref (local->fd); - } else { - ret = 0; - open_frame = frame; - open_local = local; - } + frame = create_frame(this, this->ctx->pool); + if (!frame) + goto out; - open_local->call_count = need_open_count; + local = AFR_FRAME_INIT(frame, op_errno); + if (!local) + goto out; - gf_log (this->name, GF_LOG_DEBUG, "need open count: %d", - need_open_count); + local->loc.inode = inode_ref(fd->inode); + ret = loc_path(&local->loc, NULL); + if (ret < 0) + goto out; - ia_type = open_local->fd->inode->ia_type; - GF_ASSERT (ia_type != IA_INVAL); - for (i = 0; i < priv->child_count; i++) { - if (!need_open[i]) - continue; - if (IA_IFDIR == ia_type) { - gf_log (this->name, GF_LOG_DEBUG, - "opening fd for dir %s on subvolume %s", - local->loc.path, priv->children[i]->name); - - STACK_WIND_COOKIE (open_frame, afr_openfd_fix_open_cbk, - (void*) (long) i, - priv->children[i], - priv->children[i]->fops->opendir, - &open_local->loc, open_local->fd, - NULL); - } else { - gf_log (this->name, GF_LOG_DEBUG, - "opening fd for file %s on subvolume %s", - local->loc.path, priv->children[i]->name); - - STACK_WIND_COOKIE (open_frame, afr_openfd_fix_open_cbk, - (void *)(long) i, - priv->children[i], - priv->children[i]->fops->open, - &open_local->loc, - fd_ctx->flags & (~O_TRUNC), - open_local->fd, NULL); - } + local->fd = fd_ref(fd); + local->fd_ctx = fd_ctx; + + local->call_count = call_count; + + gf_msg_debug(this->name, 0, "need open count: %d", call_count); + + for (i = 0; i < priv->child_count; i++) { + if (!need_open[i]) + continue; + if (IA_IFDIR == fd->inode->ia_type) { + gf_msg_debug(this->name, 0, "opening fd for dir %s on subvolume %s", + local->loc.path, priv->children[i]->name); + + STACK_WIND_COOKIE(frame, afr_openfd_fix_open_cbk, (void *)(long)i, + priv->children[i], + priv->children[i]->fops->opendir, &local->loc, + local->fd, NULL); + } else { + gf_msg_debug(this->name, 0, + "opening fd for file %s on subvolume %s", + local->loc.path, priv->children[i]->name); + + STACK_WIND_COOKIE(frame, afr_openfd_fix_open_cbk, (void *)(long)i, + priv->children[i], priv->children[i]->fops->open, + &local->loc, fd_ctx->flags & (~O_TRUNC), + local->fd, NULL); } - op_errno = 0; - ret = 0; + + if (!--call_count) + break; + } + + return; out: - if (op_errno) - ret = -op_errno; - if (ret && open_frame) - AFR_STACK_DESTROY (open_frame); - return ret; + if (frame) + AFR_STACK_DESTROY(frame); } |