Date: Tue, 8 Jun 2010 17:54:28 +0000 (UTC) From: Konstantin Belousov <kib@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: svn commit: r208920 - head/sys/kern Message-ID: <201006081754.o58HsSkK071689@svn.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: kib Date: Tue Jun 8 17:54:28 2010 New Revision: 208920 URL: http://svn.freebsd.org/changeset/base/208920 Log: Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks Modified: head/sys/kern/vfs_bio.c Modified: head/sys/kern/vfs_bio.c ============================================================================== --- head/sys/kern/vfs_bio.c Tue Jun 8 17:51:21 2010 (r208919) +++ head/sys/kern/vfs_bio.c Tue Jun 8 17:54:28 2010 (r208920) @@ -102,8 +102,8 @@ static void vm_hold_load_pages(struct bu static void vfs_page_set_valid(struct buf *bp, vm_ooffset_t off, vm_page_t m); static void vfs_page_set_validclean(struct buf *bp, vm_ooffset_t off, vm_page_t m); -static void vfs_clean_pages(struct buf *bp); -static void vfs_setdirty(struct buf *bp); +static void vfs_drain_busy_pages(struct buf *bp); +static void vfs_clean_pages_dirty_buf(struct buf *bp); static void vfs_setdirty_locked_object(struct buf *bp); static void vfs_vmio_release(struct buf *bp); static int vfs_bio_clcheck(struct vnode *vp, int size, @@ -1025,18 +1025,17 @@ bdwrite(struct buf *bp) } /* - * Set the *dirty* buffer range based upon the VM system dirty pages. - */ - vfs_setdirty(bp); - - /* - * We need to do this here to satisfy the vnode_pager and the - * pageout daemon, so that it thinks that the pages have been - * "cleaned". Note that since the pages are in a delayed write - * buffer -- the VFS layer "will" see that the pages get written - * out on the next sync, or perhaps the cluster will be completed. + * Set the *dirty* buffer range based upon the VM system dirty + * pages. + * + * Mark the buffer pages as clean. We need to do this here to + * satisfy the vnode_pager and the pageout daemon, so that it + * thinks that the pages have been "cleaned". Note that since + * the pages are in a delayed write buffer -- the VFS layer + * "will" see that the pages get written out on the next sync, + * or perhaps the cluster will be completed. */ - vfs_clean_pages(bp); + vfs_clean_pages_dirty_buf(bp); bqrelse(bp); /* @@ -2398,31 +2397,44 @@ notinmem: } /* - * vfs_setdirty: - * - * Sets the dirty range for a buffer based on the status of the dirty - * bits in the pages comprising the buffer. + * Set the dirty range for a buffer based on the status of the dirty + * bits in the pages comprising the buffer. The range is limited + * to the size of the buffer. * - * The range is limited to the size of the buffer. + * Tell the VM system that the pages associated with this buffer + * are clean. This is used for delayed writes where the data is + * going to go to disk eventually without additional VM intevention. * - * This routine is primarily used by NFS, but is generalized for the - * B_VMIO case. + * Note that while we only really need to clean through to b_bcount, we + * just go ahead and clean through to b_bufsize. */ static void -vfs_setdirty(struct buf *bp) +vfs_clean_pages_dirty_buf(struct buf *bp) { + vm_ooffset_t foff, noff, eoff; + vm_page_t m; + int i; - /* - * Degenerate case - empty buffer - */ - if (bp->b_bufsize == 0) + if ((bp->b_flags & B_VMIO) == 0 || bp->b_bufsize == 0) return; - if ((bp->b_flags & B_VMIO) == 0) - return; + foff = bp->b_offset; + KASSERT(bp->b_offset != NOOFFSET, + ("vfs_clean_pages_dirty_buf: no buffer offset")); VM_OBJECT_LOCK(bp->b_bufobj->bo_object); + vfs_drain_busy_pages(bp); vfs_setdirty_locked_object(bp); + for (i = 0; i < bp->b_npages; i++) { + noff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; + eoff = noff; + if (eoff > bp->b_offset + bp->b_bufsize) + eoff = bp->b_offset + bp->b_bufsize; + m = bp->b_pages[i]; + vfs_page_set_validclean(bp, foff, m); + /* vm_page_clear_dirty(m, foff & PAGE_MASK, eoff - foff); */ + foff = noff; + } VM_OBJECT_UNLOCK(bp->b_bufobj->bo_object); } @@ -3533,6 +3545,31 @@ vfs_page_set_validclean(struct buf *bp, } /* + * Ensure that all buffer pages are not busied by VPO_BUSY flag. If + * any page is busy, drain the flag. + */ +static void +vfs_drain_busy_pages(struct buf *bp) +{ + vm_page_t m; + int i, last_busied; + + VM_OBJECT_LOCK_ASSERT(bp->b_bufobj->bo_object, MA_OWNED); + last_busied = 0; + for (i = 0; i < bp->b_npages; i++) { + m = bp->b_pages[i]; + if ((m->oflags & VPO_BUSY) != 0) { + for (; last_busied < i; last_busied++) + vm_page_busy(bp->b_pages[last_busied]); + while ((m->oflags & VPO_BUSY) != 0) + vm_page_sleep(m, "vbpage"); + } + } + for (i = 0; i < last_busied; i++) + vm_page_wakeup(bp->b_pages[i]); +} + +/* * This routine is called before a device strategy routine. * It is used to tell the VM system that paging I/O is in * progress, and treat the pages associated with the buffer @@ -3560,15 +3597,9 @@ vfs_busy_pages(struct buf *bp, int clear KASSERT(bp->b_offset != NOOFFSET, ("vfs_busy_pages: no buffer offset")); VM_OBJECT_LOCK(obj); + vfs_drain_busy_pages(bp); if (bp->b_bufsize != 0) vfs_setdirty_locked_object(bp); -retry: - for (i = 0; i < bp->b_npages; i++) { - m = bp->b_pages[i]; - - if (vm_page_sleep_if_busy(m, FALSE, "vbpage")) - goto retry; - } bogus = 0; for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; @@ -3609,42 +3640,6 @@ retry: } /* - * Tell the VM system that the pages associated with this buffer - * are clean. This is used for delayed writes where the data is - * going to go to disk eventually without additional VM intevention. - * - * Note that while we only really need to clean through to b_bcount, we - * just go ahead and clean through to b_bufsize. - */ -static void -vfs_clean_pages(struct buf *bp) -{ - int i; - vm_ooffset_t foff, noff, eoff; - vm_page_t m; - - if (!(bp->b_flags & B_VMIO)) - return; - - foff = bp->b_offset; - KASSERT(bp->b_offset != NOOFFSET, - ("vfs_clean_pages: no buffer offset")); - VM_OBJECT_LOCK(bp->b_bufobj->bo_object); - for (i = 0; i < bp->b_npages; i++) { - m = bp->b_pages[i]; - noff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; - eoff = noff; - - if (eoff > bp->b_offset + bp->b_bufsize) - eoff = bp->b_offset + bp->b_bufsize; - vfs_page_set_validclean(bp, foff, m); - /* vm_page_clear_dirty(m, foff & PAGE_MASK, eoff - foff); */ - foff = noff; - } - VM_OBJECT_UNLOCK(bp->b_bufobj->bo_object); -} - -/* * vfs_bio_set_valid: * * Set the range within the buffer to valid. The range is
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201006081754.o58HsSkK071689>