From 8b04013737341442ed914b336cde866b902664ae Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Mon, 22 Jun 2020 15:04:15 -0400 Subject: pNFS/flexfiles: Fix list corruption if the mirror count changes If the mirror count changes in the new layout we pick up inside ff_layout_pg_init_write(), then we can end up adding the request to the wrong mirror and corrupting the mirror->pg_list. Fixes: d600ad1f2bdb ("NFS41: pop some layoutget errors to application") Cc: stable@vger.kernel.org Signed-off-by: Trond Myklebust Signed-off-by: Anna Schumaker --- fs/nfs/flexfilelayout/flexfilelayout.c | 11 +++++++---- 1 file changed, 7 insertions(+), 4 deletions(-) (limited to 'fs/nfs/flexfilelayout/flexfilelayout.c') diff --git a/fs/nfs/flexfilelayout/flexfilelayout.c b/fs/nfs/flexfilelayout/flexfilelayout.c index 7d399f72ebbb..de03e440b7ee 100644 --- a/fs/nfs/flexfilelayout/flexfilelayout.c +++ b/fs/nfs/flexfilelayout/flexfilelayout.c @@ -907,9 +907,8 @@ retry: goto out_mds; /* Use a direct mapping of ds_idx to pgio mirror_idx */ - if (WARN_ON_ONCE(pgio->pg_mirror_count != - FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg))) - goto out_mds; + if (pgio->pg_mirror_count != FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)) + goto out_eagain; for (i = 0; i < pgio->pg_mirror_count; i++) { mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i); @@ -931,7 +930,10 @@ retry: (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR)) pgio->pg_maxretrans = io_maxretrans; return; - +out_eagain: + pnfs_generic_pg_cleanup(pgio); + pgio->pg_error = -EAGAIN; + return; out_mds: trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode, 0, NFS4_MAX_UINT64, IOMODE_RW, @@ -941,6 +943,7 @@ out_mds: pgio->pg_lseg = NULL; pgio->pg_maxretrans = 0; nfs_pageio_reset_write_mds(pgio); + pgio->pg_error = -EAGAIN; } static unsigned int -- cgit From f97ff92bd16644c8d6f579bc6728974057de3ca5 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Thu, 18 Jun 2020 13:22:04 -0400 Subject: pNFS/flexfiles: Clean up redundant calls to pnfs_put_lseg() Both nfs_pageio_reset_read_mds() and nfs_pageio_reset_write_mds() do call pnfs_generic_pg_cleanup() for us. Signed-off-by: Trond Myklebust --- fs/nfs/flexfilelayout/flexfilelayout.c | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-) (limited to 'fs/nfs/flexfilelayout/flexfilelayout.c') diff --git a/fs/nfs/flexfilelayout/flexfilelayout.c b/fs/nfs/flexfilelayout/flexfilelayout.c index de03e440b7ee..b3ec12e5fde1 100644 --- a/fs/nfs/flexfilelayout/flexfilelayout.c +++ b/fs/nfs/flexfilelayout/flexfilelayout.c @@ -844,8 +844,7 @@ retry: if (!ds) { if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg)) goto out_mds; - pnfs_put_lseg(pgio->pg_lseg); - pgio->pg_lseg = NULL; + pnfs_generic_pg_cleanup(pgio); /* Sleep for 1 second before retrying */ ssleep(1); goto retry; @@ -871,8 +870,6 @@ out_mds: 0, NFS4_MAX_UINT64, IOMODE_READ, NFS_I(pgio->pg_inode)->layout, pgio->pg_lseg); - pnfs_put_lseg(pgio->pg_lseg); - pgio->pg_lseg = NULL; pgio->pg_maxretrans = 0; nfs_pageio_reset_read_mds(pgio); } @@ -916,8 +913,7 @@ retry: if (!ds) { if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg)) goto out_mds; - pnfs_put_lseg(pgio->pg_lseg); - pgio->pg_lseg = NULL; + pnfs_generic_pg_cleanup(pgio); /* Sleep for 1 second before retrying */ ssleep(1); goto retry; @@ -939,8 +935,6 @@ out_mds: 0, NFS4_MAX_UINT64, IOMODE_RW, NFS_I(pgio->pg_inode)->layout, pgio->pg_lseg); - pnfs_put_lseg(pgio->pg_lseg); - pgio->pg_lseg = NULL; pgio->pg_maxretrans = 0; nfs_pageio_reset_write_mds(pgio); pgio->pg_error = -EAGAIN; -- cgit From 18eb87f4443a0a23757e8a1ce3b3fa03714702fd Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Thu, 18 Jun 2020 13:30:39 -0400 Subject: pNFS/flexfiles: The mirror count could depend on the layout segment range Make sure we specify the layout segment range when calculating the mirror count. In theory, that number could depend on the range to which we're writing. Signed-off-by: Trond Myklebust --- fs/nfs/flexfilelayout/flexfilelayout.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'fs/nfs/flexfilelayout/flexfilelayout.c') diff --git a/fs/nfs/flexfilelayout/flexfilelayout.c b/fs/nfs/flexfilelayout/flexfilelayout.c index b3ec12e5fde1..b26173d72735 100644 --- a/fs/nfs/flexfilelayout/flexfilelayout.c +++ b/fs/nfs/flexfilelayout/flexfilelayout.c @@ -947,8 +947,8 @@ ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio, if (!pgio->pg_lseg) { pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req), - 0, - NFS4_MAX_UINT64, + req_offset(req), + req->wb_bytes, IOMODE_RW, false, GFP_NOFS); -- cgit From 563c53e73b8b6ec842828736f77e633f7b0911e9 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Tue, 11 Aug 2020 13:36:32 -0400 Subject: NFS: Fix flexfiles read failover The current mirrored read failover code is correctly resetting the mirror index between failed reads, however it is not able to actually flip the RPC call over to the next RPC client. The end result is that we keep resending the RPC call to the same client over and over. The fix is to use the pnfs_read_resend_pnfs() mechanism to schedule a new RPC call, but we need to add the ability to pass in a mirror index so that we always retry the next mirror in the list. Fixes: 166bd5b889ac ("pNFS/flexfiles: Fix layoutstats handling during read failovers") Signed-off-by: Trond Myklebust --- fs/nfs/flexfilelayout/flexfilelayout.c | 50 ++++++++++++++++++++++++---------- 1 file changed, 36 insertions(+), 14 deletions(-) (limited to 'fs/nfs/flexfilelayout/flexfilelayout.c') diff --git a/fs/nfs/flexfilelayout/flexfilelayout.c b/fs/nfs/flexfilelayout/flexfilelayout.c index b26173d72735..965145592750 100644 --- a/fs/nfs/flexfilelayout/flexfilelayout.c +++ b/fs/nfs/flexfilelayout/flexfilelayout.c @@ -790,6 +790,19 @@ ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg, return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx); } +static struct nfs4_pnfs_ds * +ff_layout_get_ds_for_read(struct nfs_pageio_descriptor *pgio, int *best_idx) +{ + struct pnfs_layout_segment *lseg = pgio->pg_lseg; + struct nfs4_pnfs_ds *ds; + + ds = ff_layout_choose_best_ds_for_read(lseg, pgio->pg_mirror_idx, + best_idx); + if (ds || !pgio->pg_mirror_idx) + return ds; + return ff_layout_choose_best_ds_for_read(lseg, 0, best_idx); +} + static void ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req, @@ -840,7 +853,7 @@ retry: goto out_nolseg; } - ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx); + ds = ff_layout_get_ds_for_read(pgio, &ds_idx); if (!ds) { if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg)) goto out_mds; @@ -1022,11 +1035,24 @@ static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs) } } +static void ff_layout_resend_pnfs_read(struct nfs_pgio_header *hdr) +{ + u32 idx = hdr->pgio_mirror_idx + 1; + int new_idx = 0; + + if (ff_layout_choose_any_ds_for_read(hdr->lseg, idx + 1, &new_idx)) + ff_layout_send_layouterror(hdr->lseg); + else + pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg); + pnfs_read_resend_pnfs(hdr, new_idx); +} + static void ff_layout_reset_read(struct nfs_pgio_header *hdr) { struct rpc_task *task = &hdr->task; pnfs_layoutcommit_inode(hdr->inode, false); + pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg); if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { dprintk("%s Reset task %5u for i/o through MDS " @@ -1228,6 +1254,12 @@ static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg, break; case NFS4ERR_NXIO: ff_layout_mark_ds_unreachable(lseg, idx); + /* + * Don't return the layout if this is a read and we still + * have layouts to try + */ + if (opnum == OP_READ) + break; /* Fallthrough */ default: pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, @@ -1241,7 +1273,6 @@ static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg, static int ff_layout_read_done_cb(struct rpc_task *task, struct nfs_pgio_header *hdr) { - int new_idx = hdr->pgio_mirror_idx; int err; if (task->tk_status < 0) { @@ -1261,10 +1292,6 @@ static int ff_layout_read_done_cb(struct rpc_task *task, clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags); switch (err) { case -NFS4ERR_RESET_TO_PNFS: - if (ff_layout_choose_best_ds_for_read(hdr->lseg, - hdr->pgio_mirror_idx + 1, - &new_idx)) - goto out_layouterror; set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags); return task->tk_status; case -NFS4ERR_RESET_TO_MDS: @@ -1275,10 +1302,6 @@ static int ff_layout_read_done_cb(struct rpc_task *task, } return 0; -out_layouterror: - ff_layout_read_record_layoutstats_done(task, hdr); - ff_layout_send_layouterror(hdr->lseg); - hdr->pgio_mirror_idx = new_idx; out_eagain: rpc_restart_call_prepare(task); return -EAGAIN; @@ -1405,10 +1428,9 @@ static void ff_layout_read_release(void *data) struct nfs_pgio_header *hdr = data; ff_layout_read_record_layoutstats_done(&hdr->task, hdr); - if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) { - ff_layout_send_layouterror(hdr->lseg); - pnfs_read_resend_pnfs(hdr); - } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags)) + if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) + ff_layout_resend_pnfs_read(hdr); + else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags)) ff_layout_reset_read(hdr); pnfs_generic_rw_release(data); } -- cgit