/* struct vnop_putpages_args { struct vnode *a_vp; vm_page_t *a_m; int a_count; int a_sync; int *a_rtvals; vm_ooffset_t a_offset; }; */ static int fuse_vnop_putpages(struct vop_putpages_args *ap) { struct uio uio; struct iovec iov; vm_offset_t kva; struct buf *bp; int i, error, npages, count; off_t offset; int *rtvals; struct vnode *vp; struct thread *td; struct ucred *cred; vm_page_t *pages; vm_ooffset_t fsize; FS_DEBUG2G("heh\n"); vp = ap->a_vp; KASSERT(vp->v_object, ("objectless vp passed to putpages")); fsize = vp->v_object->un_pager.vnp.vnp_size; td = curthread; /* XXX */ cred = curthread->td_ucred; /* XXX */ pages = ap->a_m; count = ap->a_count; rtvals = ap->a_rtvals; npages = btoc(count); offset = IDX_TO_OFF(pages[0]->pindex); if (!fsess_opt_mmap(vnode_mount(vp))) { FS_DEBUG("called on non-cacheable vnode??\n"); } for (i = 0; i < npages; i++) rtvals[i] = VM_PAGER_AGAIN; /* * When putting pages, do not extend file past EOF. */ if (offset + count > fsize) { count = fsize - offset; if (count < 0) count = 0; } /* * We use only the kva address for the buffer, but this is extremely * convenient and fast. */ bp = getpbuf(&fuse_pbuf_freecnt); kva = (vm_offset_t)bp->b_data; pmap_qenter(kva, pages, npages); PCPU_INC(cnt.v_vnodeout); PCPU_ADD(cnt.v_vnodepgsout, count); iov.iov_base = (caddr_t)kva; iov.iov_len = count; uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = offset; uio.uio_resid = count; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_WRITE; uio.uio_td = td; error = fuse_io_dispatch(vp, &uio, IO_DIRECT, cred); pmap_qremove(kva, npages); relpbuf(bp, &fuse_pbuf_freecnt); if (!error) { int nwritten = round_page(count - uio.uio_resid) / PAGE_SIZE; for (i = 0; i < nwritten; i++) { rtvals[i] = VM_PAGER_OK; VM_OBJECT_WLOCK(pages[i]->object); vm_page_undirty(pages[i]); VM_OBJECT_WUNLOCK(pages[i]->object); } } return rtvals[0]; }
/* * If blocks are contiguous on disk, use this to provide clustered * read ahead. We will read as many blocks as possible sequentially * and then parcel them up into logical blocks in the buffer hash table. */ static struct buf * cluster_rbuild(struct vnode *vp, u_quad_t filesize, daddr_t lbn, daddr_t blkno, long size, int run, int gbflags, struct buf *fbp) { struct buf *bp, *tbp; daddr_t bn; off_t off; long tinc, tsize; int i, inc, j, k, toff; KASSERT(size == vp->v_mount->mnt_stat.f_iosize, ("cluster_rbuild: size %ld != f_iosize %jd\n", size, (intmax_t)vp->v_mount->mnt_stat.f_iosize)); /* * avoid a division */ while ((u_quad_t) size * (lbn + run) > filesize) { --run; } if (fbp) { tbp = fbp; tbp->b_iocmd = BIO_READ; } else { tbp = getblk(vp, lbn, size, 0, 0, gbflags); if (tbp->b_flags & B_CACHE) return tbp; tbp->b_flags |= B_ASYNC | B_RAM; tbp->b_iocmd = BIO_READ; } tbp->b_blkno = blkno; if( (tbp->b_flags & B_MALLOC) || ((tbp->b_flags & B_VMIO) == 0) || (run <= 1) ) return tbp; bp = trypbuf(&cluster_pbuf_freecnt); if (bp == NULL) return tbp; /* * We are synthesizing a buffer out of vm_page_t's, but * if the block size is not page aligned then the starting * address may not be either. Inherit the b_data offset * from the original buffer. */ bp->b_flags = B_ASYNC | B_CLUSTER | B_VMIO; if ((gbflags & GB_UNMAPPED) != 0) { bp->b_data = unmapped_buf; } else { bp->b_data = (char *)((vm_offset_t)bp->b_data | ((vm_offset_t)tbp->b_data & PAGE_MASK)); } bp->b_iocmd = BIO_READ; bp->b_iodone = cluster_callback; bp->b_blkno = blkno; bp->b_lblkno = lbn; bp->b_offset = tbp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("cluster_rbuild: no buffer offset")); pbgetvp(vp, bp); TAILQ_INIT(&bp->b_cluster.cluster_head); bp->b_bcount = 0; bp->b_bufsize = 0; bp->b_npages = 0; inc = btodb(size); for (bn = blkno, i = 0; i < run; ++i, bn += inc) { if (i == 0) { VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object); vfs_drain_busy_pages(tbp); vm_object_pip_add(tbp->b_bufobj->bo_object, tbp->b_npages); for (k = 0; k < tbp->b_npages; k++) vm_page_sbusy(tbp->b_pages[k]); VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object); } else { if ((bp->b_npages * PAGE_SIZE) + round_page(size) > vp->v_mount->mnt_iosize_max) { break; } tbp = getblk(vp, lbn + i, size, 0, 0, GB_LOCK_NOWAIT | (gbflags & GB_UNMAPPED)); /* Don't wait around for locked bufs. */ if (tbp == NULL) break; /* * Stop scanning if the buffer is fully valid * (marked B_CACHE), or locked (may be doing a * background write), or if the buffer is not * VMIO backed. The clustering code can only deal * with VMIO-backed buffers. The bo lock is not * required for the BKGRDINPROG check since it * can not be set without the buf lock. */ if ((tbp->b_vflags & BV_BKGRDINPROG) || (tbp->b_flags & B_CACHE) || (tbp->b_flags & B_VMIO) == 0) { bqrelse(tbp); break; } /* * The buffer must be completely invalid in order to * take part in the cluster. If it is partially valid * then we stop. */ off = tbp->b_offset; tsize = size; VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object); for (j = 0; tsize > 0; j++) { toff = off & PAGE_MASK; tinc = tsize; if (toff + tinc > PAGE_SIZE) tinc = PAGE_SIZE - toff; VM_OBJECT_ASSERT_WLOCKED(tbp->b_pages[j]->object); if ((tbp->b_pages[j]->valid & vm_page_bits(toff, tinc)) != 0) break; if (vm_page_xbusied(tbp->b_pages[j])) break; vm_object_pip_add(tbp->b_bufobj->bo_object, 1); vm_page_sbusy(tbp->b_pages[j]); off += tinc; tsize -= tinc; } if (tsize > 0) { clean_sbusy: vm_object_pip_add(tbp->b_bufobj->bo_object, -j); for (k = 0; k < j; k++) vm_page_sunbusy(tbp->b_pages[k]); VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object); bqrelse(tbp); break; } VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object); /* * Set a read-ahead mark as appropriate */ if ((fbp && (i == 1)) || (i == (run - 1))) tbp->b_flags |= B_RAM; /* * Set the buffer up for an async read (XXX should * we do this only if we do not wind up brelse()ing?). * Set the block number if it isn't set, otherwise * if it is make sure it matches the block number we * expect. */ tbp->b_flags |= B_ASYNC; tbp->b_iocmd = BIO_READ; if (tbp->b_blkno == tbp->b_lblkno) { tbp->b_blkno = bn; } else if (tbp->b_blkno != bn) { VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object); goto clean_sbusy; } } /* * XXX fbp from caller may not be B_ASYNC, but we are going * to biodone() it in cluster_callback() anyway */ BUF_KERNPROC(tbp); TAILQ_INSERT_TAIL(&bp->b_cluster.cluster_head, tbp, b_cluster.cluster_entry); VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object); for (j = 0; j < tbp->b_npages; j += 1) { vm_page_t m; m = tbp->b_pages[j]; if ((bp->b_npages == 0) || (bp->b_pages[bp->b_npages-1] != m)) { bp->b_pages[bp->b_npages] = m; bp->b_npages++; } if (m->valid == VM_PAGE_BITS_ALL) tbp->b_pages[j] = bogus_page; } VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object); /* * Don't inherit tbp->b_bufsize as it may be larger due to * a non-page-aligned size. Instead just aggregate using * 'size'. */ if (tbp->b_bcount != size) printf("warning: tbp->b_bcount wrong %ld vs %ld\n", tbp->b_bcount, size); if (tbp->b_bufsize != size) printf("warning: tbp->b_bufsize wrong %ld vs %ld\n", tbp->b_bufsize, size); bp->b_bcount += size; bp->b_bufsize += size; } /* * Fully valid pages in the cluster are already good and do not need * to be re-read from disk. Replace the page with bogus_page */ VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); for (j = 0; j < bp->b_npages; j++) { VM_OBJECT_ASSERT_WLOCKED(bp->b_pages[j]->object); if (bp->b_pages[j]->valid == VM_PAGE_BITS_ALL) bp->b_pages[j] = bogus_page; } VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); if (bp->b_bufsize > bp->b_kvasize) panic("cluster_rbuild: b_bufsize(%ld) > b_kvasize(%d)\n", bp->b_bufsize, bp->b_kvasize); if (buf_mapped(bp)) { pmap_qenter(trunc_page((vm_offset_t) bp->b_data), (vm_page_t *)bp->b_pages, bp->b_npages); } return (bp); }
static int vnode_pager_generic_getpages_done(struct buf *bp) { vm_object_t object; off_t tfoff, nextoff; int i, error; error = (bp->b_ioflags & BIO_ERROR) != 0 ? EIO : 0; object = bp->b_vp->v_object; if (error == 0 && bp->b_bcount != bp->b_npages * PAGE_SIZE) { if (!buf_mapped(bp)) { bp->b_data = bp->b_kvabase; pmap_qenter((vm_offset_t)bp->b_data, bp->b_pages, bp->b_npages); } bzero(bp->b_data + bp->b_bcount, PAGE_SIZE * bp->b_npages - bp->b_bcount); } if (buf_mapped(bp)) { pmap_qremove((vm_offset_t)bp->b_data, bp->b_npages); bp->b_data = unmapped_buf; } VM_OBJECT_WLOCK(object); for (i = 0, tfoff = IDX_TO_OFF(bp->b_pages[0]->pindex); i < bp->b_npages; i++, tfoff = nextoff) { vm_page_t mt; nextoff = tfoff + PAGE_SIZE; mt = bp->b_pages[i]; if (nextoff <= object->un_pager.vnp.vnp_size) { /* * Read filled up entire page. */ mt->valid = VM_PAGE_BITS_ALL; KASSERT(mt->dirty == 0, ("%s: page %p is dirty", __func__, mt)); KASSERT(!pmap_page_is_mapped(mt), ("%s: page %p is mapped", __func__, mt)); } else { /* * Read did not fill up entire page. * * Currently we do not set the entire page valid, * we just try to clear the piece that we couldn't * read. */ vm_page_set_valid_range(mt, 0, object->un_pager.vnp.vnp_size - tfoff); KASSERT((mt->dirty & vm_page_bits(0, object->un_pager.vnp.vnp_size - tfoff)) == 0, ("%s: page %p is dirty", __func__, mt)); } if (i != bp->b_pager.pg_reqpage) vm_page_readahead_finish(mt); } VM_OBJECT_WUNLOCK(object); if (error != 0) printf("%s: I/O read error %d\n", __func__, error); return (error); }
/* struct vnop_getpages_args { struct vnode *a_vp; vm_page_t *a_m; int a_count; int a_reqpage; }; */ static int fuse_vnop_getpages(struct vop_getpages_args *ap) { int i, error, nextoff, size, toff, count, npages; struct uio uio; struct iovec iov; vm_offset_t kva; struct buf *bp; struct vnode *vp; struct thread *td; struct ucred *cred; vm_page_t *pages; FS_DEBUG2G("heh\n"); vp = ap->a_vp; KASSERT(vp->v_object, ("objectless vp passed to getpages")); td = curthread; /* XXX */ cred = curthread->td_ucred; /* XXX */ pages = ap->a_m; npages = ap->a_count; if (!fsess_opt_mmap(vnode_mount(vp))) { FS_DEBUG("called on non-cacheable vnode??\n"); return (VM_PAGER_ERROR); } /* * If the last page is partially valid, just return it and allow * the pager to zero-out the blanks. Partially valid pages can * only occur at the file EOF. * * XXXGL: is that true for FUSE, which is a local filesystem, * but still somewhat disconnected from the kernel? */ VM_OBJECT_WLOCK(vp->v_object); if (pages[npages - 1]->valid != 0 && --npages == 0) goto out; VM_OBJECT_WUNLOCK(vp->v_object); /* * We use only the kva address for the buffer, but this is extremely * convenient and fast. */ bp = getpbuf(&fuse_pbuf_freecnt); kva = (vm_offset_t)bp->b_data; pmap_qenter(kva, pages, npages); PCPU_INC(cnt.v_vnodein); PCPU_ADD(cnt.v_vnodepgsin, npages); count = npages << PAGE_SHIFT; iov.iov_base = (caddr_t)kva; iov.iov_len = count; uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = IDX_TO_OFF(pages[0]->pindex); uio.uio_resid = count; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_READ; uio.uio_td = td; error = fuse_io_dispatch(vp, &uio, IO_DIRECT, cred); pmap_qremove(kva, npages); relpbuf(bp, &fuse_pbuf_freecnt); if (error && (uio.uio_resid == count)) { FS_DEBUG("error %d\n", error); return VM_PAGER_ERROR; } /* * Calculate the number of bytes read and validate only that number * of bytes. Note that due to pending writes, size may be 0. This * does not mean that the remaining data is invalid! */ size = count - uio.uio_resid; VM_OBJECT_WLOCK(vp->v_object); fuse_vm_page_lock_queues(); for (i = 0, toff = 0; i < npages; i++, toff = nextoff) { vm_page_t m; nextoff = toff + PAGE_SIZE; m = pages[i]; if (nextoff <= size) { /* * Read operation filled an entire page */ m->valid = VM_PAGE_BITS_ALL; KASSERT(m->dirty == 0, ("fuse_getpages: page %p is dirty", m)); } else if (size > toff) { /* * Read operation filled a partial page. */ m->valid = 0; vm_page_set_valid_range(m, 0, size - toff); KASSERT(m->dirty == 0, ("fuse_getpages: page %p is dirty", m)); } else { /* * Read operation was short. If no error occurred * we may have hit a zero-fill section. We simply * leave valid set to 0. */ ; } } fuse_vm_page_unlock_queues(); out: VM_OBJECT_WUNLOCK(vp->v_object); if (ap->a_rbehind) *ap->a_rbehind = 0; if (ap->a_rahead) *ap->a_rahead = 0; return (VM_PAGER_OK); }
/* * small block filesystem vnode pager input */ static int vnode_pager_input_smlfs(vm_object_t object, vm_page_t m) { struct vnode *vp; struct bufobj *bo; struct buf *bp; struct sf_buf *sf; daddr_t fileaddr; vm_offset_t bsize; vm_page_bits_t bits; int error, i; error = 0; vp = object->handle; if (vp->v_iflag & VI_DOOMED) return VM_PAGER_BAD; bsize = vp->v_mount->mnt_stat.f_iosize; VOP_BMAP(vp, 0, &bo, 0, NULL, NULL); sf = sf_buf_alloc(m, 0); for (i = 0; i < PAGE_SIZE / bsize; i++) { vm_ooffset_t address; bits = vm_page_bits(i * bsize, bsize); if (m->valid & bits) continue; address = IDX_TO_OFF(m->pindex) + i * bsize; if (address >= object->un_pager.vnp.vnp_size) { fileaddr = -1; } else { error = vnode_pager_addr(vp, address, &fileaddr, NULL); if (error) break; } if (fileaddr != -1) { bp = getpbuf(&vnode_pbuf_freecnt); /* build a minimal buffer header */ bp->b_iocmd = BIO_READ; bp->b_iodone = bdone; KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred")); KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred")); bp->b_rcred = crhold(curthread->td_ucred); bp->b_wcred = crhold(curthread->td_ucred); bp->b_data = (caddr_t)sf_buf_kva(sf) + i * bsize; bp->b_blkno = fileaddr; pbgetbo(bo, bp); bp->b_vp = vp; bp->b_bcount = bsize; bp->b_bufsize = bsize; bp->b_runningbufspace = bp->b_bufsize; atomic_add_long(&runningbufspace, bp->b_runningbufspace); /* do the input */ bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); bwait(bp, PVM, "vnsrd"); if ((bp->b_ioflags & BIO_ERROR) != 0) error = EIO; /* * free the buffer header back to the swap buffer pool */ bp->b_vp = NULL; pbrelbo(bp); relpbuf(bp, &vnode_pbuf_freecnt); if (error) break; } else bzero((caddr_t)sf_buf_kva(sf) + i * bsize, bsize); KASSERT((m->dirty & bits) == 0, ("vnode_pager_input_smlfs: page %p is dirty", m)); VM_OBJECT_WLOCK(object); m->valid |= bits; VM_OBJECT_WUNLOCK(object); } sf_buf_free(sf); if (error) { return VM_PAGER_ERROR; } return VM_PAGER_OK; }
/* * This is now called from local media FS's to operate against their * own vnodes if they fail to implement VOP_GETPAGES. */ int vnode_pager_generic_getpages(struct vnode *vp, vm_page_t *m, int bytecount, int reqpage, vop_getpages_iodone_t iodone, void *arg) { vm_object_t object; struct bufobj *bo; struct buf *bp; daddr_t firstaddr, reqblock; off_t foff, pib; int pbefore, pafter, i, size, bsize, first, last, *freecnt; int count, error, before, after, secmask; KASSERT(vp->v_type != VCHR && vp->v_type != VBLK, ("vnode_pager_generic_getpages does not support devices")); if (vp->v_iflag & VI_DOOMED) return (VM_PAGER_BAD); object = vp->v_object; count = bytecount / PAGE_SIZE; bsize = vp->v_mount->mnt_stat.f_iosize; /* * Synchronous and asynchronous paging operations use different * free pbuf counters. This is done to avoid asynchronous requests * to consume all pbufs. * Allocate the pbuf at the very beginning of the function, so that * if we are low on certain kind of pbufs don't even proceed to BMAP, * but sleep. */ freecnt = iodone != NULL ? &vnode_async_pbuf_freecnt : &vnode_pbuf_freecnt; bp = getpbuf(freecnt); /* * Get the underlying device blocks for the file with VOP_BMAP(). * If the file system doesn't support VOP_BMAP, use old way of * getting pages via VOP_READ. */ error = VOP_BMAP(vp, IDX_TO_OFF(m[reqpage]->pindex) / bsize, &bo, &reqblock, &after, &before); if (error == EOPNOTSUPP) { relpbuf(bp, freecnt); VM_OBJECT_WLOCK(object); for (i = 0; i < count; i++) if (i != reqpage) { vm_page_lock(m[i]); vm_page_free(m[i]); vm_page_unlock(m[i]); } PCPU_INC(cnt.v_vnodein); PCPU_INC(cnt.v_vnodepgsin); error = vnode_pager_input_old(object, m[reqpage]); VM_OBJECT_WUNLOCK(object); return (error); } else if (error != 0) { relpbuf(bp, freecnt); vm_pager_free_nonreq(object, m, reqpage, count, FALSE); return (VM_PAGER_ERROR); /* * If the blocksize is smaller than a page size, then use * special small filesystem code. */ } else if ((PAGE_SIZE / bsize) > 1) { relpbuf(bp, freecnt); vm_pager_free_nonreq(object, m, reqpage, count, FALSE); PCPU_INC(cnt.v_vnodein); PCPU_INC(cnt.v_vnodepgsin); return (vnode_pager_input_smlfs(object, m[reqpage])); } /* * Since the caller has busied the requested page, that page's valid * field will not be changed by other threads. */ vm_page_assert_xbusied(m[reqpage]); /* * If we have a completely valid page available to us, we can * clean up and return. Otherwise we have to re-read the * media. */ if (m[reqpage]->valid == VM_PAGE_BITS_ALL) { relpbuf(bp, freecnt); vm_pager_free_nonreq(object, m, reqpage, count, FALSE); return (VM_PAGER_OK); } else if (reqblock == -1) { relpbuf(bp, freecnt); pmap_zero_page(m[reqpage]); KASSERT(m[reqpage]->dirty == 0, ("vnode_pager_generic_getpages: page %p is dirty", m)); VM_OBJECT_WLOCK(object); m[reqpage]->valid = VM_PAGE_BITS_ALL; vm_pager_free_nonreq(object, m, reqpage, count, TRUE); VM_OBJECT_WUNLOCK(object); return (VM_PAGER_OK); } else if (m[reqpage]->valid != 0) { VM_OBJECT_WLOCK(object); m[reqpage]->valid = 0; VM_OBJECT_WUNLOCK(object); } pib = IDX_TO_OFF(m[reqpage]->pindex) % bsize; pbefore = ((daddr_t)before * bsize + pib) / PAGE_SIZE; pafter = ((daddr_t)(after + 1) * bsize - pib) / PAGE_SIZE - 1; first = reqpage < pbefore ? 0 : reqpage - pbefore; last = reqpage + pafter >= count ? count - 1 : reqpage + pafter; if (first > 0 || last + 1 < count) { VM_OBJECT_WLOCK(object); for (i = 0; i < first; i++) { vm_page_lock(m[i]); vm_page_free(m[i]); vm_page_unlock(m[i]); } for (i = last + 1; i < count; i++) { vm_page_lock(m[i]); vm_page_free(m[i]); vm_page_unlock(m[i]); } VM_OBJECT_WUNLOCK(object); } /* * here on direct device I/O */ firstaddr = reqblock; firstaddr += pib / DEV_BSIZE; firstaddr -= IDX_TO_OFF(reqpage - first) / DEV_BSIZE; /* * The first and last page have been calculated now, move * input pages to be zero based, and adjust the count. */ m += first; reqpage -= first; count = last - first + 1; /* * calculate the file virtual address for the transfer */ foff = IDX_TO_OFF(m[0]->pindex); /* * calculate the size of the transfer */ size = count * PAGE_SIZE; KASSERT(count > 0, ("zero count")); if ((foff + size) > object->un_pager.vnp.vnp_size) size = object->un_pager.vnp.vnp_size - foff; KASSERT(size > 0, ("zero size")); /* * round up physical size for real devices. */ secmask = bo->bo_bsize - 1; KASSERT(secmask < PAGE_SIZE && secmask > 0, ("vnode_pager_generic_getpages: sector size %d too large", secmask + 1)); size = (size + secmask) & ~secmask; /* * and map the pages to be read into the kva, if the filesystem * requires mapped buffers. */ if ((vp->v_mount->mnt_kern_flag & MNTK_UNMAPPED_BUFS) != 0 && unmapped_buf_allowed) { bp->b_data = unmapped_buf; bp->b_offset = 0; } else { bp->b_data = bp->b_kvabase; pmap_qenter((vm_offset_t)bp->b_data, m, count); } /* build a minimal buffer header */ bp->b_iocmd = BIO_READ; KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred")); KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred")); bp->b_rcred = crhold(curthread->td_ucred); bp->b_wcred = crhold(curthread->td_ucred); bp->b_blkno = firstaddr; pbgetbo(bo, bp); bp->b_vp = vp; bp->b_bcount = size; bp->b_bufsize = size; bp->b_runningbufspace = bp->b_bufsize; for (i = 0; i < count; i++) bp->b_pages[i] = m[i]; bp->b_npages = count; bp->b_pager.pg_reqpage = reqpage; atomic_add_long(&runningbufspace, bp->b_runningbufspace); PCPU_INC(cnt.v_vnodein); PCPU_ADD(cnt.v_vnodepgsin, count); /* do the input */ bp->b_iooffset = dbtob(bp->b_blkno); if (iodone != NULL) { /* async */ bp->b_pager.pg_iodone = iodone; bp->b_caller1 = arg; bp->b_iodone = vnode_pager_generic_getpages_done_async; bp->b_flags |= B_ASYNC; BUF_KERNPROC(bp); bstrategy(bp); /* Good bye! */ } else { bp->b_iodone = bdone; bstrategy(bp); bwait(bp, PVM, "vnread"); error = vnode_pager_generic_getpages_done(bp); for (i = 0; i < bp->b_npages; i++) bp->b_pages[i] = NULL; bp->b_vp = NULL; pbrelbo(bp); relpbuf(bp, &vnode_pbuf_freecnt); } return (error != 0 ? VM_PAGER_ERROR : VM_PAGER_OK); }
/* * Lets the VM system know about a change in size for a file. * We adjust our own internal size and flush any cached pages in * the associated object that are affected by the size change. * * Note: this routine may be invoked as a result of a pager put * operation (possibly at object termination time), so we must be careful. */ void vnode_pager_setsize(struct vnode *vp, vm_ooffset_t nsize) { vm_object_t object; vm_page_t m; vm_pindex_t nobjsize; if ((object = vp->v_object) == NULL) return; /* ASSERT_VOP_ELOCKED(vp, "vnode_pager_setsize and not locked vnode"); */ VM_OBJECT_WLOCK(object); if (object->type == OBJT_DEAD) { VM_OBJECT_WUNLOCK(object); return; } KASSERT(object->type == OBJT_VNODE, ("not vnode-backed object %p", object)); if (nsize == object->un_pager.vnp.vnp_size) { /* * Hasn't changed size */ VM_OBJECT_WUNLOCK(object); return; } nobjsize = OFF_TO_IDX(nsize + PAGE_MASK); if (nsize < object->un_pager.vnp.vnp_size) { /* * File has shrunk. Toss any cached pages beyond the new EOF. */ if (nobjsize < object->size) vm_object_page_remove(object, nobjsize, object->size, 0); /* * this gets rid of garbage at the end of a page that is now * only partially backed by the vnode. * * XXX for some reason (I don't know yet), if we take a * completely invalid page and mark it partially valid * it can screw up NFS reads, so we don't allow the case. */ if ((nsize & PAGE_MASK) && (m = vm_page_lookup(object, OFF_TO_IDX(nsize))) != NULL && m->valid != 0) { int base = (int)nsize & PAGE_MASK; int size = PAGE_SIZE - base; /* * Clear out partial-page garbage in case * the page has been mapped. */ pmap_zero_page_area(m, base, size); /* * Update the valid bits to reflect the blocks that * have been zeroed. Some of these valid bits may * have already been set. */ vm_page_set_valid_range(m, base, size); /* * Round "base" to the next block boundary so that the * dirty bit for a partially zeroed block is not * cleared. */ base = roundup2(base, DEV_BSIZE); /* * Clear out partial-page dirty bits. * * note that we do not clear out the valid * bits. This would prevent bogus_page * replacement from working properly. */ vm_page_clear_dirty(m, base, PAGE_SIZE - base); } else if ((nsize & PAGE_MASK) && vm_page_is_cached(object, OFF_TO_IDX(nsize))) { vm_page_cache_free(object, OFF_TO_IDX(nsize), nobjsize); } } object->un_pager.vnp.vnp_size = nsize; object->size = nobjsize; VM_OBJECT_WUNLOCK(object); }
static boolean_t vnode_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before, int *after) { struct vnode *vp = object->handle; daddr_t bn; int err; daddr_t reqblock; int poff; int bsize; int pagesperblock, blocksperpage; VM_OBJECT_ASSERT_WLOCKED(object); /* * If no vp or vp is doomed or marked transparent to VM, we do not * have the page. */ if (vp == NULL || vp->v_iflag & VI_DOOMED) return FALSE; /* * If the offset is beyond end of file we do * not have the page. */ if (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size) return FALSE; bsize = vp->v_mount->mnt_stat.f_iosize; pagesperblock = bsize / PAGE_SIZE; blocksperpage = 0; if (pagesperblock > 0) { reqblock = pindex / pagesperblock; } else { blocksperpage = (PAGE_SIZE / bsize); reqblock = pindex * blocksperpage; } VM_OBJECT_WUNLOCK(object); err = VOP_BMAP(vp, reqblock, NULL, &bn, after, before); VM_OBJECT_WLOCK(object); if (err) return TRUE; if (bn == -1) return FALSE; if (pagesperblock > 0) { poff = pindex - (reqblock * pagesperblock); if (before) { *before *= pagesperblock; *before += poff; } if (after) { /* * The BMAP vop can report a partial block in the * 'after', but must not report blocks after EOF. * Assert the latter, and truncate 'after' in case * of the former. */ KASSERT((reqblock + *after) * pagesperblock < roundup2(object->size, pagesperblock), ("%s: reqblock %jd after %d size %ju", __func__, (intmax_t )reqblock, *after, (uintmax_t )object->size)); *after *= pagesperblock; *after += pagesperblock - (poff + 1); if (pindex + *after >= object->size) *after = object->size - 1 - pindex; } } else { if (before) { *before /= blocksperpage; } if (after) { *after /= blocksperpage; } } return TRUE; }
/* * Allocate (or lookup) pager for a vnode. * Handle is a vnode pointer. * * MPSAFE */ vm_object_t vnode_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t offset, struct ucred *cred) { vm_object_t object; struct vnode *vp; /* * Pageout to vnode, no can do yet. */ if (handle == NULL) return (NULL); vp = (struct vnode *) handle; /* * If the object is being terminated, wait for it to * go away. */ retry: while ((object = vp->v_object) != NULL) { VM_OBJECT_WLOCK(object); if ((object->flags & OBJ_DEAD) == 0) break; vm_object_set_flag(object, OBJ_DISCONNECTWNT); VM_OBJECT_SLEEP(object, object, PDROP | PVM, "vadead", 0); } KASSERT(vp->v_usecount != 0, ("vnode_pager_alloc: no vnode reference")); if (object == NULL) { /* * Add an object of the appropriate size */ object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size))); object->un_pager.vnp.vnp_size = size; object->un_pager.vnp.writemappings = 0; object->handle = handle; VI_LOCK(vp); if (vp->v_object != NULL) { /* * Object has been created while we were sleeping */ VI_UNLOCK(vp); VM_OBJECT_WLOCK(object); KASSERT(object->ref_count == 1, ("leaked ref %p %d", object, object->ref_count)); object->type = OBJT_DEAD; object->ref_count = 0; VM_OBJECT_WUNLOCK(object); vm_object_destroy(object); goto retry; } vp->v_object = object; VI_UNLOCK(vp); } else { object->ref_count++; #if VM_NRESERVLEVEL > 0 vm_object_color(object, 0); #endif VM_OBJECT_WUNLOCK(object); } vref(vp); return (object); }
/* * This is now called from local media FS's to operate against their * own vnodes if they fail to implement VOP_PUTPAGES. * * This is typically called indirectly via the pageout daemon and * clustering has already typically occured, so in general we ask the * underlying filesystem to write the data out asynchronously rather * then delayed. */ int vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *ma, int bytecount, int flags, int *rtvals) { int i; vm_object_t object; vm_page_t m; int count; int maxsize, ncount; vm_ooffset_t poffset; struct uio auio; struct iovec aiov; int error; int ioflags; int ppscheck = 0; static struct timeval lastfail; static int curfail; object = vp->v_object; count = bytecount / PAGE_SIZE; for (i = 0; i < count; i++) rtvals[i] = VM_PAGER_ERROR; if ((int64_t)ma[0]->pindex < 0) { printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%lx)\n", (long)ma[0]->pindex, (u_long)ma[0]->dirty); rtvals[0] = VM_PAGER_BAD; return VM_PAGER_BAD; } maxsize = count * PAGE_SIZE; ncount = count; poffset = IDX_TO_OFF(ma[0]->pindex); /* * If the page-aligned write is larger then the actual file we * have to invalidate pages occuring beyond the file EOF. However, * there is an edge case where a file may not be page-aligned where * the last page is partially invalid. In this case the filesystem * may not properly clear the dirty bits for the entire page (which * could be VM_PAGE_BITS_ALL due to the page having been mmap()d). * With the page locked we are free to fix-up the dirty bits here. * * We do not under any circumstances truncate the valid bits, as * this will screw up bogus page replacement. */ VM_OBJECT_WLOCK(object); if (maxsize + poffset > object->un_pager.vnp.vnp_size) { if (object->un_pager.vnp.vnp_size > poffset) { int pgoff; maxsize = object->un_pager.vnp.vnp_size - poffset; ncount = btoc(maxsize); if ((pgoff = (int)maxsize & PAGE_MASK) != 0) { /* * If the object is locked and the following * conditions hold, then the page's dirty * field cannot be concurrently changed by a * pmap operation. */ m = ma[ncount - 1]; vm_page_assert_sbusied(m); KASSERT(!pmap_page_is_write_mapped(m), ("vnode_pager_generic_putpages: page %p is not read-only", m)); vm_page_clear_dirty(m, pgoff, PAGE_SIZE - pgoff); } } else { maxsize = 0; ncount = 0; } if (ncount < count) { for (i = ncount; i < count; i++) { rtvals[i] = VM_PAGER_BAD; } } } VM_OBJECT_WUNLOCK(object); /* * pageouts are already clustered, use IO_ASYNC to force a bawrite() * rather then a bdwrite() to prevent paging I/O from saturating * the buffer cache. Dummy-up the sequential heuristic to cause * large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set, * the system decides how to cluster. */ ioflags = IO_VMIO; if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL)) ioflags |= IO_SYNC; else if ((flags & VM_PAGER_CLUSTER_OK) == 0) ioflags |= IO_ASYNC; ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0; ioflags |= IO_SEQMAX << IO_SEQSHIFT; aiov.iov_base = (caddr_t) 0; aiov.iov_len = maxsize; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = poffset; auio.uio_segflg = UIO_NOCOPY; auio.uio_rw = UIO_WRITE; auio.uio_resid = maxsize; auio.uio_td = (struct thread *) 0; error = VOP_WRITE(vp, &auio, ioflags, curthread->td_ucred); PCPU_INC(cnt.v_vnodeout); PCPU_ADD(cnt.v_vnodepgsout, ncount); if (error) { if ((ppscheck = ppsratecheck(&lastfail, &curfail, 1))) printf("vnode_pager_putpages: I/O error %d\n", error); } if (auio.uio_resid) { if (ppscheck || ppsratecheck(&lastfail, &curfail, 1)) printf("vnode_pager_putpages: residual I/O %zd at %lu\n", auio.uio_resid, (u_long)ma[0]->pindex); } for (i = 0; i < ncount; i++) { rtvals[i] = VM_PAGER_OK; } return rtvals[0]; }
/* * This is now called from local media FS's to operate against their * own vnodes if they fail to implement VOP_GETPAGES. */ int vnode_pager_generic_getpages(struct vnode *vp, vm_page_t *m, int count, int *a_rbehind, int *a_rahead, vop_getpages_iodone_t iodone, void *arg) { vm_object_t object; struct bufobj *bo; struct buf *bp; off_t foff; #ifdef INVARIANTS off_t blkno0; #endif int bsize, pagesperblock, *freecnt; int error, before, after, rbehind, rahead, poff, i; int bytecount, secmask; KASSERT(vp->v_type != VCHR && vp->v_type != VBLK, ("%s does not support devices", __func__)); if (vp->v_iflag & VI_DOOMED) return (VM_PAGER_BAD); object = vp->v_object; foff = IDX_TO_OFF(m[0]->pindex); bsize = vp->v_mount->mnt_stat.f_iosize; pagesperblock = bsize / PAGE_SIZE; KASSERT(foff < object->un_pager.vnp.vnp_size, ("%s: page %p offset beyond vp %p size", __func__, m[0], vp)); KASSERT(count <= sizeof(bp->b_pages), ("%s: requested %d pages", __func__, count)); /* * The last page has valid blocks. Invalid part can only * exist at the end of file, and the page is made fully valid * by zeroing in vm_pager_get_pages(). */ if (m[count - 1]->valid != 0 && --count == 0) { if (iodone != NULL) iodone(arg, m, 1, 0); return (VM_PAGER_OK); } /* * Synchronous and asynchronous paging operations use different * free pbuf counters. This is done to avoid asynchronous requests * to consume all pbufs. * Allocate the pbuf at the very beginning of the function, so that * if we are low on certain kind of pbufs don't even proceed to BMAP, * but sleep. */ freecnt = iodone != NULL ? &vnode_async_pbuf_freecnt : &vnode_pbuf_freecnt; bp = getpbuf(freecnt); /* * Get the underlying device blocks for the file with VOP_BMAP(). * If the file system doesn't support VOP_BMAP, use old way of * getting pages via VOP_READ. */ error = VOP_BMAP(vp, foff / bsize, &bo, &bp->b_blkno, &after, &before); if (error == EOPNOTSUPP) { relpbuf(bp, freecnt); VM_OBJECT_WLOCK(object); for (i = 0; i < count; i++) { PCPU_INC(cnt.v_vnodein); PCPU_INC(cnt.v_vnodepgsin); error = vnode_pager_input_old(object, m[i]); if (error) break; } VM_OBJECT_WUNLOCK(object); return (error); } else if (error != 0) { relpbuf(bp, freecnt); return (VM_PAGER_ERROR); } /* * If the file system supports BMAP, but blocksize is smaller * than a page size, then use special small filesystem code. */ if (pagesperblock == 0) { relpbuf(bp, freecnt); for (i = 0; i < count; i++) { PCPU_INC(cnt.v_vnodein); PCPU_INC(cnt.v_vnodepgsin); error = vnode_pager_input_smlfs(object, m[i]); if (error) break; } return (error); } /* * A sparse file can be encountered only for a single page request, * which may not be preceded by call to vm_pager_haspage(). */ if (bp->b_blkno == -1) { KASSERT(count == 1, ("%s: array[%d] request to a sparse file %p", __func__, count, vp)); relpbuf(bp, freecnt); pmap_zero_page(m[0]); KASSERT(m[0]->dirty == 0, ("%s: page %p is dirty", __func__, m[0])); VM_OBJECT_WLOCK(object); m[0]->valid = VM_PAGE_BITS_ALL; VM_OBJECT_WUNLOCK(object); return (VM_PAGER_OK); } #ifdef INVARIANTS blkno0 = bp->b_blkno; #endif bp->b_blkno += (foff % bsize) / DEV_BSIZE; /* Recalculate blocks available after/before to pages. */ poff = (foff % bsize) / PAGE_SIZE; before *= pagesperblock; before += poff; after *= pagesperblock; after += pagesperblock - (poff + 1); if (m[0]->pindex + after >= object->size) after = object->size - 1 - m[0]->pindex; KASSERT(count <= after + 1, ("%s: %d pages asked, can do only %d", __func__, count, after + 1)); after -= count - 1; /* Trim requested rbehind/rahead to possible values. */ rbehind = a_rbehind ? *a_rbehind : 0; rahead = a_rahead ? *a_rahead : 0; rbehind = min(rbehind, before); rbehind = min(rbehind, m[0]->pindex); rahead = min(rahead, after); rahead = min(rahead, object->size - m[count - 1]->pindex); /* * Check that total amount of pages fit into buf. Trim rbehind and * rahead evenly if not. */ if (rbehind + rahead + count > nitems(bp->b_pages)) { int trim, sum; trim = rbehind + rahead + count - nitems(bp->b_pages) + 1; sum = rbehind + rahead; if (rbehind == before) { /* Roundup rbehind trim to block size. */ rbehind -= roundup(trim * rbehind / sum, pagesperblock); if (rbehind < 0) rbehind = 0; } else rbehind -= trim * rbehind / sum; rahead -= trim * rahead / sum; } KASSERT(rbehind + rahead + count <= nitems(bp->b_pages), ("%s: behind %d ahead %d count %d", __func__, rbehind, rahead, count)); /* * Fill in the bp->b_pages[] array with requested and optional * read behind or read ahead pages. Read behind pages are looked * up in a backward direction, down to a first cached page. Same * for read ahead pages, but there is no need to shift the array * in case of encountering a cached page. */ i = bp->b_npages = 0; if (rbehind) { vm_pindex_t startpindex, tpindex; vm_page_t p; VM_OBJECT_WLOCK(object); startpindex = m[0]->pindex - rbehind; if ((p = TAILQ_PREV(m[0], pglist, listq)) != NULL && p->pindex >= startpindex) startpindex = p->pindex + 1; /* tpindex is unsigned; beware of numeric underflow. */ for (tpindex = m[0]->pindex - 1; tpindex >= startpindex && tpindex < m[0]->pindex; tpindex--, i++) { p = vm_page_alloc(object, tpindex, VM_ALLOC_NORMAL); if (p == NULL) { /* Shift the array. */ for (int j = 0; j < i; j++) bp->b_pages[j] = bp->b_pages[j + tpindex + 1 - startpindex]; break; } bp->b_pages[tpindex - startpindex] = p; } bp->b_pgbefore = i; bp->b_npages += i; bp->b_blkno -= IDX_TO_OFF(i) / DEV_BSIZE; } else bp->b_pgbefore = 0; /* Requested pages. */ for (int j = 0; j < count; j++, i++) bp->b_pages[i] = m[j]; bp->b_npages += count; if (rahead) { vm_pindex_t endpindex, tpindex; vm_page_t p; if (!VM_OBJECT_WOWNED(object)) VM_OBJECT_WLOCK(object); endpindex = m[count - 1]->pindex + rahead + 1; if ((p = TAILQ_NEXT(m[count - 1], listq)) != NULL && p->pindex < endpindex) endpindex = p->pindex; if (endpindex > object->size) endpindex = object->size; for (tpindex = m[count - 1]->pindex + 1; tpindex < endpindex; i++, tpindex++) { p = vm_page_alloc(object, tpindex, VM_ALLOC_NORMAL); if (p == NULL) break; bp->b_pages[i] = p; } bp->b_pgafter = i - bp->b_npages; bp->b_npages = i; } else bp->b_pgafter = 0; if (VM_OBJECT_WOWNED(object)) VM_OBJECT_WUNLOCK(object); /* Report back actual behind/ahead read. */ if (a_rbehind) *a_rbehind = bp->b_pgbefore; if (a_rahead) *a_rahead = bp->b_pgafter; #ifdef INVARIANTS KASSERT(bp->b_npages <= nitems(bp->b_pages), ("%s: buf %p overflowed", __func__, bp)); for (int j = 1, prev = 1; j < bp->b_npages; j++) { if (bp->b_pages[j] == bogus_page) continue; KASSERT(bp->b_pages[j]->pindex - bp->b_pages[prev]->pindex == j - prev, ("%s: pages array not consecutive, bp %p", __func__, bp)); prev = j; } #endif /* * Recalculate first offset and bytecount with regards to read behind. * Truncate bytecount to vnode real size and round up physical size * for real devices. */ foff = IDX_TO_OFF(bp->b_pages[0]->pindex); bytecount = bp->b_npages << PAGE_SHIFT; if ((foff + bytecount) > object->un_pager.vnp.vnp_size) bytecount = object->un_pager.vnp.vnp_size - foff; secmask = bo->bo_bsize - 1; KASSERT(secmask < PAGE_SIZE && secmask > 0, ("%s: sector size %d too large", __func__, secmask + 1)); bytecount = (bytecount + secmask) & ~secmask; /* * And map the pages to be read into the kva, if the filesystem * requires mapped buffers. */ if ((vp->v_mount->mnt_kern_flag & MNTK_UNMAPPED_BUFS) != 0 && unmapped_buf_allowed) { bp->b_data = unmapped_buf; bp->b_offset = 0; } else { bp->b_data = bp->b_kvabase; pmap_qenter((vm_offset_t)bp->b_data, bp->b_pages, bp->b_npages); } /* Build a minimal buffer header. */ bp->b_iocmd = BIO_READ; KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred")); KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred")); bp->b_rcred = crhold(curthread->td_ucred); bp->b_wcred = crhold(curthread->td_ucred); pbgetbo(bo, bp); bp->b_vp = vp; bp->b_bcount = bp->b_bufsize = bp->b_runningbufspace = bytecount; bp->b_iooffset = dbtob(bp->b_blkno); KASSERT(IDX_TO_OFF(m[0]->pindex - bp->b_pages[0]->pindex) == (blkno0 - bp->b_blkno) * DEV_BSIZE + IDX_TO_OFF(m[0]->pindex) % bsize, ("wrong offsets bsize %d m[0] %ju b_pages[0] %ju " "blkno0 %ju b_blkno %ju", bsize, (uintmax_t)m[0]->pindex, (uintmax_t)bp->b_pages[0]->pindex, (uintmax_t)blkno0, (uintmax_t)bp->b_blkno)); atomic_add_long(&runningbufspace, bp->b_runningbufspace); PCPU_INC(cnt.v_vnodein); PCPU_ADD(cnt.v_vnodepgsin, bp->b_npages); if (iodone != NULL) { /* async */ bp->b_pgiodone = iodone; bp->b_caller1 = arg; bp->b_iodone = vnode_pager_generic_getpages_done_async; bp->b_flags |= B_ASYNC; BUF_KERNPROC(bp); bstrategy(bp); return (VM_PAGER_OK); } else { bp->b_iodone = bdone; bstrategy(bp); bwait(bp, PVM, "vnread"); error = vnode_pager_generic_getpages_done(bp); for (i = 0; i < bp->b_npages; i++) bp->b_pages[i] = NULL; bp->b_vp = NULL; pbrelbo(bp); relpbuf(bp, &vnode_pbuf_freecnt); return (error != 0 ? VM_PAGER_ERROR : VM_PAGER_OK); } }