[Patch v4 2/3] CIFS: Add support for direct I/O write
Tom Talpey
ttalpey at microsoft.com
Thu Nov 29 21:45:02 UTC 2018
> -----Original Message-----
> From: linux-cifs-owner at vger.kernel.org <linux-cifs-owner at vger.kernel.org> On
> Behalf Of Long Li
> Sent: Thursday, November 29, 2018 4:30 PM
> To: Pavel Shilovsky <piastryyy at gmail.com>
> Cc: Steve French <sfrench at samba.org>; linux-cifs <linux-cifs at vger.kernel.org>;
> samba-technical <samba-technical at lists.samba.org>; Kernel Mailing List
> <linux-kernel at vger.kernel.org>
> Subject: RE: [Patch v4 2/3] CIFS: Add support for direct I/O write
>
> > Subject: Re: [Patch v4 2/3] CIFS: Add support for direct I/O write
> >
> > ср, 28 нояб. 2018 г. в 18:20, Long Li <longli at microsoft.com>:
> > >
> > > > Subject: Re: [Patch v4 2/3] CIFS: Add support for direct I/O write
> > > >
> > > > ср, 31 окт. 2018 г. в 15:26, Long Li <longli at linuxonhyperv.com>:
> > > > >
> > > > > From: Long Li <longli at microsoft.com>
> > > > >
> > > > > With direct I/O write, user supplied buffers are pinned to the
> > > > > memory and data are transferred directly from user buffers to the
> > transport layer.
> > > > >
> > > > > Change in v3: add support for kernel AIO
> > > > >
> > > > > Change in v4:
> > > > > Refactor common write code to __cifs_writev for direct and non-direct
> > I/O.
> > > > > Retry on direct I/O failure.
> > > > >
> > > > > Signed-off-by: Long Li <longli at microsoft.com>
> > > > > ---
> > > > > fs/cifs/cifsfs.h | 1 +
> > > > > fs/cifs/file.c | 194
> > +++++++++++++++++++++++++++++++++++++++++++----
> > > > --------
> > > > > 2 files changed, 154 insertions(+), 41 deletions(-)
> > > > >
> > > > > diff --git a/fs/cifs/cifsfs.h b/fs/cifs/cifsfs.h index
> > > > > 7fba9aa..e9c5103 100644
> > > > > --- a/fs/cifs/cifsfs.h
> > > > > +++ b/fs/cifs/cifsfs.h
> > > > > @@ -105,6 +105,7 @@ extern ssize_t cifs_user_readv(struct kiocb
> > > > > *iocb, struct iov_iter *to); extern ssize_t
> > > > > cifs_direct_readv(struct kiocb *iocb, struct iov_iter *to);
> > > > > extern ssize_t cifs_strict_readv(struct kiocb *iocb, struct
> > > > > iov_iter *to); extern ssize_t cifs_user_writev(struct kiocb
> > > > > *iocb, struct iov_iter *from);
> > > > > +extern ssize_t cifs_direct_writev(struct kiocb *iocb, struct
> > > > > +iov_iter *from);
> > > > > extern ssize_t cifs_strict_writev(struct kiocb *iocb, struct
> > > > > iov_iter *from); extern int cifs_lock(struct file *, int, struct
> > > > > file_lock *); extern int cifs_fsync(struct file *, loff_t, loff_t,
> > > > > int); diff --git a/fs/cifs/file.c b/fs/cifs/file.c index
> > > > > daab878..1a41c04 100644
> > > > > --- a/fs/cifs/file.c
> > > > > +++ b/fs/cifs/file.c
> > > > > @@ -2524,6 +2524,55 @@ wdata_fill_from_iovec(struct cifs_writedata
> > > > > *wdata, struct iov_iter *from, }
> > > > >
> > > > > static int
> > > > > +cifs_resend_wdata(struct cifs_writedata *wdata, struct list_head
> > > > > +*wdata_list, struct cifs_aio_ctx *ctx) {
> > > > > + int wait_retry = 0;
> > > > > + unsigned int wsize, credits;
> > > > > + int rc;
> > > > > + struct TCP_Server_Info *server =
> > > > > +tlink_tcon(wdata->cfile->tlink)->ses->server;
> > > > > +
> > > > > + /*
> > > > > + * Try to resend this wdata, waiting for credits up to 3 seconds.
> > > > > + * Note: we are attempting to resend the whole wdata not
> > > > > + in
> > > > segments
> > > > > + */
> > > > > + do {
> > > > > + rc = server->ops->wait_mtu_credits(server,
> > > > > + wdata->bytes, &wsize, &credits);
> > > > > +
> > > > > + if (rc)
> > > > > + break;
> > > > > +
> > > > > + if (wsize < wdata->bytes) {
> > > > > + add_credits_and_wake_if(server, credits, 0);
> > > > > + msleep(1000);
> > > > > + wait_retry++;
> > > > > + }
> > > > > + } while (wsize < wdata->bytes && wait_retry < 3);
> > > > > +
> > > > > + if (wsize < wdata->bytes) {
> > > > > + rc = -EBUSY;
> > > > > + goto out;
> > > > > + }
> > > > > +
> > > > > + rc = -EAGAIN;
> > > > > + while (rc == -EAGAIN)
> > > > > + if (!wdata->cfile->invalidHandle ||
> > > > > + !(rc = cifs_reopen_file(wdata->cfile, false)))
> > > > > + rc = server->ops->async_writev(wdata,
> > > > > +
> > > > > + cifs_uncached_writedata_release);
> > > > > +
> > > > > + if (!rc) {
> > > > > + list_add_tail(&wdata->list, wdata_list);
> > > > > + return 0;
> > > > > + }
> > > > > +
> > > > > + add_credits_and_wake_if(server, wdata->credits, 0);
> > > > > +out:
> > > > > + kref_put(&wdata->refcount,
> > > > > +cifs_uncached_writedata_release);
> > > > > +
> > > > > + return rc;
> > > > > +}
> > > > > +
> > > > > +static int
> > > > > cifs_write_from_iter(loff_t offset, size_t len, struct iov_iter *from,
> > > > > struct cifsFileInfo *open_file,
> > > > > struct cifs_sb_info *cifs_sb, struct
> > > > > list_head *wdata_list, @@ -2537,6 +2586,8 @@
> > > > > cifs_write_from_iter(loff_t offset,
> > > > size_t len, struct iov_iter *from,
> > > > > loff_t saved_offset = offset;
> > > > > pid_t pid;
> > > > > struct TCP_Server_Info *server;
> > > > > + struct page **pagevec;
> > > > > + size_t start;
> > > > >
> > > > > if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
> > > > > pid = open_file->pid; @@ -2553,38 +2604,74 @@
> > > > > cifs_write_from_iter(loff_t offset, size_t len,
> > > > struct iov_iter *from,
> > > > > if (rc)
> > > > > break;
> > > > >
> > > > > - nr_pages = get_numpages(wsize, len, &cur_len);
> > > > > - wdata = cifs_writedata_alloc(nr_pages,
> > > > > + if (ctx->direct_io) {
> > > > > + cur_len = iov_iter_get_pages_alloc(
> > > > > + from, &pagevec, wsize, &start);
> > > > > + if (cur_len < 0) {
> > > > > + cifs_dbg(VFS,
> > > > > + "direct_writev couldn't get user pages "
> > > > > + "(rc=%zd) iter type %d iov_offset %zd count"
> > > > > + " %zd\n",
> > > > > + cur_len, from->type,
> > > > > + from->iov_offset, from->count);
> > > > > + dump_stack();
> > > > > + break;
> > > > > + }
> > > > > + iov_iter_advance(from, cur_len);
> > > > > +
> > > > > + nr_pages = (cur_len + start + PAGE_SIZE -
> > > > > + 1) / PAGE_SIZE;
> > > > > +
> > > > > + wdata =
> > > > > + cifs_writedata_direct_alloc(pagevec,
> > > > > cifs_uncached_writev_complete);
> > > > > - if (!wdata) {
> > > > > - rc = -ENOMEM;
> > > > > - add_credits_and_wake_if(server, credits, 0);
> > > > > - break;
> > > > > - }
> > > > > + if (!wdata) {
> > > > > + rc = -ENOMEM;
> > > > > + add_credits_and_wake_if(server, credits, 0);
> > > > > + break;
> > > > > + }
> > > > >
> > > > > - rc = cifs_write_allocate_pages(wdata->pages, nr_pages);
> > > > > - if (rc) {
> > > > > - kfree(wdata);
> > > > > - add_credits_and_wake_if(server, credits, 0);
> > > > > - break;
> > > > > - }
> > > > >
> > > > > - num_pages = nr_pages;
> > > > > - rc = wdata_fill_from_iovec(wdata, from, &cur_len,
> > &num_pages);
> > > > > - if (rc) {
> > > > > - for (i = 0; i < nr_pages; i++)
> > > > > - put_page(wdata->pages[i]);
> > > > > - kfree(wdata);
> > > > > - add_credits_and_wake_if(server, credits, 0);
> > > > > - break;
> > > > > - }
> > > > > + wdata->page_offset = start;
> > > > > + wdata->tailsz =
> > > > > + nr_pages > 1 ?
> > > > > + cur_len - (PAGE_SIZE - start) -
> > > > > + (nr_pages - 2) * PAGE_SIZE :
> > > > > + cur_len;
> > > > > + } else {
> > > > > + nr_pages = get_numpages(wsize, len, &cur_len);
> > > > > + wdata = cifs_writedata_alloc(nr_pages,
> > > > > + cifs_uncached_writev_complete);
> > > > > + if (!wdata) {
> > > > > + rc = -ENOMEM;
> > > > > + add_credits_and_wake_if(server, credits, 0);
> > > > > + break;
> > > > > + }
> > > > >
> > > > > - /*
> > > > > - * Bring nr_pages down to the number of pages we actually
> > used,
> > > > > - * and free any pages that we didn't use.
> > > > > - */
> > > > > - for ( ; nr_pages > num_pages; nr_pages--)
> > > > > - put_page(wdata->pages[nr_pages - 1]);
> > > > > + rc = cifs_write_allocate_pages(wdata->pages, nr_pages);
> > > > > + if (rc) {
> > > > > + kfree(wdata);
> > > > > + add_credits_and_wake_if(server, credits, 0);
> > > > > + break;
> > > > > + }
> > > > > +
> > > > > + num_pages = nr_pages;
> > > > > + rc = wdata_fill_from_iovec(wdata, from,
> > > > > + &cur_len,
> > > > &num_pages);
> > > > > + if (rc) {
> > > > > + for (i = 0; i < nr_pages; i++)
> > > > > + put_page(wdata->pages[i]);
> > > > > + kfree(wdata);
> > > > > + add_credits_and_wake_if(server, credits, 0);
> > > > > + break;
> > > > > + }
> > > > > +
> > > > > + /*
> > > > > + * Bring nr_pages down to the number of
> > > > > + pages we actually
> > > > used,
> > > > > + * and free any pages that we didn't use.
> > > > > + */
> > > > > + for ( ; nr_pages > num_pages; nr_pages--)
> > > > > + put_page(wdata->pages[nr_pages -
> > > > > + 1]);
> > > > > +
> > > > > + wdata->tailsz = cur_len - ((nr_pages - 1) * PAGE_SIZE);
> > > > > + }
> > > > >
> > > > > wdata->sync_mode = WB_SYNC_ALL;
> > > > > wdata->nr_pages = nr_pages; @@ -2593,7 +2680,6 @@
> > > > > cifs_write_from_iter(loff_t offset, size_t len, struct iov_iter *from,
> > > > > wdata->pid = pid;
> > > > > wdata->bytes = cur_len;
> > > > > wdata->pagesz = PAGE_SIZE;
> > > > > - wdata->tailsz = cur_len - ((nr_pages - 1) * PAGE_SIZE);
> > > > > wdata->credits = credits;
> > > > > wdata->ctx = ctx;
> > > > > kref_get(&ctx->refcount); @@ -2668,13 +2754,17 @@
> > > > > static void collect_uncached_write_data(struct cifs_aio_ctx *ctx)
> > > > > INIT_LIST_HEAD(&tmp_list);
> > > > > list_del_init(&wdata->list);
> > > > >
> > > > > - iov_iter_advance(&tmp_from,
> > > > > + if (ctx->direct_io)
> > > > > + rc = cifs_resend_wdata(wdata, &tmp_list, ctx);
> > > > > + else {
> > > > > +
> > > > > + iov_iter_advance(&tmp_from,
> > > > > wdata->offset -
> > > > > ctx->pos);
> > > > >
> > > > > - rc = cifs_write_from_iter(wdata->offset,
> > > > > + rc =
> > > > > + cifs_write_from_iter(wdata->offset,
> > > > > wdata->bytes, &tmp_from,
> > > > > ctx->cfile, cifs_sb, &tmp_list,
> > > > > ctx);
> > > > > + }
> > > > >
> > > > > list_splice(&tmp_list,
> > > > > &ctx->list);
> > > > >
> > > > > @@ -2687,8 +2777,9 @@ static void
> > > > > collect_uncached_write_data(struct
> > > > cifs_aio_ctx *ctx)
> > > > > kref_put(&wdata->refcount,
> > cifs_uncached_writedata_release);
> > > > > }
> > > > >
> > > > > - for (i = 0; i < ctx->npages; i++)
> > > > > - put_page(ctx->bv[i].bv_page);
> > > > > + if (!ctx->direct_io)
> > > > > + for (i = 0; i < ctx->npages; i++)
> > > > > + put_page(ctx->bv[i].bv_page);
> > > > >
> > > > > cifs_stats_bytes_written(tcon, ctx->total_len);
> > > > > set_bit(CIFS_INO_INVALID_MAPPING,
> > > > > &CIFS_I(dentry->d_inode)->flags); @@ -2703,7 +2794,7 @@ static
> > > > > void
> > > > collect_uncached_write_data(struct cifs_aio_ctx *ctx)
> > > > > complete(&ctx->done); }
> > > > >
> > > > > -ssize_t cifs_user_writev(struct kiocb *iocb, struct iov_iter
> > > > > *from)
> > > > > +static ssize_t __cifs_writev(struct kiocb *iocb, struct iov_iter
> > > > > +*from, bool direct)
> > > > > {
> > > > > struct file *file = iocb->ki_filp;
> > > > > ssize_t total_written = 0; @@ -2712,13 +2803,18 @@ ssize_t
> > > > > cifs_user_writev(struct kiocb *iocb,
> > > > struct iov_iter *from)
> > > > > struct cifs_sb_info *cifs_sb;
> > > > > struct cifs_aio_ctx *ctx;
> > > > > struct iov_iter saved_from = *from;
> > > > > + size_t len = iov_iter_count(from);
> > > > > int rc;
> > > > >
> > > > > /*
> > > > > - * BB - optimize the way when signing is disabled. We can drop this
> > > > > - * extra memory-to-memory copying and use iovec buffers for
> > > > constructing
> > > > > - * write request.
> > > > > + * iov_iter_get_pages_alloc doesn't work with ITER_KVEC.
> > > > > + * In this case, fall back to non-direct write function.
> > > > > + * this could be improved by getting pages directly in
> > > > > + ITER_KVEC
> > > > > */
> > > > > + if (direct && from->type & ITER_KVEC) {
> > > > > + cifs_dbg(FYI, "use non-direct cifs_writev for kvec I/O\n");
> > > > > + direct = false;
> > > > > + }
> > > > >
> > > > > rc = generic_write_checks(iocb, from);
> > > > > if (rc <= 0)
> > > > > @@ -2742,10 +2838,16 @@ ssize_t cifs_user_writev(struct kiocb
> > > > > *iocb, struct iov_iter *from)
> > > > >
> > > > > ctx->pos = iocb->ki_pos;
> > > > >
> > > > > - rc = setup_aio_ctx_iter(ctx, from, WRITE);
> > > > > - if (rc) {
> > > > > - kref_put(&ctx->refcount, cifs_aio_ctx_release);
> > > > > - return rc;
> > > > > + if (direct) {
> > > > > + ctx->direct_io = true;
> > > > > + ctx->iter = *from;
> > > > > + ctx->len = len;
> > > > > + } else {
> > > > > + rc = setup_aio_ctx_iter(ctx, from, WRITE);
> > > > > + if (rc) {
> > > > > + kref_put(&ctx->refcount, cifs_aio_ctx_release);
> > > > > + return rc;
> > > > > + }
> > > > > }
> > > > >
> > > > > /* grab a lock here due to read response handlers can
> > > > > access ctx */ @@ -2795,6 +2897,16 @@ ssize_t
> > > > > cifs_user_writev(struct kiocb
> > > > *iocb, struct iov_iter *from)
> > > > > return total_written;
> > > > > }
> > > > >
> > > > > +ssize_t cifs_direct_writev(struct kiocb *iocb, struct iov_iter
> > > > > +*from) {
> > > > > + return __cifs_writev(iocb, from, true); }
> > > > > +
> > > > > +ssize_t cifs_user_writev(struct kiocb *iocb, struct iov_iter *from) {
> > > > > + return __cifs_writev(iocb, from, false); }
> > > > > +
> > > > > static ssize_t
> > > > > cifs_writev(struct kiocb *iocb, struct iov_iter *from) {
> > > > > --
> > > > > 2.7.4
> > > > >
> > > >
> > > > Did you measure the performance benefit of using O_DIRECT with your
> > > > patches for non-RDMA mode?
> > >
> > > Here are some of the results measured on TCP/IP over IPoIB 40G Infiniband.
> > Please note that IPoIB is done in software so it's slower than a 40G Ethernet
> > NIC.
> > >
> > > All servers are running 2 X Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz.
> > Tested on FIO 64k read or write with queue depths 1 or 16. All the tests are
> > running with 1 FIO job, and with "direct=1".
> > >
> > > Without direct I/O:
> > > read 64k d1 113669KB/s
> > > read 64k d16 618428KB/s
> > > write 64k d1 134911KB/s
> > > write 64k d16 457005KB/s
> > >
> > > With direct I/O:
> > > read 64k d1 127134KB/s
> > > read 64k d16 714629KB/s
> > > write 64k d1 129268KB/s
> > > write 64k d16 461054KB/s
> > >
> > > Direct I/O improvement:
> > > read 64k d1 11%
> > > read 64k d16 15%
> > > write 64k d1 -5%
> > ^^^
> > This is strange. Is it consistent across multiple runs?
> >
> > > write 64k d16 1%
> >
> > Good read performance improvements overall and it seems write
> > performance results need more investigations.
>
> My apologies, I found out that I was doing it wrong for comparing write tests.
> They were actually all direct I/O, 5% difference is due to benchmark variation at
> the time of tests.
>
> Here are the re-run (3 times, each time for 90 seconds) for 64k write. This time,
> Linux client, Windows server and 40G IB switch are rebooted prior to tests.
>
> direct I/O
>
> 64k write d16
> 470m/s
> 616m/s
> 479m/s
>
> 64k write d1
> 176m/s
> 160m/s
> 156m/s
>
>
> without direct I/O
>
> 64k write d16
> 534m/s
> 405m/s
> 406m/s
>
> 64k write d1
> 146m/s
> 147m/s
> 147m/s
>
> I'm not sure why test results at depth 16 are jumpy, but overall there are
> improvements in direct I/O for 64k writes.
Better! Regarding the "jumpy" results, it's likely from queuing effects creeping in.
You might want to try doing a sweep of write queue depth, say 1, 4, 8, 16, and more,
then comparing the results across them all, and ignoring the jumpy ones - assuming
the jumpiness appears at about the same point. The idea being to compare the
improvement at the maximum stable point. Getting onto a decent 10GbE NIC would
also help. Still, it's clear there's a win here.
I'll chat with you offline about testing with John Hubbard's get_user_pages() RFC.
Would be very interesting to compare using that, too.
Tom.
More information about the samba-technical
mailing list