diff options
Diffstat (limited to 'xlators/performance')
| -rw-r--r-- | xlators/performance/read-ahead/src/page.c | 574 | ||||
| -rw-r--r-- | xlators/performance/read-ahead/src/read-ahead-mem-types.h | 33 | ||||
| -rw-r--r-- | xlators/performance/read-ahead/src/read-ahead.c | 1103 | ||||
| -rw-r--r-- | xlators/performance/read-ahead/src/read-ahead.h | 126 | 
4 files changed, 924 insertions, 912 deletions
diff --git a/xlators/performance/read-ahead/src/page.c b/xlators/performance/read-ahead/src/page.c index 925bc253d65..d410986513a 100644 --- a/xlators/performance/read-ahead/src/page.c +++ b/xlators/performance/read-ahead/src/page.c @@ -32,81 +32,81 @@  ra_page_t *  ra_page_get (ra_file_t *file, off_t offset)  { -	ra_page_t *page = NULL; -	off_t     rounded_offset = 0; +        ra_page_t *page           = NULL; +        off_t      rounded_offset = 0; -	page = file->pages.next; -	rounded_offset = floor (offset, file->page_size); +        page = file->pages.next; +        rounded_offset = floor (offset, file->page_size); -	while (page != &file->pages && page->offset < rounded_offset) -		page = page->next; +        while (page != &file->pages && page->offset < rounded_offset) +                page = page->next; -	if (page == &file->pages || page->offset != rounded_offset) -		page = NULL; +        if (page == &file->pages || page->offset != rounded_offset) +                page = NULL; -	return page; +        return page;  }  ra_page_t *  ra_page_create (ra_file_t *file, off_t offset)  { -	ra_page_t  *page      = NULL; -	off_t      rounded_offset = 0; -	ra_page_t  *newpage   = NULL; +        ra_page_t  *page           = NULL; +        off_t       rounded_offset = 0; +        ra_page_t  *newpage        = NULL; -	page           = file->pages.next; -	rounded_offset = floor (offset, file->page_size); +        page           = file->pages.next; +        rounded_offset = floor (offset, file->page_size); -	while (page != &file->pages && page->offset < rounded_offset) -		page = page->next; +        while (page != &file->pages && page->offset < rounded_offset) +                page = page->next; -	if (page == &file->pages || page->offset != rounded_offset) { -		newpage = GF_CALLOC (1, sizeof (*newpage), +        if (page == &file->pages || page->offset != rounded_offset) { +                newpage = GF_CALLOC (1, sizeof (*newpage),                                       gf_ra_mt_ra_page_t); -		if (!newpage) -			return NULL; +                if (!newpage) +                        return NULL; -		newpage->offset = rounded_offset; -		newpage->prev = page->prev; -		newpage->next = page; -		newpage->file = file; -		page->prev->next = newpage; -		page->prev = newpage; +                newpage->offset = rounded_offset; +                newpage->prev = page->prev; +                newpage->next = page; +                newpage->file = file; +                page->prev->next = newpage; +                page->prev = newpage; -		page = newpage; -	} +                page = newpage; +        } -	return page; +        return page;  }  void  ra_wait_on_page (ra_page_t *page, call_frame_t *frame)  { -	ra_waitq_t *waitq = NULL; -	ra_local_t *local = NULL; +        ra_waitq_t *waitq = NULL; +        ra_local_t *local = NULL; -	local = frame->local; -	waitq = GF_CALLOC (1, sizeof (*waitq), +        local = frame->local; +        waitq = GF_CALLOC (1, sizeof (*waitq),                             gf_ra_mt_ra_waitq_t); -	if (!waitq) { -		gf_log (frame->this->name, GF_LOG_ERROR, -			"out of memory"); +        if (!waitq) { +                gf_log (frame->this->name, GF_LOG_ERROR, +                        "out of memory");                  local->op_ret = -1;                  local->op_errno = ENOMEM;                  goto out; -	} +        } -	waitq->data = frame; -	waitq->next = page->waitq; -	page->waitq = waitq; +        waitq->data = frame; +        waitq->next = page->waitq; +        page->waitq = waitq; -	ra_local_lock (local); -	{ -		local->wait_count++; -	} -	ra_local_unlock (local); +        ra_local_lock (local); +        { +                local->wait_count++; +        } +        ra_local_unlock (local);  out:          return; @@ -116,112 +116,112 @@ out:  void  ra_waitq_return (ra_waitq_t *waitq)  { -	ra_waitq_t   *trav = NULL; -	ra_waitq_t   *next = NULL; -	call_frame_t *frame = NULL; +        ra_waitq_t   *trav  = NULL; +        ra_waitq_t   *next  = NULL; +        call_frame_t *frame = NULL; -	for (trav = waitq; trav; trav = next) { -		next = trav->next; +        for (trav = waitq; trav; trav = next) { +                next = trav->next; -		frame = trav->data; -		ra_frame_return (frame); -		GF_FREE (trav); -	} +                frame = trav->data; +                ra_frame_return (frame); +                GF_FREE (trav); +        }  }  int  ra_fault_cbk (call_frame_t *frame, void *cookie, xlator_t *this, -	      int32_t op_ret, int32_t op_errno, struct iovec *vector, -	      int32_t count, struct iatt *stbuf, struct iobref *iobref) +              int32_t op_ret, int32_t op_errno, struct iovec *vector, +              int32_t count, struct iatt *stbuf, struct iobref *iobref)  { -	ra_local_t   *local = NULL; -	off_t        pending_offset = 0; -	ra_file_t    *file = NULL; -	ra_page_t    *page = NULL; -	ra_waitq_t   *waitq = NULL; -	fd_t         *fd = NULL; -	uint64_t     tmp_file = 0; - -	local = frame->local; -	fd  = local->fd; - -	fd_ctx_get (fd, this, &tmp_file); - -	file = (ra_file_t *)(long)tmp_file; -	pending_offset = local->pending_offset; - -	ra_file_lock (file); -	{ -		if (op_ret >= 0) -			file->stbuf = *stbuf; - -		if (op_ret < 0) { -			page = ra_page_get (file, pending_offset); -			if (page) -				waitq = ra_page_error (page, op_ret, op_errno); -			goto unlock; -		} - -		page = ra_page_get (file, pending_offset); -		if (!page) { -			gf_log (this->name, GF_LOG_DEBUG, -				"wasted copy: %"PRId64"[+%"PRId64"] file=%p",  -				pending_offset, file->page_size, file); -			goto unlock; -		} - -		if (page->vector) { -			iobref_unref (page->iobref); -			GF_FREE (page->vector); -		} - -		page->vector = iov_dup (vector, count); +        ra_local_t   *local          = NULL; +        off_t         pending_offset = 0; +        ra_file_t    *file           = NULL; +        ra_page_t    *page           = NULL; +        ra_waitq_t   *waitq          = NULL; +        fd_t         *fd             = NULL; +        uint64_t      tmp_file       = 0; + +        local = frame->local; +        fd  = local->fd; + +        fd_ctx_get (fd, this, &tmp_file); + +        file = (ra_file_t *)(long)tmp_file; +        pending_offset = local->pending_offset; + +        ra_file_lock (file); +        { +                if (op_ret >= 0) +                        file->stbuf = *stbuf; + +                if (op_ret < 0) { +                        page = ra_page_get (file, pending_offset); +                        if (page) +                                waitq = ra_page_error (page, op_ret, op_errno); +                        goto unlock; +                } + +                page = ra_page_get (file, pending_offset); +                if (!page) { +                        gf_log (this->name, GF_LOG_DEBUG, +                                "wasted copy: %"PRId64"[+%"PRId64"] file=%p", +                                pending_offset, file->page_size, file); +                        goto unlock; +                } + +                if (page->vector) { +                        iobref_unref (page->iobref); +                        GF_FREE (page->vector); +                } + +                page->vector = iov_dup (vector, count);                  if (page->vector == NULL) {                          waitq = ra_page_error (page, -1, ENOMEM);                          goto unlock;                  } -		page->count = count; -		page->iobref = iobref_ref (iobref); -		page->ready = 1; +                page->count = count; +                page->iobref = iobref_ref (iobref); +                page->ready = 1; -		page->size = iov_length (vector, count); +                page->size = iov_length (vector, count); -		waitq = ra_page_wakeup (page); -	} +                waitq = ra_page_wakeup (page); +        }  unlock: -	ra_file_unlock (file); +        ra_file_unlock (file); -	ra_waitq_return (waitq); +        ra_waitq_return (waitq); -	fd_unref (local->fd); +        fd_unref (local->fd); -	GF_FREE (frame->local); -	frame->local = NULL; +        GF_FREE (frame->local); +        frame->local = NULL; -	STACK_DESTROY (frame->root); -	return 0; +        STACK_DESTROY (frame->root); +        return 0;  }  void  ra_page_fault (ra_file_t *file, call_frame_t *frame, off_t offset)  { -	call_frame_t *fault_frame = NULL; -	ra_local_t   *fault_local = NULL; -        ra_page_t    *page = NULL; -        ra_waitq_t   *waitq = NULL; -        int32_t      op_ret = -1, op_errno = -1;  +        call_frame_t *fault_frame = NULL; +        ra_local_t   *fault_local = NULL; +        ra_page_t    *page        = NULL; +        ra_waitq_t   *waitq       = NULL; +        int32_t       op_ret      = -1, op_errno = -1; -	fault_frame = copy_frame (frame); +        fault_frame = copy_frame (frame);          if (fault_frame == NULL) {                  op_ret = -1;                  op_errno = ENOMEM;                  goto err;          } -	fault_local = GF_CALLOC (1, sizeof (ra_local_t), +        fault_local = GF_CALLOC (1, sizeof (ra_local_t),                                   gf_ra_mt_ra_local_t);          if (fault_local == NULL) {                  STACK_DESTROY (fault_frame->root); @@ -230,18 +230,18 @@ ra_page_fault (ra_file_t *file, call_frame_t *frame, off_t offset)                  goto err;          } -	fault_frame->local = fault_local; -	fault_local->pending_offset = offset; -	fault_local->pending_size = file->page_size; +        fault_frame->local = fault_local; +        fault_local->pending_offset = offset; +        fault_local->pending_size = file->page_size; -	fault_local->fd = fd_ref (file->fd); +        fault_local->fd = fd_ref (file->fd); -	STACK_WIND (fault_frame, ra_fault_cbk, -		    FIRST_CHILD (fault_frame->this), -		    FIRST_CHILD (fault_frame->this)->fops->readv, -		    file->fd, file->page_size, offset); +        STACK_WIND (fault_frame, ra_fault_cbk, +                    FIRST_CHILD (fault_frame->this), +                    FIRST_CHILD (fault_frame->this)->fops->readv, +                    file->fd, file->page_size, offset); -	return; +        return;  err:          ra_file_lock (file); @@ -252,7 +252,7 @@ err:                                                 op_errno);          }          ra_file_unlock (file); -         +          if (waitq != NULL) {                  ra_waitq_return (waitq);          } @@ -261,54 +261,54 @@ err:  void  ra_frame_fill (ra_page_t *page, call_frame_t *frame)  { -	ra_local_t *local = NULL; -	ra_fill_t  *fill = NULL; -	off_t      src_offset = 0; -	off_t      dst_offset = 0; -	ssize_t    copy_size = 0; -	ra_fill_t  *new = NULL; - -	local = frame->local; -	fill  = &local->fill; - -	if (local->op_ret != -1 && page->size) { -		if (local->offset > page->offset) -			src_offset = local->offset - page->offset; -		else -			dst_offset = page->offset - local->offset; - -		copy_size = min (page->size - src_offset, -				 local->size - dst_offset); - -		if (copy_size < 0) { -			/* if page contains fewer bytes and the required offset -			   is beyond the page size in the page */ -			copy_size = src_offset = 0; -		} - -		fill = fill->next; -		while (fill != &local->fill) { -			if (fill->offset > page->offset) { -				break; -			} -			fill = fill->next; -		} - -		new = GF_CALLOC (1, sizeof (*new), -                                gf_ra_mt_ra_fill_t); +        ra_local_t *local      = NULL; +        ra_fill_t  *fill       = NULL; +        off_t       src_offset = 0; +        off_t       dst_offset = 0; +        ssize_t     copy_size  = 0; +        ra_fill_t  *new        = NULL; + +        local = frame->local; +        fill  = &local->fill; + +        if (local->op_ret != -1 && page->size) { +                if (local->offset > page->offset) +                        src_offset = local->offset - page->offset; +                else +                        dst_offset = page->offset - local->offset; + +                copy_size = min (page->size - src_offset, +                                 local->size - dst_offset); + +                if (copy_size < 0) { +                        /* if page contains fewer bytes and the required offset +                           is beyond the page size in the page */ +                        copy_size = src_offset = 0; +                } + +                fill = fill->next; +                while (fill != &local->fill) { +                        if (fill->offset > page->offset) { +                                break; +                        } +                        fill = fill->next; +                } + +                new = GF_CALLOC (1, sizeof (*new), +                                 gf_ra_mt_ra_fill_t);                  if (new == NULL) {                          local->op_ret = -1;                          local->op_errno = ENOMEM;                          goto out;                  } -		new->offset = page->offset; -		new->size = copy_size; -		new->iobref = iobref_ref (page->iobref); -		new->count = iov_subset (page->vector, page->count, -					 src_offset, src_offset+copy_size, -					 NULL); -		new->vector = GF_CALLOC (new->count, sizeof (struct iovec), +                new->offset = page->offset; +                new->size = copy_size; +                new->iobref = iobref_ref (page->iobref); +                new->count = iov_subset (page->vector, page->count, +                                         src_offset, src_offset+copy_size, +                                         NULL); +                new->vector = GF_CALLOC (new->count, sizeof (struct iovec),                                           gf_ra_mt_iovec);                  if (new->vector == NULL) {                          local->op_ret = -1; @@ -317,17 +317,17 @@ ra_frame_fill (ra_page_t *page, call_frame_t *frame)                          goto out;                  } -		new->count = iov_subset (page->vector, page->count, -					 src_offset, src_offset+copy_size, -					 new->vector); +                new->count = iov_subset (page->vector, page->count, +                                         src_offset, src_offset+copy_size, +                                         new->vector); -		new->next = fill; -		new->prev = new->next->prev; -		new->next->prev = new; -		new->prev->next = new; +                new->next = fill; +                new->prev = new->next->prev; +                new->next->prev = new; +                new->prev->next = new; -		local->op_ret += copy_size; -	} +                local->op_ret += copy_size; +        }  out:          return; @@ -337,34 +337,34 @@ out:  void  ra_frame_unwind (call_frame_t *frame)  { -	ra_local_t    *local = NULL; -	ra_fill_t     *fill = NULL; -	int32_t       count = 0; -	struct iovec  *vector; -	int32_t       copied = 0; -        struct iobref *iobref = NULL; -	ra_fill_t     *next = NULL; -	fd_t          *fd = NULL; -	ra_file_t     *file = NULL; -	uint64_t      tmp_file = 0; - -	local = frame->local; -	fill  = local->fill.next; - -	iobref  = iobref_new (); +        ra_local_t    *local    = NULL; +        ra_fill_t     *fill     = NULL; +        int32_t        count    = 0; +        struct iovec  *vector   = NULL; +        int32_t        copied   = 0; +        struct iobref *iobref   = NULL; +        ra_fill_t     *next     = NULL; +        fd_t          *fd       = NULL; +        ra_file_t     *file     = NULL; +        uint64_t       tmp_file = 0; + +        local = frame->local; +        fill  = local->fill.next; + +        iobref  = iobref_new ();          if (iobref == NULL) {                  local->op_ret = -1;                  local->op_errno = ENOMEM;          } -	frame->local = NULL; +        frame->local = NULL; -	while (fill != &local->fill) { -		count += fill->count; -		fill = fill->next; -	} +        while (fill != &local->fill) { +                count += fill->count; +                fill = fill->next; +        } -	vector = GF_CALLOC (count, sizeof (*vector), +        vector = GF_CALLOC (count, sizeof (*vector),                              gf_ra_mt_iovec);          if (vector == NULL) {                  local->op_ret = -1; @@ -373,42 +373,42 @@ ra_frame_unwind (call_frame_t *frame)                  iobref = NULL;          } -	fill = local->fill.next; +        fill = local->fill.next; -	while (fill != &local->fill) { -		next = fill->next; +        while (fill != &local->fill) { +                next = fill->next;                  if ((vector != NULL) && (iobref != NULL)) {                          memcpy (((char *)vector) + copied, fill->vector,                                  fill->count * sizeof (*vector)); -                         +                          copied += (fill->count * sizeof (*vector));                          iobref_merge (iobref, fill->iobref);                  } -		fill->next->prev = fill->prev; -		fill->prev->next = fill->prev; +                fill->next->prev = fill->prev; +                fill->prev->next = fill->prev; -		iobref_unref (fill->iobref); -		GF_FREE (fill->vector); -		GF_FREE (fill); +                iobref_unref (fill->iobref); +                GF_FREE (fill->vector); +                GF_FREE (fill); -		fill = next; -	} +                fill = next; +        } -	fd = local->fd; -	fd_ctx_get (fd, frame->this, &tmp_file); -	file = (ra_file_t *)(long)tmp_file; +        fd = local->fd; +        fd_ctx_get (fd, frame->this, &tmp_file); +        file = (ra_file_t *)(long)tmp_file; -	STACK_UNWIND_STRICT (readv, frame, local->op_ret, local->op_errno, +        STACK_UNWIND_STRICT (readv, frame, local->op_ret, local->op_errno,                               vector, count, &file->stbuf, iobref); -	iobref_unref (iobref); -	pthread_mutex_destroy (&local->local_lock); -	GF_FREE (local); -	GF_FREE (vector); +        iobref_unref (iobref); +        pthread_mutex_destroy (&local->local_lock); +        GF_FREE (local); +        GF_FREE (vector); -	return; +        return;  }  /* @@ -419,25 +419,25 @@ ra_frame_unwind (call_frame_t *frame)  void  ra_frame_return (call_frame_t *frame)  { -	ra_local_t  *local = NULL; -	int32_t     wait_count = 0; +        ra_local_t  *local      = NULL; +        int32_t      wait_count = 0; -	local = frame->local; -	GF_ASSERT (local->wait_count > 0); +        local = frame->local; +        GF_ASSERT (local->wait_count > 0); -	ra_local_lock (local); -	{ -		wait_count = --local->wait_count; -	} -	ra_local_unlock (local); +        ra_local_lock (local); +        { +                wait_count = --local->wait_count; +        } +        ra_local_unlock (local); -	if (!wait_count) -		ra_frame_unwind (frame); +        if (!wait_count) +                ra_frame_unwind (frame); -	return; +        return;  } -/*  +/*   * ra_page_wakeup -   * @page:   * @@ -445,18 +445,18 @@ ra_frame_return (call_frame_t *frame)  ra_waitq_t *  ra_page_wakeup (ra_page_t *page)  { -	ra_waitq_t *waitq = NULL, *trav = NULL; -	call_frame_t *frame; +        ra_waitq_t   *waitq = NULL, *trav = NULL; +        call_frame_t *frame = NULL; -	waitq = page->waitq; -	page->waitq = NULL; +        waitq = page->waitq; +        page->waitq = NULL; -	for (trav = waitq; trav; trav = trav->next) { -		frame = trav->data; -		ra_frame_fill (page, frame); -	} +        for (trav = waitq; trav; trav = trav->next) { +                frame = trav->data; +                ra_frame_fill (page, frame); +        } -	return waitq; +        return waitq;  }  /* @@ -467,14 +467,14 @@ ra_page_wakeup (ra_page_t *page)  void  ra_page_purge (ra_page_t *page)  { -	page->prev->next = page->next; -	page->next->prev = page->prev; - -	if (page->iobref) { -		iobref_unref (page->iobref); -	} -	GF_FREE (page->vector); -	GF_FREE (page); +        page->prev->next = page->next; +        page->next->prev = page->prev; + +        if (page->iobref) { +                iobref_unref (page->iobref); +        } +        GF_FREE (page->vector); +        GF_FREE (page);  }  /* @@ -488,30 +488,30 @@ ra_waitq_t *  ra_page_error (ra_page_t *page, int32_t op_ret, int32_t op_errno)  { -	ra_waitq_t   *waitq = NULL; -	ra_waitq_t   *trav = NULL; -	call_frame_t *frame = NULL; -	ra_local_t   *local = NULL; +        ra_waitq_t   *waitq = NULL; +        ra_waitq_t   *trav  = NULL; +        call_frame_t *frame = NULL; +        ra_local_t   *local = NULL; -	waitq = page->waitq; -	page->waitq = NULL; +        waitq = page->waitq; +        page->waitq = NULL; -	for (trav = waitq; trav; trav = trav->next) { -		frame = trav->data; +        for (trav = waitq; trav; trav = trav->next) { +                frame = trav->data; -		local = frame->local; -		if (local->op_ret != -1) { -			local->op_ret   = op_ret; -			local->op_errno = op_errno; -		} -	} +                local = frame->local; +                if (local->op_ret != -1) { +                        local->op_ret   = op_ret; +                        local->op_errno = op_errno; +                } +        } -	ra_page_purge (page); +        ra_page_purge (page); -	return waitq; +        return waitq;  } -/*  +/*   * ra_file_destroy -   * @file:   * @@ -519,24 +519,24 @@ ra_page_error (ra_page_t *page, int32_t op_ret, int32_t op_errno)  void  ra_file_destroy (ra_file_t *file)  { -	ra_conf_t *conf = NULL; -	ra_page_t *trav = NULL; - -	conf = file->conf; - -	ra_conf_lock (conf); -	{ -		file->prev->next = file->next; -		file->next->prev = file->prev; -	} -	ra_conf_unlock (conf); - -	trav = file->pages.next; -	while (trav != &file->pages) { -		ra_page_error (trav, -1, EINVAL); -		trav = file->pages.next; -	} - -	pthread_mutex_destroy (&file->file_lock); -	GF_FREE (file); +        ra_conf_t *conf = NULL; +        ra_page_t *trav = NULL; + +        conf = file->conf; + +        ra_conf_lock (conf); +        { +                file->prev->next = file->next; +                file->next->prev = file->prev; +        } +        ra_conf_unlock (conf); + +        trav = file->pages.next; +        while (trav != &file->pages) { +                ra_page_error (trav, -1, EINVAL); +                trav = file->pages.next; +        } + +        pthread_mutex_destroy (&file->file_lock); +        GF_FREE (file);  } diff --git a/xlators/performance/read-ahead/src/read-ahead-mem-types.h b/xlators/performance/read-ahead/src/read-ahead-mem-types.h index 3b76069bced..1c703a059d0 100644 --- a/xlators/performance/read-ahead/src/read-ahead-mem-types.h +++ b/xlators/performance/read-ahead/src/read-ahead-mem-types.h @@ -1,20 +1,20 @@  /* -   Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com> -   This file is part of GlusterFS. - -   GlusterFS is free software; you can redistribute it and/or modify -   it under the terms of the GNU Affero General Public License as published -   by the Free Software Foundation; either version 3 of the License, -   or (at your option) any later version. - -   GlusterFS is distributed in the hope that it will be useful, but -   WITHOUT ANY WARRANTY; without even the implied warranty of -   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU -   Affero General Public License for more details. - -   You should have received a copy of the GNU Affero General Public License -   along with this program.  If not, see -   <http://www.gnu.org/licenses/>. +  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com> +  This file is part of GlusterFS. + +  GlusterFS is free software; you can redistribute it and/or modify +  it under the terms of the GNU Affero General Public License as published +  by the Free Software Foundation; either version 3 of the License, +  or (at your option) any later version. + +  GlusterFS is distributed in the hope that it will be useful, but +  WITHOUT ANY WARRANTY; without even the implied warranty of +  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU +  Affero General Public License for more details. + +  You should have received a copy of the GNU Affero General Public License +  along with this program.  If not, see +  <http://www.gnu.org/licenses/>.  */ @@ -34,4 +34,3 @@ enum gf_ra_mem_types_ {          gf_ra_mt_end  };  #endif - diff --git a/xlators/performance/read-ahead/src/read-ahead.c b/xlators/performance/read-ahead/src/read-ahead.c index f16e59c4237..8317be4a7d3 100644 --- a/xlators/performance/read-ahead/src/read-ahead.c +++ b/xlators/performance/read-ahead/src/read-ahead.c @@ -17,11 +17,11 @@    <http://www.gnu.org/licenses/>.  */ -/*  -   TODO: -   - handle O_DIRECT -   - maintain offset, flush on lseek -   - ensure efficient memory managment in case of random seek +/* +  TODO: +  - handle O_DIRECT +  - maintain offset, flush on lseek +  - ensure efficient memory managment in case of random seek  */  #ifndef _CONFIG_H @@ -46,75 +46,75 @@ int  ra_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,               int32_t op_ret, int32_t op_errno, fd_t *fd)  { -	ra_conf_t  *conf = NULL; -	ra_file_t  *file = NULL; -	int         ret = 0; +        ra_conf_t  *conf    = NULL; +        ra_file_t  *file    = NULL; +        int         ret     = 0;          long        wbflags = 0; -	conf  = this->private; +        conf  = this->private; -	if (op_ret == -1) { -		goto unwind; -	} +        if (op_ret == -1) { +                goto unwind; +        }          wbflags = (long)frame->local; -	file = GF_CALLOC (1, sizeof (*file), gf_ra_mt_ra_file_t); -	if (!file) { +        file = GF_CALLOC (1, sizeof (*file), gf_ra_mt_ra_file_t); +        if (!file) {                  op_ret = -1;                  op_errno = ENOMEM; -		gf_log (this->name, GF_LOG_ERROR, -			"out of memory"); -		goto unwind; -	} +                gf_log (this->name, GF_LOG_ERROR, +                        "out of memory"); +                goto unwind; +        } -	/* If O_DIRECT open, we disable caching on it */ +        /* If O_DIRECT open, we disable caching on it */ -	if ((fd->flags & O_DIRECT) || ((fd->flags & O_ACCMODE) == O_WRONLY)) -		file->disabled = 1; +        if ((fd->flags & O_DIRECT) || ((fd->flags & O_ACCMODE) == O_WRONLY)) +                file->disabled = 1;          if (wbflags & GF_OPEN_NOWB) {                  file->disabled = 1;          } -	file->offset = (unsigned long long) 0; -	file->conf = conf; -	file->pages.next = &file->pages; -	file->pages.prev = &file->pages; -	file->pages.offset = (unsigned long long) 0; -	file->pages.file = file; - -	ra_conf_lock (conf); -	{ -		file->next = conf->files.next; -		conf->files.next = file; -		file->next->prev = file; -		file->prev = &conf->files; -	} -	ra_conf_unlock (conf); - -	file->fd = fd; -	file->page_count = conf->page_count; -	file->page_size = conf->page_size; -	pthread_mutex_init (&file->file_lock, NULL); - -	if (!file->disabled) { -		file->page_count = 1; -	} - -	ret = fd_ctx_set (fd, this, (uint64_t)(long)file); +        file->offset = (unsigned long long) 0; +        file->conf = conf; +        file->pages.next = &file->pages; +        file->pages.prev = &file->pages; +        file->pages.offset = (unsigned long long) 0; +        file->pages.file = file; + +        ra_conf_lock (conf); +        { +                file->next = conf->files.next; +                conf->files.next = file; +                file->next->prev = file; +                file->prev = &conf->files; +        } +        ra_conf_unlock (conf); + +        file->fd = fd; +        file->page_count = conf->page_count; +        file->page_size = conf->page_size; +        pthread_mutex_init (&file->file_lock, NULL); + +        if (!file->disabled) { +                file->page_count = 1; +        } + +        ret = fd_ctx_set (fd, this, (uint64_t)(long)file);          if (ret == -1) {                  ra_file_destroy (file);                  op_ret = -1;                  op_errno = ENOMEM;          } -         +  unwind:          frame->local = NULL; -	STACK_UNWIND_STRICT (open, frame, op_ret, op_errno, fd); +        STACK_UNWIND_STRICT (open, frame, op_ret, op_errno, fd); -	return 0; +        return 0;  } @@ -124,53 +124,53 @@ ra_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,                 struct iatt *buf, struct iatt *preparent,                 struct iatt *postparent)  { -	ra_conf_t  *conf = NULL; -	ra_file_t  *file = NULL; -	int         ret = 0; +        ra_conf_t  *conf = NULL; +        ra_file_t  *file = NULL; +        int         ret  = 0; -	conf  = this->private; +        conf  = this->private; -	if (op_ret == -1) { -		goto unwind; -	} +        if (op_ret == -1) { +                goto unwind; +        } -	file = GF_CALLOC (1, sizeof (*file), gf_ra_mt_ra_file_t); -	if (!file) { +        file = GF_CALLOC (1, sizeof (*file), gf_ra_mt_ra_file_t); +        if (!file) {                  op_ret = -1;                  op_errno = ENOMEM; -		gf_log (this->name, GF_LOG_ERROR, -			"out of memory"); -		goto unwind; -	} - -	/* If O_DIRECT open, we disable caching on it */ - -	if ((fd->flags & O_DIRECT) || ((fd->flags & O_ACCMODE) == O_WRONLY)) -			file->disabled = 1; - -	file->offset = (unsigned long long) 0; -	//file->size = fd->inode->buf.ia_size; -	file->conf = conf; -	file->pages.next = &file->pages; -	file->pages.prev = &file->pages; -	file->pages.offset = (unsigned long long) 0; -	file->pages.file = file; - -	ra_conf_lock (conf); -	{ -		file->next = conf->files.next; -		conf->files.next = file; -		file->next->prev = file; -		file->prev = &conf->files; -	} -	ra_conf_unlock (conf); - -	file->fd = fd; -	file->page_count = conf->page_count; -	file->page_size = conf->page_size; -	pthread_mutex_init (&file->file_lock, NULL); - -	ret = fd_ctx_set (fd, this, (uint64_t)(long)file); +                gf_log (this->name, GF_LOG_ERROR, +                        "out of memory"); +                goto unwind; +        } + +        /* If O_DIRECT open, we disable caching on it */ + +        if ((fd->flags & O_DIRECT) || ((fd->flags & O_ACCMODE) == O_WRONLY)) +                file->disabled = 1; + +        file->offset = (unsigned long long) 0; +        //file->size = fd->inode->buf.ia_size; +        file->conf = conf; +        file->pages.next = &file->pages; +        file->pages.prev = &file->pages; +        file->pages.offset = (unsigned long long) 0; +        file->pages.file = file; + +        ra_conf_lock (conf); +        { +                file->next = conf->files.next; +                conf->files.next = file; +                file->next->prev = file; +                file->prev = &conf->files; +        } +        ra_conf_unlock (conf); + +        file->fd = fd; +        file->page_count = conf->page_count; +        file->page_size = conf->page_size; +        pthread_mutex_init (&file->file_lock, NULL); + +        ret = fd_ctx_set (fd, this, (uint64_t)(long)file);          if (ret == -1) {                  ra_file_destroy (file);                  op_ret = -1; @@ -178,10 +178,10 @@ ra_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,          }  unwind: -	STACK_UNWIND_STRICT (create, frame, op_ret, op_errno, fd, inode, buf, +        STACK_UNWIND_STRICT (create, frame, op_ret, op_errno, fd, inode, buf,                               preparent, postparent); -	return 0; +        return 0;  } @@ -191,24 +191,25 @@ ra_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,  {          frame->local = (void *)(long)wbflags; -	STACK_WIND (frame, ra_open_cbk, -		    FIRST_CHILD (this), -		    FIRST_CHILD (this)->fops->open, -		    loc, flags, fd, wbflags); +        STACK_WIND (frame, ra_open_cbk, +                    FIRST_CHILD (this), +                    FIRST_CHILD (this)->fops->open, +                    loc, flags, fd, wbflags); -	return 0; +        return 0;  } +  int  ra_create (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,             mode_t mode, fd_t *fd, dict_t *params)  { -	STACK_WIND (frame, ra_create_cbk, -		    FIRST_CHILD(this), -		    FIRST_CHILD(this)->fops->create, -		    loc, flags, mode, fd, params); +        STACK_WIND (frame, ra_create_cbk, +                    FIRST_CHILD(this), +                    FIRST_CHILD(this)->fops->create, +                    loc, flags, mode, fd, params); -	return 0; +        return 0;  }  /* free cache pages between offset and offset+size, @@ -218,109 +219,110 @@ ra_create (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,  static void  flush_region (call_frame_t *frame, ra_file_t *file, off_t offset, off_t size)  { -	ra_page_t *trav = NULL; -	ra_page_t *next = NULL; - -	ra_file_lock (file); -	{ -		trav = file->pages.next; -		while (trav != &file->pages -		       && trav->offset < (offset + size)) { - -			next = trav->next; -			if (trav->offset >= offset && !trav->waitq) { -				ra_page_purge (trav); -			} -			trav = next; -		} -	} -	ra_file_unlock (file); +        ra_page_t *trav = NULL; +        ra_page_t *next = NULL; + +        ra_file_lock (file); +        { +                trav = file->pages.next; +                while (trav != &file->pages +                       && trav->offset < (offset + size)) { + +                        next = trav->next; +                        if (trav->offset >= offset && !trav->waitq) { +                                ra_page_purge (trav); +                        } +                        trav = next; +                } +        } +        ra_file_unlock (file);  }  int  ra_release (xlator_t *this, fd_t *fd)  { -	uint64_t tmp_file = 0; -	int      ret = 0; +        uint64_t tmp_file = 0; +        int      ret      = 0; -	ret = fd_ctx_del (fd, this, &tmp_file); -	 -	if (!ret) { -		ra_file_destroy ((ra_file_t *)(long)tmp_file); -	} +        ret = fd_ctx_del (fd, this, &tmp_file); -	return 0; +        if (!ret) { +                ra_file_destroy ((ra_file_t *)(long)tmp_file); +        } + +        return 0;  }  void  read_ahead (call_frame_t *frame, ra_file_t *file)  { -	off_t      ra_offset = 0; -	size_t     ra_size = 0; -	off_t      trav_offset = 0; -	ra_page_t  *trav = NULL; -	off_t      cap = 0; -	char       fault = 0; - -	if (!file->page_count) -		return; - -	ra_size   = file->page_size * file->page_count; -	ra_offset = floor (file->offset, file->page_size); -	cap       = file->size ? file->size : file->offset + ra_size; - -	while (ra_offset < min (file->offset + ra_size, cap)) { - -		ra_file_lock (file); -		{ -			trav = ra_page_get (file, ra_offset); -		} -		ra_file_unlock (file); - -		if (!trav) -			break; - -		ra_offset += file->page_size; -	} - -	if (trav) -		/* comfortable enough */ -		return; - -	trav_offset = ra_offset; - -	cap  = file->size ? file->size : ra_offset + ra_size; - -	while (trav_offset < min(ra_offset + ra_size, cap)) { -		fault = 0; -		ra_file_lock (file); -		{ -			trav = ra_page_get (file, trav_offset); -			if (!trav) { -				fault = 1; -				trav = ra_page_create (file, trav_offset); -				if (trav)  -					trav->dirty = 1; -			} -		} -		ra_file_unlock (file); - -		if (!trav) { -			/* OUT OF MEMORY */ -			break; -		} - -		if (fault) { -			gf_log (frame->this->name, GF_LOG_TRACE, -				"RA at offset=%"PRId64, trav_offset); -			ra_page_fault (file, frame, trav_offset); -		} -		trav_offset += file->page_size; -	} - -	return; +        off_t      ra_offset   = 0; +        size_t     ra_size     = 0; +        off_t      trav_offset = 0; +        ra_page_t *trav        = NULL; +        off_t      cap         = 0; +        char       fault       = 0; + +        if (!file->page_count) +                return; + +        ra_size   = file->page_size * file->page_count; +        ra_offset = floor (file->offset, file->page_size); +        cap       = file->size ? file->size : file->offset + ra_size; + +        while (ra_offset < min (file->offset + ra_size, cap)) { + +                ra_file_lock (file); +                { +                        trav = ra_page_get (file, ra_offset); +                } +                ra_file_unlock (file); + +                if (!trav) +                        break; + +                ra_offset += file->page_size; +        } + +        if (trav) { +                /* comfortable enough */ +                return; +        } + +        trav_offset = ra_offset; + +        cap  = file->size ? file->size : ra_offset + ra_size; + +        while (trav_offset < min(ra_offset + ra_size, cap)) { +                fault = 0; +                ra_file_lock (file); +                { +                        trav = ra_page_get (file, trav_offset); +                        if (!trav) { +                                fault = 1; +                                trav = ra_page_create (file, trav_offset); +                                if (trav) +                                        trav->dirty = 1; +                        } +                } +                ra_file_unlock (file); + +                if (!trav) { +                        /* OUT OF MEMORY */ +                        break; +                } + +                if (fault) { +                        gf_log (frame->this->name, GF_LOG_TRACE, +                                "RA at offset=%"PRId64, trav_offset); +                        ra_page_fault (file, frame, trav_offset); +                } +                trav_offset += file->page_size; +        } + +        return;  } @@ -329,92 +331,92 @@ ra_need_atime_cbk (call_frame_t *frame, void *cookie, xlator_t *this,                     int32_t op_ret, int32_t op_errno, struct iovec *vector,                     int32_t count, struct iatt *stbuf, struct iobref *iobref)  { -	STACK_DESTROY (frame->root); -	return 0; +        STACK_DESTROY (frame->root); +        return 0;  }  static void  dispatch_requests (call_frame_t *frame, ra_file_t *file)  { -	ra_local_t    *local = NULL; -	ra_conf_t     *conf = NULL; -	off_t         rounded_offset = 0; -	off_t         rounded_end = 0; -	off_t         trav_offset = 0; -	ra_page_t     *trav = NULL; -	call_frame_t  *ra_frame = NULL; -	char          need_atime_update = 1; -	char          fault = 0; - -	local = frame->local; -	conf  = file->conf; - -	rounded_offset = floor (local->offset, file->page_size); -	rounded_end    = roof (local->offset + local->size, file->page_size); - -	trav_offset = rounded_offset; - -	while (trav_offset < rounded_end) { -		fault = 0; - -		ra_file_lock (file); -		{ -			trav = ra_page_get (file, trav_offset); -			if (!trav) { -				trav = ra_page_create (file, trav_offset); -				fault = 1; -				need_atime_update = 0; -			} - -			if (!trav) { +        ra_local_t   *local             = NULL; +        ra_conf_t    *conf              = NULL; +        off_t         rounded_offset    = 0; +        off_t         rounded_end       = 0; +        off_t         trav_offset       = 0; +        ra_page_t    *trav              = NULL; +        call_frame_t *ra_frame          = NULL; +        char          need_atime_update = 1; +        char          fault             = 0; + +        local = frame->local; +        conf  = file->conf; + +        rounded_offset = floor (local->offset, file->page_size); +        rounded_end    = roof (local->offset + local->size, file->page_size); + +        trav_offset = rounded_offset; + +        while (trav_offset < rounded_end) { +                fault = 0; + +                ra_file_lock (file); +                { +                        trav = ra_page_get (file, trav_offset); +                        if (!trav) { +                                trav = ra_page_create (file, trav_offset); +                                fault = 1; +                                need_atime_update = 0; +                        } + +                        if (!trav) {                                  local->op_ret = -1;                                  local->op_errno = ENOMEM; - 				goto unlock; +                                goto unlock; +                        } + +                        if (trav->ready) { +                                gf_log (frame->this->name, GF_LOG_TRACE, +                                        "HIT at offset=%"PRId64".", +                                        trav_offset); +                                ra_frame_fill (trav, frame); +                        } else { +                                gf_log (frame->this->name, GF_LOG_TRACE, +                                        "IN-TRANSIT at offset=%"PRId64".", +                                        trav_offset); +                                ra_wait_on_page (trav, frame); +                                need_atime_update = 0;                          } +                } +        unlock: +                ra_file_unlock (file); + +                if (fault) { +                        gf_log (frame->this->name, GF_LOG_TRACE, +                                "MISS at offset=%"PRId64".", +                                trav_offset); +                        ra_page_fault (file, frame, trav_offset); +                } + +                trav_offset += file->page_size; +        } -			if (trav->ready) { -				gf_log (frame->this->name, GF_LOG_TRACE, -					"HIT at offset=%"PRId64".", -					trav_offset); -				ra_frame_fill (trav, frame); -			} else { -				gf_log (frame->this->name, GF_LOG_TRACE, -					"IN-TRANSIT at offset=%"PRId64".", -					trav_offset); -				ra_wait_on_page (trav, frame); -				need_atime_update = 0; -			} -		} -	unlock: -		ra_file_unlock (file); - -		if (fault) { -			gf_log (frame->this->name, GF_LOG_TRACE, -				"MISS at offset=%"PRId64".", -				trav_offset); -			ra_page_fault (file, frame, trav_offset); -		} - -		trav_offset += file->page_size; -	} - -	if (need_atime_update && conf->force_atime_update) { -		/* TODO: use untimens() since readv() can confuse underlying -		   io-cache and others */ -		ra_frame = copy_frame (frame); +        if (need_atime_update && conf->force_atime_update) { +                /* TODO: use untimens() since readv() can confuse underlying +                   io-cache and others */ +                ra_frame = copy_frame (frame);                  if (ra_frame == NULL) {                          goto out;                  } -		STACK_WIND (ra_frame, ra_need_atime_cbk, -			    FIRST_CHILD (frame->this),  -			    FIRST_CHILD (frame->this)->fops->readv, -			    file->fd, 1, 1); -	} +                STACK_WIND (ra_frame, ra_need_atime_cbk, +                            FIRST_CHILD (frame->this), +                            FIRST_CHILD (frame->this)->fops->readv, +                            file->fd, 1, 1); +        }  out: -	return ; +        return ;  } @@ -423,10 +425,10 @@ ra_readv_disabled_cbk (call_frame_t *frame, void *cookie, xlator_t *this,                         int32_t op_ret, int32_t op_errno, struct iovec *vector,                         int32_t count, struct iatt *stbuf, struct iobref *iobref)  { -	STACK_UNWIND_STRICT (readv, frame, op_ret, op_errno, vector, count, +        STACK_UNWIND_STRICT (readv, frame, op_ret, op_errno, vector, count,                               stbuf, iobref); -	return 0; +        return 0;  } @@ -434,21 +436,21 @@ int  ra_readv (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,            off_t offset)  { -	ra_file_t    *file = NULL; -	ra_local_t   *local = NULL; -	ra_conf_t    *conf = NULL; -	int          op_errno = 0; -	char         expected_offset = 1; -	uint64_t     tmp_file = 0; +        ra_file_t   *file            = NULL; +        ra_local_t  *local           = NULL; +        ra_conf_t   *conf            = NULL; +        int          op_errno        = 0; +        char         expected_offset = 1; +        uint64_t     tmp_file        = 0; -	conf = this->private; +        conf = this->private; -	gf_log (this->name, GF_LOG_TRACE, -		"NEW REQ at offset=%"PRId64" for size=%"GF_PRI_SIZET"", -		offset, size); +        gf_log (this->name, GF_LOG_TRACE, +                "NEW REQ at offset=%"PRId64" for size=%"GF_PRI_SIZET"", +                offset, size); -	fd_ctx_get (fd, this, &tmp_file); -	file = (ra_file_t *)(long)tmp_file; +        fd_ctx_get (fd, this, &tmp_file); +        file = (ra_file_t *)(long)tmp_file;          if (file == NULL) {                  op_errno = EBADF; @@ -457,73 +459,74 @@ ra_readv (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,                  goto unwind;          } -	if (file->offset != offset) { -		gf_log (this->name, GF_LOG_DEBUG, -			"unexpected offset (%"PRId64" != %"PRId64") resetting", -			file->offset, offset); - -		expected_offset = file->expected = file->page_count = 0; -	} else { -		gf_log (this->name, GF_LOG_TRACE, -			"expected offset (%"PRId64") when page_count=%d", -			offset, file->page_count); - -		if (file->expected < (conf->page_size * conf->page_count)) { -			file->expected += size; -			file->page_count = min ((file->expected / file->page_size), -						conf->page_count); -		} -	} - -	if (!expected_offset) { -		flush_region (frame, file, 0, file->pages.prev->offset + 1); -	} - -	if (file->disabled) { -		STACK_WIND (frame, ra_readv_disabled_cbk, -			    FIRST_CHILD (frame->this),  -			    FIRST_CHILD (frame->this)->fops->readv, -			    file->fd, size, offset); -		return 0; -	} - -	local = (void *) GF_CALLOC (1, sizeof (*local), +        if (file->offset != offset) { +                gf_log (this->name, GF_LOG_DEBUG, +                        "unexpected offset (%"PRId64" != %"PRId64") resetting", +                        file->offset, offset); + +                expected_offset = file->expected = file->page_count = 0; +        } else { +                gf_log (this->name, GF_LOG_TRACE, +                        "expected offset (%"PRId64") when page_count=%d", +                        offset, file->page_count); + +                if (file->expected < (conf->page_size * conf->page_count)) { +                        file->expected += size; +                        file->page_count = min ((file->expected +                                                 / file->page_size), +                                                conf->page_count); +                } +        } + +        if (!expected_offset) { +                flush_region (frame, file, 0, file->pages.prev->offset + 1); +        } + +        if (file->disabled) { +                STACK_WIND (frame, ra_readv_disabled_cbk, +                            FIRST_CHILD (frame->this), +                            FIRST_CHILD (frame->this)->fops->readv, +                            file->fd, size, offset); +                return 0; +        } + +        local = (void *) GF_CALLOC (1, sizeof (*local),                                      gf_ra_mt_ra_local_t); -	if (!local) { -		gf_log (this->name, GF_LOG_ERROR, -			"out of memory"); -		op_errno = ENOMEM; -		goto unwind; -	} +        if (!local) { +                gf_log (this->name, GF_LOG_ERROR, +                        "out of memory"); +                op_errno = ENOMEM; +                goto unwind; +        } -	local->fd         = fd; -	local->offset     = offset; -	local->size       = size; -	local->wait_count = 1; +        local->fd         = fd; +        local->offset     = offset; +        local->size       = size; +        local->wait_count = 1; -	local->fill.next  = &local->fill; -	local->fill.prev  = &local->fill; +        local->fill.next  = &local->fill; +        local->fill.prev  = &local->fill; -	pthread_mutex_init (&local->local_lock, NULL); +        pthread_mutex_init (&local->local_lock, NULL); -	frame->local = local; +        frame->local = local; -	dispatch_requests (frame, file); +        dispatch_requests (frame, file); -	flush_region (frame, file, 0, floor (offset, file->page_size)); +        flush_region (frame, file, 0, floor (offset, file->page_size));          read_ahead (frame, file); -	ra_frame_return (frame); +        ra_frame_return (frame); -	file->offset = offset + size; +        file->offset = offset + size; -	return 0; +        return 0;  unwind: -	STACK_UNWIND_STRICT (readv, frame, -1, op_errno, NULL, 0, NULL, NULL); +        STACK_UNWIND_STRICT (readv, frame, -1, op_errno, NULL, 0, NULL, NULL); -	return 0; +        return 0;  } @@ -531,8 +534,8 @@ int  ra_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,                int32_t op_errno)  { -	STACK_UNWIND_STRICT (flush, frame, op_ret, op_errno); -	return 0; +        STACK_UNWIND_STRICT (flush, frame, op_ret, op_errno); +        return 0;  } @@ -541,20 +544,21 @@ int  ra_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)  { -	STACK_UNWIND_STRICT (fsync, frame, op_ret, op_errno, prebuf, postbuf); -	return 0; +        STACK_UNWIND_STRICT (fsync, frame, op_ret, op_errno, prebuf, postbuf); +        return 0;  }  int  ra_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)  { -	ra_file_t *file = NULL; -	uint64_t  tmp_file = 0; -        int32_t   op_errno = 0; +        ra_file_t *file     = NULL; +        uint64_t   tmp_file = 0; +        int32_t    op_errno = 0; + +        fd_ctx_get (fd, this, &tmp_file); -	fd_ctx_get (fd, this, &tmp_file); -	file = (ra_file_t *)(long)tmp_file; +        file = (ra_file_t *)(long)tmp_file;          if (file == NULL) {                  op_errno = EBADF;                  gf_log (this->name, GF_LOG_DEBUG, "flush received on fd with no" @@ -564,14 +568,14 @@ ra_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)          flush_region (frame, file, 0, file->pages.prev->offset+1); -	STACK_WIND (frame, ra_flush_cbk, -		    FIRST_CHILD (this), -		    FIRST_CHILD (this)->fops->flush, -		    fd); -	return 0; +        STACK_WIND (frame, ra_flush_cbk, +                    FIRST_CHILD (this), +                    FIRST_CHILD (this)->fops->flush, +                    fd); +        return 0;  unwind: -	STACK_UNWIND_STRICT (flush, frame, -1, op_errno); +        STACK_UNWIND_STRICT (flush, frame, -1, op_errno);          return 0;  } @@ -579,12 +583,13 @@ unwind:  int  ra_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync)  { -	ra_file_t *file = NULL; -	uint64_t  tmp_file = 0; -        int32_t   op_errno = 0; +        ra_file_t *file     = NULL; +        uint64_t   tmp_file = 0; +        int32_t    op_errno = 0; -	fd_ctx_get (fd, this, &tmp_file); -	file = (ra_file_t *)(long)tmp_file; +        fd_ctx_get (fd, this, &tmp_file); + +        file = (ra_file_t *)(long)tmp_file;          if (file == NULL) {                  op_errno = EBADF;                  gf_log (this->name, GF_LOG_DEBUG, "fsync received on fd with no" @@ -592,18 +597,18 @@ ra_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync)                  goto unwind;          } -	if (file) { -		flush_region (frame, file, 0, file->pages.prev->offset+1); -	} +        if (file) { +                flush_region (frame, file, 0, file->pages.prev->offset+1); +        } -	STACK_WIND (frame, ra_fsync_cbk, -		    FIRST_CHILD (this), -		    FIRST_CHILD (this)->fops->fsync, -		    fd, datasync); -	return 0; +        STACK_WIND (frame, ra_fsync_cbk, +                    FIRST_CHILD (this), +                    FIRST_CHILD (this)->fops->fsync, +                    fd, datasync); +        return 0;  unwind: -	STACK_UNWIND_STRICT (fsync, frame, -1, op_errno, NULL, NULL); +        STACK_UNWIND_STRICT (fsync, frame, -1, op_errno, NULL, NULL);          return 0;  } @@ -613,20 +618,20 @@ ra_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,                 int32_t op_ret, int32_t op_errno, struct iatt *prebuf,                 struct iatt *postbuf)  { -	fd_t      *fd = NULL; -	ra_file_t *file = NULL; -	uint64_t  tmp_file = 0; +        fd_t      *fd       = NULL; +        ra_file_t *file     = NULL; +        uint64_t   tmp_file = 0; -	fd = frame->local; +        fd = frame->local; -	fd_ctx_get (fd, this, &tmp_file); -	file = (ra_file_t *)(long)tmp_file; +        fd_ctx_get (fd, this, &tmp_file); +        file = (ra_file_t *)(long)tmp_file;          flush_region (frame, file, 0, file->pages.prev->offset+1); -	frame->local = NULL; -	STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf); -	return 0; +        frame->local = NULL; +        STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf); +        return 0;  } @@ -634,12 +639,12 @@ int  ra_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,             int32_t count, off_t offset, struct iobref *iobref)  { -	ra_file_t *file = NULL; -	uint64_t  tmp_file = 0; +        ra_file_t *file    = NULL; +        uint64_t  tmp_file = 0;          int32_t   op_errno = 0; -	fd_ctx_get (fd, this, &tmp_file); -	file = (ra_file_t *)(long)tmp_file; +        fd_ctx_get (fd, this, &tmp_file); +        file = (ra_file_t *)(long)tmp_file;          if (file == NULL) {                  op_errno = EBADF;                  gf_log (this->name, GF_LOG_DEBUG, "writev received on fd with" @@ -652,17 +657,17 @@ ra_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,          /* reset the read-ahead counters too */          file->expected = file->page_count = 0; -	frame->local = fd; +        frame->local = fd; -	STACK_WIND (frame, ra_writev_cbk, -		    FIRST_CHILD(this), -		    FIRST_CHILD(this)->fops->writev, -		    fd, vector, count, offset, iobref); +        STACK_WIND (frame, ra_writev_cbk, +                    FIRST_CHILD(this), +                    FIRST_CHILD(this)->fops->writev, +                    fd, vector, count, offset, iobref); -	return 0; +        return 0;  unwind: -	STACK_UNWIND_STRICT (writev, frame, -1, op_errno, NULL, NULL); +        STACK_UNWIND_STRICT (writev, frame, -1, op_errno, NULL, NULL);          return 0;  } @@ -672,122 +677,123 @@ ra_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,                   struct iatt *postbuf)  { -	STACK_UNWIND_STRICT (truncate, frame, op_ret, op_errno, prebuf, +        STACK_UNWIND_STRICT (truncate, frame, op_ret, op_errno, prebuf,                               postbuf); -	return 0; +        return 0;  }  int  ra_attr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, -	     int32_t op_ret, int32_t op_errno, struct iatt *buf) +             int32_t op_ret, int32_t op_errno, struct iatt *buf)  { -	STACK_UNWIND_STRICT (stat, frame, op_ret, op_errno, buf); -	return 0; +        STACK_UNWIND_STRICT (stat, frame, op_ret, op_errno, buf); +        return 0;  }  int  ra_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset)  { -	ra_file_t *file = NULL; -	fd_t      *iter_fd = NULL; -	inode_t   *inode = NULL; -	uint64_t  tmp_file = 0; - -	inode = loc->inode; - -	LOCK (&inode->lock); -	{ -		list_for_each_entry (iter_fd, &inode->fd_list, inode_list) { -			fd_ctx_get (iter_fd, this, &tmp_file); -			file = (ra_file_t *)(long)tmp_file; - -			if (!file) -				continue; -			flush_region (frame, file, 0, -				      file->pages.prev->offset + 1); -		} -	} -	UNLOCK (&inode->lock); - -	STACK_WIND (frame, ra_truncate_cbk, -		    FIRST_CHILD (this), -		    FIRST_CHILD (this)->fops->truncate, -		    loc, offset); -	return 0; +        ra_file_t *file     = NULL; +        fd_t      *iter_fd  = NULL; +        inode_t   *inode    = NULL; +        uint64_t   tmp_file = 0; + +        inode = loc->inode; + +        LOCK (&inode->lock); +        { +                list_for_each_entry (iter_fd, &inode->fd_list, inode_list) { +                        fd_ctx_get (iter_fd, this, &tmp_file); +                        file = (ra_file_t *)(long)tmp_file; + +                        if (!file) +                                continue; +                        flush_region (frame, file, 0, +                                      file->pages.prev->offset + 1); +                } +        } +        UNLOCK (&inode->lock); + +        STACK_WIND (frame, ra_truncate_cbk, +                    FIRST_CHILD (this), +                    FIRST_CHILD (this)->fops->truncate, +                    loc, offset); +        return 0;  }  int  ra_fstat (call_frame_t *frame, xlator_t *this, fd_t *fd)  { -	ra_file_t *file = NULL; -	fd_t      *iter_fd = NULL; -	inode_t   *inode = NULL; -	uint64_t  tmp_file = 0; - -	inode = fd->inode; - -	LOCK (&inode->lock); -	{ -		list_for_each_entry (iter_fd, &inode->fd_list, inode_list) { -			fd_ctx_get (iter_fd, this, &tmp_file); -			file = (ra_file_t *)(long)tmp_file; - -			if (!file) -				continue; -			flush_region (frame, file, 0, -				      file->pages.prev->offset + 1); -		} -	} -	UNLOCK (&inode->lock); - -	STACK_WIND (frame, ra_attr_cbk, -		    FIRST_CHILD (this), -		    FIRST_CHILD (this)->fops->fstat, -		    fd); -	return 0; +        ra_file_t *file     = NULL; +        fd_t      *iter_fd  = NULL; +        inode_t   *inode    = NULL; +        uint64_t   tmp_file = 0; + +        inode = fd->inode; + +        LOCK (&inode->lock); +        { +                list_for_each_entry (iter_fd, &inode->fd_list, inode_list) { +                        fd_ctx_get (iter_fd, this, &tmp_file); +                        file = (ra_file_t *)(long)tmp_file; + +                        if (!file) +                                continue; +                        flush_region (frame, file, 0, +                                      file->pages.prev->offset + 1); +                } +        } +        UNLOCK (&inode->lock); + +        STACK_WIND (frame, ra_attr_cbk, +                    FIRST_CHILD (this), +                    FIRST_CHILD (this)->fops->fstat, +                    fd); +        return 0;  }  int  ra_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset)  { -	ra_file_t *file = NULL; -	fd_t      *iter_fd = NULL; -	inode_t   *inode = NULL; -	uint64_t  tmp_file = 0; - -	inode = fd->inode; - -	LOCK (&inode->lock); -	{ -		list_for_each_entry (iter_fd, &inode->fd_list, inode_list) { -			fd_ctx_get (iter_fd, this, &tmp_file); -			file = (ra_file_t *)(long)tmp_file; -			if (!file) -				continue; -			flush_region (frame, file, 0, -				      file->pages.prev->offset + 1); -		} -	} -	UNLOCK (&inode->lock); - -	STACK_WIND (frame, ra_truncate_cbk, -		    FIRST_CHILD (this), -		    FIRST_CHILD (this)->fops->ftruncate, -		    fd, offset); -	return 0; +        ra_file_t *file    = NULL; +        fd_t      *iter_fd = NULL; +        inode_t   *inode   = NULL; +        uint64_t  tmp_file = 0; + +        inode = fd->inode; + +        LOCK (&inode->lock); +        { +                list_for_each_entry (iter_fd, &inode->fd_list, inode_list) { +                        fd_ctx_get (iter_fd, this, &tmp_file); +                        file = (ra_file_t *)(long)tmp_file; +                        if (!file) +                                continue; +                        flush_region (frame, file, 0, +                                      file->pages.prev->offset + 1); +                } +        } +        UNLOCK (&inode->lock); + +        STACK_WIND (frame, ra_truncate_cbk, +                    FIRST_CHILD (this), +                    FIRST_CHILD (this)->fops->ftruncate, +                    fd, offset); +        return 0;  } +  int  ra_priv_dump (xlator_t *this)  { -        ra_conf_t       *conf = NULL; -        int             ret = -1; -        char            key[GF_DUMP_MAX_BUF_LEN]; -        char            key_prefix[GF_DUMP_MAX_BUF_LEN]; +        ra_conf_t       *conf                           = NULL; +        int             ret                             = -1; +        char            key[GF_DUMP_MAX_BUF_LEN]        = {0, }; +        char            key_prefix[GF_DUMP_MAX_BUF_LEN] = {0, };          if (!this)                  return -1; @@ -806,9 +812,7 @@ ra_priv_dump (xlator_t *this)                  return -1;          } - -        gf_proc_dump_build_key (key_prefix, -                                "xlator.performance.read-ahead", +        gf_proc_dump_build_key (key_prefix, "xlator.performance.read-ahead",                                  "priv");          gf_proc_dump_add_section (key_prefix); @@ -824,19 +828,21 @@ ra_priv_dump (xlator_t *this)          return 0;  } +  int32_t  mem_acct_init (xlator_t *this)  {          int     ret = -1; -        if (!this) +        if (!this) {                  return ret; +        }          ret = xlator_mem_acct_init (this, gf_ra_mt_end + 1); -         +          if (ret != 0) {                  gf_log (this->name, GF_LOG_ERROR, "Memory accounting init" -                                "failed"); +                        "failed");                  return ret;          } @@ -846,24 +852,24 @@ mem_acct_init (xlator_t *this)  int  init (xlator_t *this)  { -	ra_conf_t *conf = NULL; -	dict_t    *options = this->options; -	char      *page_count_string = NULL; -        int32_t   ret = -1; - -	if (!this->children || this->children->next) { -		gf_log (this->name,  GF_LOG_ERROR, -			"FATAL: read-ahead not configured with exactly one" +        ra_conf_t *conf              = NULL; +        dict_t    *options           = this->options; +        char      *page_count_string = NULL; +        int32_t    ret               = -1; + +        if (!this->children || this->children->next) { +                gf_log (this->name,  GF_LOG_ERROR, +                        "FATAL: read-ahead not configured with exactly one"                          " child");                  goto out; -	} - -	if (!this->parents) { -		gf_log (this->name, GF_LOG_WARNING, -			"dangling volume. check volfile "); -	} -  -	conf = (void *) GF_CALLOC (1, sizeof (*conf), +        } + +        if (!this->parents) { +                gf_log (this->name, GF_LOG_WARNING, +                        "dangling volume. check volfile "); +        } + +        conf = (void *) GF_CALLOC (1, sizeof (*conf),                                     gf_ra_mt_ra_conf_t);          if (conf == NULL) {                  gf_log (this->name, GF_LOG_ERROR, @@ -871,48 +877,54 @@ init (xlator_t *this)                  goto out;          } -	conf->page_size = this->ctx->page_size; -	conf->page_count = 4; - -	if (dict_get (options, "page-count")) -		page_count_string = data_to_str (dict_get (options,  -							   "page-count")); -	if (page_count_string) -	{ -		if (gf_string2uint_base10 (page_count_string, &conf->page_count) -                    != 0) -		{ -			gf_log ("read-ahead",  -				GF_LOG_ERROR,  -				"invalid number format \"%s\" of \"option " -                                "page-count\"",  -				page_count_string); +        conf->page_size = this->ctx->page_size; +        conf->page_count = 4; + +        if (dict_get (options, "page-count")) { +                page_count_string = data_to_str (dict_get (options, +                                                           "page-count")); +        } + +        if (page_count_string) { +                if (gf_string2uint_base10 (page_count_string, &conf->page_count) +                    != 0) { +                        gf_log ("read-ahead", GF_LOG_ERROR, +                                "invalid number format \"%s\" of \"option " +                                "page-count\"", +                                page_count_string);                          goto out; -		} -		gf_log (this->name, GF_LOG_DEBUG, "Using conf->page_count = %u", -			conf->page_count); -	} -   -	if (dict_get (options, "force-atime-update")) { -		char *force_atime_update_str = data_to_str (dict_get (options, -								      "force-atime-update")); -		if (gf_string2boolean (force_atime_update_str, +                } + +                gf_log (this->name, GF_LOG_DEBUG, "Using conf->page_count = %u", +                        conf->page_count); +        } + +        if (dict_get (options, "force-atime-update")) { +                char *force_atime_update_str = NULL; + +                force_atime_update_str +                        = data_to_str (dict_get (options, +                                                 "force-atime-update")); + +                if (gf_string2boolean (force_atime_update_str,                                         &conf->force_atime_update) == -1) { -			gf_log (this->name, GF_LOG_ERROR, -				"'force-atime-update' takes only boolean " +                        gf_log (this->name, GF_LOG_ERROR, +                                "'force-atime-update' takes only boolean "                                  "options");                          goto out; -		} -		if (conf->force_atime_update) -			gf_log (this->name, GF_LOG_DEBUG, "Forcing atime " +                } + +                if (conf->force_atime_update) { +                        gf_log (this->name, GF_LOG_DEBUG, "Forcing atime "                                  "updates on cache hit"); -	} +                } +        } -	conf->files.next = &conf->files; -	conf->files.prev = &conf->files; +        conf->files.next = &conf->files; +        conf->files.prev = &conf->files; -	pthread_mutex_init (&conf->conf_lock, NULL); -	this->private = conf; +        pthread_mutex_init (&conf->conf_lock, NULL); +        this->private = conf;          ret = 0;  out: @@ -925,35 +937,36 @@ out:          return ret;  } +  void  fini (xlator_t *this)  { -	ra_conf_t *conf = this->private; +        ra_conf_t *conf = this->private;          if (conf == NULL)                  return; -	pthread_mutex_destroy (&conf->conf_lock); -	GF_FREE (conf); +        pthread_mutex_destroy (&conf->conf_lock); +        GF_FREE (conf); -	this->private = NULL; -	return; +        this->private = NULL; +        return;  }  struct xlator_fops fops = { -	.open        = ra_open, -	.create      = ra_create, -	.readv       = ra_readv, -	.writev      = ra_writev, -	.flush       = ra_flush, -	.fsync       = ra_fsync, -	.truncate    = ra_truncate, -	.ftruncate   = ra_ftruncate, -	.fstat       = ra_fstat, +        .open        = ra_open, +        .create      = ra_create, +        .readv       = ra_readv, +        .writev      = ra_writev, +        .flush       = ra_flush, +        .fsync       = ra_fsync, +        .truncate    = ra_truncate, +        .ftruncate   = ra_ftruncate, +        .fstat       = ra_fstat,  };  struct xlator_cbks cbks = { -	.release       = ra_release, +        .release       = ra_release,  };  struct xlator_dumpops dumpops = { @@ -961,13 +974,13 @@ struct xlator_dumpops dumpops = {  };  struct volume_options options[] = { -	{ .key  = {"force-atime-update"},  -	  .type = GF_OPTION_TYPE_BOOL  -	}, -	{ .key  = {"page-count"},  -	  .type = GF_OPTION_TYPE_INT,  -	  .min  = 1,  -	  .max  = 16  -	}, -	{ .key = {NULL} }, +        { .key  = {"force-atime-update"}, +          .type = GF_OPTION_TYPE_BOOL +        }, +        { .key  = {"page-count"}, +          .type = GF_OPTION_TYPE_INT, +          .min  = 1, +          .max  = 16 +        }, +        { .key = {NULL} },  }; diff --git a/xlators/performance/read-ahead/src/read-ahead.h b/xlators/performance/read-ahead/src/read-ahead.h index 6eb6ffdc8fc..ed918094fbc 100644 --- a/xlators/performance/read-ahead/src/read-ahead.h +++ b/xlators/performance/read-ahead/src/read-ahead.h @@ -41,77 +41,77 @@ struct ra_waitq;  struct ra_waitq { -	struct ra_waitq *next; -	void            *data; +        struct ra_waitq *next; +        void            *data;  };  struct ra_fill { -	struct ra_fill *next; -	struct ra_fill *prev; -	off_t           offset; -	size_t          size; -	struct iovec   *vector; -	int32_t         count; +        struct ra_fill *next; +        struct ra_fill *prev; +        off_t           offset; +        size_t          size; +        struct iovec   *vector; +        int32_t         count;          struct iobref  *iobref;  };  struct ra_local { -	mode_t            mode; -	struct ra_fill    fill; -	off_t             offset; -	size_t            size; -	int32_t           op_ret; -	int32_t           op_errno; -	off_t             pending_offset; -	size_t            pending_size; -	fd_t             *fd; -	int32_t           wait_count; -	pthread_mutex_t   local_lock; +        mode_t            mode; +        struct ra_fill    fill; +        off_t             offset; +        size_t            size; +        int32_t           op_ret; +        int32_t           op_errno; +        off_t             pending_offset; +        size_t            pending_size; +        fd_t             *fd; +        int32_t           wait_count; +        pthread_mutex_t   local_lock;  };  struct ra_page { -	struct ra_page   *next; -	struct ra_page   *prev; -	struct ra_file   *file; -	char              dirty; -	char              ready; -	struct iovec     *vector; -	int32_t           count; -	off_t             offset; -	size_t            size; -	struct ra_waitq  *waitq; +        struct ra_page   *next; +        struct ra_page   *prev; +        struct ra_file   *file; +        char              dirty; +        char              ready; +        struct iovec     *vector; +        int32_t           count; +        off_t             offset; +        size_t            size; +        struct ra_waitq  *waitq;          struct iobref    *iobref;  };  struct ra_file { -	struct ra_file    *next; -	struct ra_file    *prev; -	struct ra_conf    *conf; -	fd_t              *fd; -	int                disabled; -	size_t             expected; -	struct ra_page     pages; -	off_t              offset; -	size_t             size; -	int32_t            refcount; -	pthread_mutex_t    file_lock; -	struct iatt        stbuf; -	uint64_t           page_size; -	uint32_t           page_count; +        struct ra_file    *next; +        struct ra_file    *prev; +        struct ra_conf    *conf; +        fd_t              *fd; +        int                disabled; +        size_t             expected; +        struct ra_page     pages; +        off_t              offset; +        size_t             size; +        int32_t            refcount; +        pthread_mutex_t    file_lock; +        struct iatt        stbuf; +        uint64_t           page_size; +        uint32_t           page_count;  };  struct ra_conf { -	uint64_t          page_size; -	uint32_t          page_count; -	void             *cache_block; -	struct ra_file    files; -	gf_boolean_t      force_atime_update; -	pthread_mutex_t   conf_lock; +        uint64_t          page_size; +        uint32_t          page_count; +        void             *cache_block; +        struct ra_file    files; +        gf_boolean_t      force_atime_update; +        pthread_mutex_t   conf_lock;  }; @@ -124,19 +124,19 @@ typedef struct ra_fill ra_fill_t;  ra_page_t *  ra_page_get (ra_file_t *file, -	     off_t offset); +             off_t offset);  ra_page_t *  ra_page_create (ra_file_t *file, -		off_t offset); +                off_t offset);  void  ra_page_fault (ra_file_t *file, -	       call_frame_t *frame, -	       off_t offset); +               call_frame_t *frame, +               off_t offset);  void  ra_wait_on_page (ra_page_t *page, -		 call_frame_t *frame); +                 call_frame_t *frame);  ra_waitq_t *  ra_page_wakeup (ra_page_t *page); @@ -146,8 +146,8 @@ ra_page_flush (ra_page_t *page);  ra_waitq_t *  ra_page_error (ra_page_t *page, -	       int32_t op_ret, -	       int32_t op_errno); +               int32_t op_ret, +               int32_t op_errno);  void  ra_page_purge (ra_page_t *page); @@ -156,7 +156,7 @@ ra_frame_return (call_frame_t *frame);  void  ra_frame_fill (ra_page_t *page, -	       call_frame_t *frame); +               call_frame_t *frame);  void  ra_file_destroy (ra_file_t *file); @@ -164,36 +164,36 @@ ra_file_destroy (ra_file_t *file);  static inline void  ra_file_lock (ra_file_t *file)  { -	pthread_mutex_lock (&file->file_lock); +        pthread_mutex_lock (&file->file_lock);  }  static inline void  ra_file_unlock (ra_file_t *file)  { -	pthread_mutex_unlock (&file->file_lock); +        pthread_mutex_unlock (&file->file_lock);  }  static inline void  ra_conf_lock (ra_conf_t *conf)  { -	pthread_mutex_lock (&conf->conf_lock); +        pthread_mutex_lock (&conf->conf_lock);  }  static inline void  ra_conf_unlock (ra_conf_t *conf)  { -	pthread_mutex_unlock (&conf->conf_lock); +        pthread_mutex_unlock (&conf->conf_lock);  }  static inline void  ra_local_lock (ra_local_t *local)  { -	pthread_mutex_lock (&local->local_lock); +        pthread_mutex_lock (&local->local_lock);  }  static inline void  ra_local_unlock (ra_local_t *local)  { -	pthread_mutex_unlock (&local->local_lock); +        pthread_mutex_unlock (&local->local_lock);  }  #endif /* __READ_AHEAD_H */  | 
