releases/5.0.18/libnvdimm-namespace-fix-label-tracking-error.patch

   1 From c4703ce11c23423d4b46e3d59aef7979814fd608 Mon Sep 17 00:00:00 2001
   2 From: Dan Williams <dan.j.williams@intel.com>
   3 Date: Tue, 30 Apr 2019 21:51:21 -0700
   4 Subject: libnvdimm/namespace: Fix label tracking error
   5
   6 From: Dan Williams <dan.j.williams@intel.com>
   7
   8 commit c4703ce11c23423d4b46e3d59aef7979814fd608 upstream.
   9
  10 Users have reported intermittent occurrences of DIMM initialization
  11 failures due to duplicate allocations of address capacity detected in
  12 the labels, or errors of the form below, both have the same root cause.
  13
  14     nd namespace1.4: failed to track label: 0
  15     WARNING: CPU: 17 PID: 1381 at drivers/nvdimm/label.c:863
  16
  17     RIP: 0010:__pmem_label_update+0x56c/0x590 [libnvdimm]
  18     Call Trace:
  19      ? nd_pmem_namespace_label_update+0xd6/0x160 [libnvdimm]
  20      nd_pmem_namespace_label_update+0xd6/0x160 [libnvdimm]
  21      uuid_store+0x17e/0x190 [libnvdimm]
  22      kernfs_fop_write+0xf0/0x1a0
  23      vfs_write+0xb7/0x1b0
  24      ksys_write+0x57/0xd0
  25      do_syscall_64+0x60/0x210
  26
  27 Unfortunately those reports were typically with a busy parallel
  28 namespace creation / destruction loop making it difficult to see the
  29 components of the bug. However, Jane provided a simple reproducer using
  30 the work-in-progress sub-section implementation.
  31
  32 When ndctl is reconfiguring a namespace it may take an existing defunct
  33 / disabled namespace and reconfigure it with a new uuid and other
  34 parameters. Critically namespace_update_uuid() takes existing address
  35 resources and renames them for the new namespace to use / reconfigure as
  36 it sees fit. The bug is that this rename only happens in the resource
  37 tracking tree. Existing labels with the old uuid are not reaped leading
  38 to a scenario where multiple active labels reference the same span of
  39 address range.
  40
  41 Teach namespace_update_uuid() to flag any references to the old uuid for
  42 reaping at the next label update attempt.
  43
  44 Cc: <stable@vger.kernel.org>
  45 Fixes: bf9bccc14c05 ("libnvdimm: pmem label sets and namespace instantiation")
  46 Link: https://github.com/pmem/ndctl/issues/91
  47 Reported-by: Jane Chu <jane.chu@oracle.com>
  48 Reported-by: Jeff Moyer <jmoyer@redhat.com>
  49 Reported-by: Erwin Tsaur <erwin.tsaur@oracle.com>
  50 Cc: Johannes Thumshirn <jthumshirn@suse.de>
  51 Signed-off-by: Dan Williams <dan.j.williams@intel.com>
  52 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  53
  54 ---
  55  drivers/nvdimm/label.c          |   29 ++++++++++++++++-------------
  56  drivers/nvdimm/namespace_devs.c |   15 +++++++++++++++
  57  drivers/nvdimm/nd.h             |    4 ++++
  58  3 files changed, 35 insertions(+), 13 deletions(-)
  59
  60 --- a/drivers/nvdimm/label.c
  61 +++ b/drivers/nvdimm/label.c
  62 @@ -753,6 +753,17 @@ static const guid_t *to_abstraction_guid
  63                 return &guid_null;
  64  }
  65
  66 +static void reap_victim(struct nd_mapping *nd_mapping,
  67 +               struct nd_label_ent *victim)
  68 +{
  69 +       struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
  70 +       u32 slot = to_slot(ndd, victim->label);
  71 +
  72 +       dev_dbg(ndd->dev, "free: %d\n", slot);
  73 +       nd_label_free_slot(ndd, slot);
  74 +       victim->label = NULL;
  75 +}
  76 +
  77  static int __pmem_label_update(struct nd_region *nd_region,
  78                 struct nd_mapping *nd_mapping, struct nd_namespace_pmem *nspm,
  79                 int pos, unsigned long flags)
  80 @@ -760,9 +771,9 @@ static int __pmem_label_update(struct nd
  81         struct nd_namespace_common *ndns = &nspm->nsio.common;
  82         struct nd_interleave_set *nd_set = nd_region->nd_set;
  83         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
  84 -       struct nd_label_ent *label_ent, *victim = NULL;
  85         struct nd_namespace_label *nd_label;
  86         struct nd_namespace_index *nsindex;
  87 +       struct nd_label_ent *label_ent;
  88         struct nd_label_id label_id;
  89         struct resource *res;
  90         unsigned long *free;
  91 @@ -831,18 +842,10 @@ static int __pmem_label_update(struct nd
  92         list_for_each_entry(label_ent, &nd_mapping->labels, list) {
  93                 if (!label_ent->label)
  94                         continue;
  95 -               if (memcmp(nspm->uuid, label_ent->label->uuid,
  96 -                                       NSLABEL_UUID_LEN) != 0)
  97 -                       continue;
  98 -               victim = label_ent;
  99 -               list_move_tail(&victim->list, &nd_mapping->labels);
 100 -               break;
 101 -       }
 102 -       if (victim) {
 103 -               dev_dbg(ndd->dev, "free: %d\n", slot);
 104 -               slot = to_slot(ndd, victim->label);
 105 -               nd_label_free_slot(ndd, slot);
 106 -               victim->label = NULL;
 107 +               if (test_and_clear_bit(ND_LABEL_REAP, &label_ent->flags)
 108 +                               || memcmp(nspm->uuid, label_ent->label->uuid,
 109 +                                       NSLABEL_UUID_LEN) == 0)
 110 +                       reap_victim(nd_mapping, label_ent);
 111         }
 112
 113         /* update index */
 114 --- a/drivers/nvdimm/namespace_devs.c
 115 +++ b/drivers/nvdimm/namespace_devs.c
 116 @@ -1247,12 +1247,27 @@ static int namespace_update_uuid(struct
 117         for (i = 0; i < nd_region->ndr_mappings; i++) {
 118                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 119                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 120 +               struct nd_label_ent *label_ent;
 121                 struct resource *res;
 122
 123                 for_each_dpa_resource(ndd, res)
 124                         if (strcmp(res->name, old_label_id.id) == 0)
 125                                 sprintf((void *) res->name, "%s",
 126                                                 new_label_id.id);
 127 +
 128 +               mutex_lock(&nd_mapping->lock);
 129 +               list_for_each_entry(label_ent, &nd_mapping->labels, list) {
 130 +                       struct nd_namespace_label *nd_label = label_ent->label;
 131 +                       struct nd_label_id label_id;
 132 +
 133 +                       if (!nd_label)
 134 +                               continue;
 135 +                       nd_label_gen_id(&label_id, nd_label->uuid,
 136 +                                       __le32_to_cpu(nd_label->flags));
 137 +                       if (strcmp(old_label_id.id, label_id.id) == 0)
 138 +                               set_bit(ND_LABEL_REAP, &label_ent->flags);
 139 +               }
 140 +               mutex_unlock(&nd_mapping->lock);
 141         }
 142         kfree(*old_uuid);
 143   out:
 144 --- a/drivers/nvdimm/nd.h
 145 +++ b/drivers/nvdimm/nd.h
 146 @@ -113,8 +113,12 @@ struct nd_percpu_lane {
 147         spinlock_t lock;
 148  };
 149
 150 +enum nd_label_flags {
 151 +       ND_LABEL_REAP,
 152 +};
 153  struct nd_label_ent {
 154         struct list_head list;
 155 +       unsigned long flags;
 156         struct nd_namespace_label *label;
 157  };
 158