[Crash-utility] [PATCH v6 1/7] Implement install and remove operations for mod_symname_hash

[Tao Liu]

Hello lijiang,

Thanks for reviewing the patch!

On Wed, Oct 20, 2021 at 4:46 PM lijiang <lijiang at redhat.com> wrote:
>
> Hi, Tao
> Thank you for the update.
>
> On Sat, Oct 16, 2021 at 1:21 PM Tao Liu <ltao at redhat.com> wrote:
> >
> > This patch indroduces mod_symname_hash, and its install/remove operations.
> > Since symbol_search() has to return the lowest address symbol and
> > symbol_search_next() returns the next lowest symbol, thus the installation
> > should be sorted ascendingly.
> >
> > In mod_symname_hash_install_range scenario, spn are already arranged
> > ascendingly, so for mod_symname_hash_install:
> >
> > Install spn previous to sp:
> >
> > If sp is the start of bucket, and
> > 1) spn->value is smaller than sp->value.
> >
> > Install spn next to sp:
> >
> > 1) sp->name_hash_next is NULL or
> > 2) sp->name_hash_next->value larger than spn->value
> >
> > spn->value is the kernel address of the symbol and will not change.
> > So we use it mainly to determine the sequence. When spn->value equals
> > sp->value, they must be symbols within a kernel module.
> >
> > Signed-off-by: Tao Liu <ltao at redhat.com>
> > ---
> >  defs.h    |  1 +
> >  symbols.c | 73 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
> >  2 files changed, 74 insertions(+)
> >
> > diff --git a/defs.h b/defs.h
> > index cbd45e5..bbdca79 100644
> > --- a/defs.h
> > +++ b/defs.h
> > @@ -2755,6 +2755,7 @@ struct symbol_table_data {
> >          double val_hash_searches;
> >          double val_hash_iterations;
> >          struct syment *symname_hash[SYMNAME_HASH];
> > +       struct syment *mod_symname_hash[SYMNAME_HASH];
> >         struct symbol_namespace kernel_namespace;
> >         struct syment *ext_module_symtable;
> >         struct syment *ext_module_symend;
> > diff --git a/symbols.c b/symbols.c
> > index 69dccdb..ad12d1c 100644
> > --- a/symbols.c
> > +++ b/symbols.c
> > @@ -1157,6 +1157,79 @@ symname_hash_install(struct syment *spn)
> >         }
> >  }
> >
> > +/*
> > + *  Install a single kernel module symbol into the mod_symname_hash.
> > + */
> > +static void
> > +mod_symname_hash_install(struct syment *spn)
> > +{
> > +       struct syment *sp;
> > +       int index;
> > +
> > +       if (!spn)
> > +               return;
> > +
> > +       index = SYMNAME_HASH_INDEX(spn->name);
> > +
> > +       sp = st->mod_symname_hash[index];
> > +
> > +       if (!sp || (spn->value < sp->value)) {
> > +               st->mod_symname_hash[index] = spn;
>                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> This could overwrite the existing syment if (sp && (spn->value <
> sp->value)), right?
>
I think it won't overwrite existing syment. if (sp && (spn->value <
> sp->value)), then what we want to do is inserting spn as the start of the bucket, and making sp to be the 2nd one. So st->mod_symname_hash[index] = spn, making spn to be the start of the bucket, then spn->name_hash_next = sp, making sp right after spn.

> > +               spn->name_hash_next = sp;
> > +               return;
> > +       }
> > +       for (; sp; sp = sp->name_hash_next) {
> > +               if (!sp->name_hash_next ||
> > +                   spn->value < sp->name_hash_next->value) {
> > +                       spn->name_hash_next = sp->name_hash_next;
> > +                       sp->name_hash_next = spn;
> > +                       return;
> > +               }
> > +       }
> > +}
> > +
> > +static void
> > +mod_symname_hash_remove(struct syment *spn)
> > +{
> > +       struct syment *sp;
> > +       int index;
> > +
> > +       if (!spn)
> > +               return;
> > +
> > +       index = SYMNAME_HASH_INDEX(spn->name);
> > +
> > +       if (st->mod_symname_hash[index] == spn) {
> > +               st->mod_symname_hash[index] = spn->name_hash_next;
> > +               return;
> > +       }
> > +
> > +       for (sp = st->mod_symname_hash[index]; sp; sp = sp->name_hash_next) {
> > +               if (sp->name_hash_next == spn) {
> > +                       sp->name_hash_next = spn->name_hash_next;
> > +                       return;
> > +               }
> > +       }
> > +}
>
> Can the above mod_symname_hash_remove() be simplified into the
> following implementation? The code may become more readable, and I
> didn't see any obvious performance issues as below.
>
> +static void
> +mod_symname_hash_remove(struct syment *spn)
> +{
> +       int index;
> +       struct syment *sp;
> +
> +       if (!spn)
> +               return;
> +
> +       index = SYMNAME_HASH_INDEX(spn->name);
> +       sp = st->mod_symname_hash[index];
> +
> +       while (sp) {
> +               if (sp == spn) {
> +                       sp = spn->name_hash_next;
> +                       spn->name_hash_next = NULL;
> +                       return;
> +               }
> +               sp = sp->name_hash_next;
> +       }
> +}
>

I don't think it can work. if (sp == spn), then sp should be removed
from the hash table. Since it is a singly linked list, the
name_hash_next field of the one which is prior to sp should be
updated. But in the code it is not.

Thanks,
Tao Liu
>
> > +
> > +static void
> > +mod_symtable_hash_install_range(struct syment *from, struct syment *to)
> > +{
> > +       struct syment *sp;
> > +
> > +       for (sp = from; sp <= to; sp++)
> > +               mod_symname_hash_install(sp);
> > +}
> > +
> > +static void
> > +mod_symtable_hash_remove_range(struct syment *from, struct syment *to)
> > +{
> > +       struct syment *sp;
> > +
> > +       for (sp = from; sp <= to; sp++)
> > +               mod_symname_hash_remove(sp);
> > +}
> > +
> >  /*
> >   *  Static kernel symbol value search
> >   */
> > --
> > 2.29.2
> >
>