From: Steven Whitehouse Date: Mon, 21 Nov 2011 11:12:26 +0000 Subject: [Cluster-devel] [GFS2 PATCH TRY2] GFS2: split function rgblk_search In-Reply-To: References: Message-ID: <1321873946.2710.23.camel@menhir> List-Id: To: cluster-devel.redhat.com MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Hi, On Fri, 2011-11-18 at 13:15 -0500, Bob Peterson wrote: > ----- Original Message ----- > | Hi, > | > | On Wed, 2011-11-16 at 14:59 -0500, Bob Peterson wrote: > | > ----- Original Message ----- > | > | Hi, > | > | > | > | As a follow up to this patch, I'd quite like to see rgblk_search > | > | split > | > | into two functions. One to do the search and another to actually > | > | make > | > | changes to the bitmap when a block is found. That should help > | > | development of further alloc changes on top. > | > > | > Hi, > | > > | > So something like the patch that follows? > | > > | > This patch splits function rgblk_search into two functions: > | > the finding and the bit setting function, now called > | > gfs2_alloc_extent. > | > > | Yes, but lets make gfs2_alloc_extent() a totally separate function > | and > | not call it from rgblk_search, that way when looking for unlinked > | inodes > | we don't need to care about gfs2_alloc_extent at all. The only issue > | is > | passing a couple of things from rgblk_search to gfs2_alloc_extent > | (the > | bi (either by index or by pointer) and the current position in the > | bitmap. > | > | However my thought was to separate these two completely rather than > | leaving one calling the other, > | > | Steve. > > Hi, > > This is another revision of a previously posted patch based on Steve's > suggestions. The idea here is to split function rgblk_search into two > functions: one to search for a block in the proper state and another > to assign multiple blocks starting at that location. > > Splitting the two allowed for some code optimizations. > > Regards, > > Bob Peterson > Red Hat File Systems > > Signed-off-by: Bob Peterson > -- > GFS2: split function rgblk_search > > This patch splits function rgblk_search into a function that finds > blocks to allocate (rgblk_search) and a function that assigns those > blocks (gfs2_alloc_extent). > > diff --git a/fs/gfs2/rgrp.c b/fs/gfs2/rgrp.c > index b8935af..8f87cd0 100644 > --- a/fs/gfs2/rgrp.c > +++ b/fs/gfs2/rgrp.c > @@ -65,8 +65,8 @@ static const char valid_change[16] = { > }; > > static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal, > - unsigned char old_state, bool dinode, > - unsigned int *ndata); > + unsigned char old_state, bool dinode, > + struct gfs2_bitmap **rbi); > > /** > * gfs2_setbit - Set a bit in the bitmaps > @@ -912,19 +912,20 @@ static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip > u32 goal = 0, block; > u64 no_addr; > struct gfs2_sbd *sdp = rgd->rd_sbd; > - unsigned int n; > struct gfs2_glock *gl; > struct gfs2_inode *ip; > int error; > int found = 0; > + struct gfs2_bitmap *bi; > > while (goal < rgd->rd_data) { > down_write(&sdp->sd_log_flush_lock); > - n = 1; > - block = rgblk_search(rgd, goal, GFS2_BLKST_UNLINKED, 0, &n); > + block = rgblk_search(rgd, goal, GFS2_BLKST_UNLINKED, 0, &bi); > up_write(&sdp->sd_log_flush_lock); > if (block == BFITNOENT) > break; > + > + block = GFS2_BI2RGD_BLK(bi, block); > /* rgblk_search can return a block < goal, so we need to > keep it marching forward. */ > no_addr = block + rgd->rd_data0; > @@ -1109,38 +1110,35 @@ static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block) > } > > /** > - * rgblk_search - find a block in @old_state, change allocation > - * state to @new_state > + * rgblk_search - find a block in @old_state > * @rgd: the resource group descriptor > * @goal: the goal block within the RG (start here to search for avail block) > * @old_state: GFS2_BLKST_XXX the before-allocation state to find > * @dinode: TRUE if the first block we allocate is for a dinode > - * @n: The extent length > + * @rbi: address of the pointer to the bitmap containing the block found > * > * Walk rgrp's bitmap to find bits that represent a block in @old_state. > - * Add the found bitmap buffer to the transaction. > - * Set the found bits to @new_state to change block's allocation state. > * > * This function never fails, because we wouldn't call it unless we > * know (from reservation results, etc.) that a block is available. > * > - * Scope of @goal and returned block is just within rgrp, not the whole > - * filesystem. > + * Scope of @goal is just within rgrp, not the whole filesystem. > + * Scope of @returned block is just within bitmap, not the whole filesystem. > * > - * Returns: the block number allocated > + * Returns: the block number found relative to the bitmap rbi > */ > > static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal, > - unsigned char old_state, bool dinode, unsigned int *n) > + unsigned char old_state, bool dinode, > + struct gfs2_bitmap **rbi) > { > struct gfs2_bitmap *bi = NULL; > const u32 length = rgd->rd_length; > u32 blk = BFITNOENT; > unsigned int buf, x; > - const unsigned int elen = *n; > const u8 *buffer = NULL; > > - *n = 0; > + *rbi = NULL; > /* Find bitmap block that contains bits for goal block */ > for (buf = 0; buf < length; buf++) { > bi = rgd->rd_bits + buf; > @@ -1187,12 +1185,31 @@ skip: > goal = 0; > } > > - if (blk == BFITNOENT) > - return blk; > + if (blk != BFITNOENT) > + *rbi = bi; > + Since bi is set to NULL above, this could presumably be unconditional > + return blk; > +} > > - if (old_state == GFS2_BLKST_UNLINKED) > - goto out; > +/** > + * gfs2_alloc_extent - allocate an extent from a given bitmap > + * @rgd: the resource group descriptor > + * @bi: the bitmap within the rgrp > + * @blk: the block within the bitmap > + * @dinode: TRUE if the first block we allocate is for a dinode > + * @n: The extent length > + * > + * Add the found bitmap buffer to the transaction. > + * Set the found bits to @new_state to change block's allocation state. > + */ > +static void gfs2_alloc_extent(struct gfs2_rgrpd *rgd, struct gfs2_bitmap *bi, > + u32 blk, bool dinode, unsigned int *n) > +{ > + const unsigned int elen = *n; > + u32 goal; > + const u8 *buffer = NULL; > > + buffer = bi->bi_bh->b_data + bi->bi_offset; > gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1); > gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset, > bi, blk, dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); > @@ -1210,8 +1227,6 @@ skip: > bi, goal, GFS2_BLKST_USED); > (*n)++; > } > -out: > - return (bi->bi_start * GFS2_NBBY) + blk; > } > > /** > @@ -1319,6 +1334,7 @@ int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *ndata, > u32 goal, extlen, blk; /* block, within the rgrp scope */ > u64 block; /* block, within the file system scope */ > int error; > + struct gfs2_bitmap *bi; > > /* Only happens if there is a bug in gfs2, return something distinctive > * to ensure that it is noticed. > @@ -1333,12 +1349,16 @@ int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *ndata, > else > goal = rgd->rd_last_alloc; > > - blk = rgblk_search(rgd, goal, GFS2_BLKST_FREE, dinode, ndata); > + blk = rgblk_search(rgd, goal, GFS2_BLKST_FREE, dinode, &bi); > > + *ndata = 0; > /* Since all blocks are reserved in advance, this shouldn't happen */ > if (blk == BFITNOENT) > goto rgrp_error; > > + gfs2_alloc_extent(rgd, bi, blk, dinode, ndata); > + > + blk = GFS2_BI2RGD_BLK(bi, blk); > rgd->rd_last_alloc = blk; > block = rgd->rd_data0 + blk; If rd_last_alloc was set in gfs2_alloc_extent() then it could be set (correctly) to the end of the extent, and not the beginning. Also gfs2_alloc_extent could then return the starting block of the extent, so these lines would just become: block = gfs2_alloc_extent(rgd, bi, blk, dinode, ndata); which would be a bit neater. > if (!dinode) { > diff --git a/fs/gfs2/rgrp.h b/fs/gfs2/rgrp.h > index b3b61b8..aacb3fd 100644 > --- a/fs/gfs2/rgrp.h > +++ b/fs/gfs2/rgrp.h > @@ -12,6 +12,8 @@ > > #include > > +#define GFS2_BI2RGD_BLK(bi, blk) (bi->bi_start * GFS2_NBBY) + blk > + Could you make this an inline function in rgrp.c, since it is not needed outside of that file? A macro written like this can potentially have some nasty side effects, as there are not enough brackets in the expression, Steve. > struct gfs2_rgrpd; > struct gfs2_sbd; > struct gfs2_holder;