From patchwork Thu Jul 21 19:34:19 2011 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Jim Rees X-Patchwork-Id: 996942 Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by demeter2.kernel.org (8.14.4/8.14.4) with ESMTP id p6LJYV0q015518 for ; Thu, 21 Jul 2011 19:34:49 GMT Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753715Ab1GUTes (ORCPT ); Thu, 21 Jul 2011 15:34:48 -0400 Received: from merit-proxy01.merit.edu ([207.75.116.193]:46759 "EHLO merit-proxy01.merit.edu" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753046Ab1GUTer (ORCPT ); Thu, 21 Jul 2011 15:34:47 -0400 Received: from localhost (localhost.localdomain [127.0.0.1]) by merit-proxy01.merit.edu (Postfix) with ESMTP id DF0C12039AE0; Thu, 21 Jul 2011 15:34:46 -0400 (EDT) X-Virus-Scanned: amavisd-new at merit-proxy01.merit.edu Received: from merit-proxy01.merit.edu ([127.0.0.1]) by localhost (merit-proxy01.merit.edu [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id N8sM619OJB87; Thu, 21 Jul 2011 15:34:46 -0400 (EDT) Received: from merit.edu (castelmezzano.citi.umich.edu [141.212.112.22]) by merit-proxy01.merit.edu (Postfix) with ESMTPSA id 362BA2039AED; Thu, 21 Jul 2011 15:34:46 -0400 (EDT) From: Jim Rees To: Trond Myklebust Cc: linux-nfs@vger.kernel.org, peter honeyman Subject: [PATCH v2 19/25] pnfsblock: cleanup_layoutcommit Date: Thu, 21 Jul 2011 15:34:19 -0400 Message-Id: <1311276865-29484-20-git-send-email-rees@umich.edu> X-Mailer: git-send-email 1.7.4.1 In-Reply-To: <1311276865-29484-1-git-send-email-rees@umich.edu> References: <1311276865-29484-1-git-send-email-rees@umich.edu> Sender: linux-nfs-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-nfs@vger.kernel.org X-Greylist: IP, sender and recipient auto-whitelisted, not delayed by milter-greylist-4.2.6 (demeter2.kernel.org [140.211.167.43]); Thu, 21 Jul 2011 19:34:49 +0000 (UTC) From: Fred Isaman In blocklayout driver. There are two things happening while layoutcommit/cleanup. 1. the modified extents are encoded. 2. On cleanup the extents are put back on the layout rw extents list, for reads. In the new system where actual xdr encoding is done in encode_layoutcommit() directly into xdr buffer, these are the new commit stages: 1. On setup_layoutcommit, the range is adjusted as before and a structure is allocated for communication with bl_encode_layoutcommit && bl_cleanup_layoutcommit (Generic layer provides a void-star to hang it on) 2. bl_encode_layoutcommit is called to do the actual encoding directly into xdr. The commit-extent-list is not freed and is stored on above structure. FIXME: The code is not yet converted to the new XDR cleanup 3. On cleanup the commit-extent-list is put back by a call to set_to_rw() as before, but with no need for XDR decoding of the list as before. And the commit-extent-list is freed. Finally allocated structure is freed. [SQUASHME: pnfs: blocklayout: port block layout code] Signed-off-by: Peng Tao [pnfsblock: SQUASHME: adjust to API change] Signed-off-by: Fred Isaman [blocklayout: encode_layoutcommit implementation] Signed-off-by: Boaz Harrosh [pnfsblock: fix bug setting up layoutcommit.] Signed-off-by: Tao Guo [pnfsblock: cleanup_layoutcommit wants a status parameter] Signed-off-by: Boaz Harrosh Signed-off-by: Benny Halevy Signed-off-by: Benny Halevy --- fs/nfs/blocklayout/blocklayout.c | 2 + fs/nfs/blocklayout/blocklayout.h | 3 + fs/nfs/blocklayout/extents.c | 210 ++++++++++++++++++++++++++++++++++++++ 3 files changed, 215 insertions(+), 0 deletions(-) diff --git a/fs/nfs/blocklayout/blocklayout.c b/fs/nfs/blocklayout/blocklayout.c index 36fd19c..300a678 100644 --- a/fs/nfs/blocklayout/blocklayout.c +++ b/fs/nfs/blocklayout/blocklayout.c @@ -168,6 +168,8 @@ static void bl_cleanup_layoutcommit(struct pnfs_layout_hdr *lo, struct nfs4_layoutcommit_data *lcdata) { + dprintk("%s enter\n", __func__); + clean_pnfs_block_layoutupdate(BLK_LO2EXT(lo), &lcdata->args, lcdata->res.status); } static void free_blk_mountid(struct block_mount_id *mid) diff --git a/fs/nfs/blocklayout/blocklayout.h b/fs/nfs/blocklayout/blocklayout.h index edbe6a9..197f919 100644 --- a/fs/nfs/blocklayout/blocklayout.h +++ b/fs/nfs/blocklayout/blocklayout.h @@ -197,6 +197,9 @@ int is_sector_initialized(struct pnfs_inval_markings *marks, sector_t isect); int encode_pnfs_block_layoutupdate(struct pnfs_block_layout *bl, struct xdr_stream *xdr, const struct nfs4_layoutcommit_args *arg); +void clean_pnfs_block_layoutupdate(struct pnfs_block_layout *bl, + const struct nfs4_layoutcommit_args *arg, + int status); int bl_add_merge_extent(struct pnfs_block_layout *bl, struct pnfs_block_extent *new); diff --git a/fs/nfs/blocklayout/extents.c b/fs/nfs/blocklayout/extents.c index 462a270..483c235 100644 --- a/fs/nfs/blocklayout/extents.c +++ b/fs/nfs/blocklayout/extents.c @@ -329,6 +329,73 @@ void print_clist(struct list_head *list, unsigned int count) } } +/* Note: In theory, we should do more checking that devid's match between + * old and new, but if they don't, the lists are too corrupt to salvage anyway. + */ +/* Note this is very similar to bl_add_merge_extent */ +static void add_to_commitlist(struct pnfs_block_layout *bl, + struct pnfs_block_short_extent *new) +{ + struct list_head *clist = &bl->bl_commit; + struct pnfs_block_short_extent *old, *save; + sector_t end = new->bse_f_offset + new->bse_length; + + dprintk("%s enter\n", __func__); + print_short_extent(new); + print_clist(clist, bl->bl_count); + bl->bl_count++; + /* Scan for proper place to insert, extending new to the left + * as much as possible. + */ + list_for_each_entry_safe(old, save, clist, bse_node) { + if (new->bse_f_offset < old->bse_f_offset) + break; + if (end <= old->bse_f_offset + old->bse_length) { + /* Range is already in list */ + bl->bl_count--; + kfree(new); + return; + } else if (new->bse_f_offset <= + old->bse_f_offset + old->bse_length) { + /* new overlaps or abuts existing be */ + if (new->bse_mdev == old->bse_mdev) { + /* extend new to fully replace old */ + new->bse_length += new->bse_f_offset - + old->bse_f_offset; + new->bse_f_offset = old->bse_f_offset; + list_del(&old->bse_node); + bl->bl_count--; + kfree(old); + } + } + } + /* Note that if we never hit the above break, old will not point to a + * valid extent. However, in that case &old->bse_node==list. + */ + list_add_tail(&new->bse_node, &old->bse_node); + /* Scan forward for overlaps. If we find any, extend new and + * remove the overlapped extent. + */ + old = list_prepare_entry(new, clist, bse_node); + list_for_each_entry_safe_continue(old, save, clist, bse_node) { + if (end < old->bse_f_offset) + break; + /* new overlaps or abuts old */ + if (new->bse_mdev == old->bse_mdev) { + if (end < old->bse_f_offset + old->bse_length) { + /* extend new to fully cover old */ + end = old->bse_f_offset + old->bse_length; + new->bse_length = end - new->bse_f_offset; + } + list_del(&old->bse_node); + bl->bl_count--; + kfree(old); + } + } + dprintk("%s: after merging\n", __func__); + print_clist(clist, bl->bl_count); +} + static void print_bl_extent(struct pnfs_block_extent *be) { dprintk("PRINT EXTENT extent %p\n", be); @@ -547,6 +614,34 @@ bl_find_get_extent(struct pnfs_block_layout *bl, sector_t isect, return ret; } +/* Similar to bl_find_get_extent, but called with lock held, and ignores cow */ +static struct pnfs_block_extent * +bl_find_get_extent_locked(struct pnfs_block_layout *bl, sector_t isect) +{ + struct pnfs_block_extent *be, *ret = NULL; + int i; + + dprintk("%s enter with isect %llu\n", __func__, (u64)isect); + for (i = 0; i < EXTENT_LISTS; i++) { + if (ret) + break; + list_for_each_entry_reverse(be, &bl->bl_extents[i], be_node) { + if (isect >= be->be_f_offset + be->be_length) + break; + if (isect >= be->be_f_offset) { + /* We have found an extent */ + dprintk("%s Get %p (%i)\n", __func__, be, + atomic_read(&be->be_refcnt.refcount)); + kref_get(&be->be_refcnt); + ret = be; + break; + } + } + } + print_bl_extent(ret); + return ret; +} + int encode_pnfs_block_layoutupdate(struct pnfs_block_layout *bl, struct xdr_stream *xdr, @@ -636,3 +731,118 @@ _front_merge(struct pnfs_block_extent *be, struct list_head *head, kfree(storage); return be; } + +static u64 +set_to_rw(struct pnfs_block_layout *bl, u64 offset, u64 length) +{ + u64 rv = offset + length; + struct pnfs_block_extent *be, *e1, *e2, *e3, *new, *old; + struct pnfs_block_extent *children[3]; + struct pnfs_block_extent *merge1 = NULL, *merge2 = NULL; + int i = 0, j; + + dprintk("%s(%llu, %llu)\n", __func__, offset, length); + /* Create storage for up to three new extents e1, e2, e3 */ + e1 = kmalloc(sizeof(*e1), GFP_ATOMIC); + e2 = kmalloc(sizeof(*e2), GFP_ATOMIC); + e3 = kmalloc(sizeof(*e3), GFP_ATOMIC); + /* BUG - we are ignoring any failure */ + if (!e1 || !e2 || !e3) + goto out_nosplit; + + spin_lock(&bl->bl_ext_lock); + be = bl_find_get_extent_locked(bl, offset); + rv = be->be_f_offset + be->be_length; + if (be->be_state != PNFS_BLOCK_INVALID_DATA) { + spin_unlock(&bl->bl_ext_lock); + goto out_nosplit; + } + /* Add e* to children, bumping e*'s krefs */ + if (be->be_f_offset != offset) { + _prep_new_extent(e1, be, be->be_f_offset, + offset - be->be_f_offset, + PNFS_BLOCK_INVALID_DATA); + children[i++] = e1; + print_bl_extent(e1); + } else + merge1 = e1; + _prep_new_extent(e2, be, offset, + min(length, be->be_f_offset + be->be_length - offset), + PNFS_BLOCK_READWRITE_DATA); + children[i++] = e2; + print_bl_extent(e2); + if (offset + length < be->be_f_offset + be->be_length) { + _prep_new_extent(e3, be, e2->be_f_offset + e2->be_length, + be->be_f_offset + be->be_length - + offset - length, + PNFS_BLOCK_INVALID_DATA); + children[i++] = e3; + print_bl_extent(e3); + } else + merge2 = e3; + + /* Remove be from list, and insert the e* */ + /* We don't get refs on e*, since this list is the base reference + * set when init'ed. + */ + if (i < 3) + children[i] = NULL; + new = children[0]; + list_replace(&be->be_node, &new->be_node); + bl_put_extent(be); + new = _front_merge(new, &bl->bl_extents[RW_EXTENT], merge1); + for (j = 1; j < i; j++) { + old = new; + new = children[j]; + list_add(&new->be_node, &old->be_node); + } + if (merge2) { + /* This is a HACK, should just create a _back_merge function */ + new = list_entry(new->be_node.next, + struct pnfs_block_extent, be_node); + new = _front_merge(new, &bl->bl_extents[RW_EXTENT], merge2); + } + spin_unlock(&bl->bl_ext_lock); + + /* Since we removed the base reference above, be is now scheduled for + * destruction. + */ + bl_put_extent(be); + dprintk("%s returns %llu after split\n", __func__, rv); + return rv; + + out_nosplit: + kfree(e1); + kfree(e2); + kfree(e3); + dprintk("%s returns %llu without splitting\n", __func__, rv); + return rv; +} + +void +clean_pnfs_block_layoutupdate(struct pnfs_block_layout *bl, + const struct nfs4_layoutcommit_args *arg, + int status) +{ + struct pnfs_block_short_extent *lce, *save; + + dprintk("%s status %d\n", __func__, status); + list_for_each_entry_safe(lce, save, &bl->bl_committing, bse_node) { + if (likely(!status)) { + u64 offset = lce->bse_f_offset; + u64 end = offset + lce->bse_length; + + do { + offset = set_to_rw(bl, offset, end - offset); + } while (offset < end); + list_del(&lce->bse_node); + + kfree(lce); + } else { + list_del(&lce->bse_node); + spin_lock(&bl->bl_ext_lock); + add_to_commitlist(bl, lce); + spin_unlock(&bl->bl_ext_lock); + } + } +}