[bug report] [JFFS2] Add erase block summary support (mount time improvement)

From: Dan Carpenter <dan.carpenter@oracle.com>
To: havasi@inf.u-szeged.hu
Cc: linux-mtd@lists.infradead.org
Subject: [bug report] [JFFS2] Add erase block summary support (mount time improvement)
Date: Thu, 7 Nov 2019 11:00:41 +0300	[thread overview]
Message-ID: <20191107080041.GA14320@mwanda> (raw)

Hi JFFS2 devs,

The patch e631ddba5887: "[JFFS2] Add erase block summary support
(mount time improvement)" from Sep 7, 2005, leads to the following
static checker warning:

fs/jffs2/wbuf.c:934 jffs2_flash_writev()
warn: inconsistent returns 'c->wbuf_sem'.
  Locked on  : 908
  Unlocked on: 916,934

fs/jffs2/wbuf.c
   859          for (invec = 0; invec < count; invec++) {
   860                  int vlen = invecs[invec].iov_len;
   861                  uint8_t *v = invecs[invec].iov_base;
   862  
   863                  wbuf_retlen = jffs2_fill_wbuf(c, v, vlen);
   864  
   865                  if (c->wbuf_len == c->wbuf_pagesize) {
   866                          ret = __jffs2_flush_wbuf(c, NOPAD);
   867                          if (ret)
   868                                  goto outerr;
   869                  }
   870                  vlen -= wbuf_retlen;
   871                  outvec_to += wbuf_retlen;
   872                  donelen += wbuf_retlen;
   873                  v += wbuf_retlen;
   874  
   875                  if (vlen >= c->wbuf_pagesize) {
   876                          ret = mtd_write(c->mtd, outvec_to, PAGE_DIV(vlen),
   877                                          &wbuf_retlen, v);
   878                          if (ret < 0 || wbuf_retlen != PAGE_DIV(vlen))
   879                                  goto outfile;
   880  
   881                          vlen -= wbuf_retlen;
   882                          outvec_to += wbuf_retlen;
   883                          c->wbuf_ofs = outvec_to;
   884                          donelen += wbuf_retlen;
   885                          v += wbuf_retlen;
   886                  }
   887  
   888                  wbuf_retlen = jffs2_fill_wbuf(c, v, vlen);
   889                  if (c->wbuf_len == c->wbuf_pagesize) {
   890                          ret = __jffs2_flush_wbuf(c, NOPAD);
   891                          if (ret)
   892                                  goto outerr;
   893                  }
   894  
   895                  outvec_to += wbuf_retlen;
   896                  donelen += wbuf_retlen;
   897          }
   898  
   899          /*
   900           * If there's a remainder in the wbuf and it's a non-GC write,
   901           * remember that the wbuf affects this ino
   902           */
   903          *retlen = donelen;
   904  
   905          if (jffs2_sum_active()) {
   906                  int res = jffs2_sum_add_kvec(c, invecs, count, (uint32_t) to);
   907                  if (res)
   908                          return res;
                                ^^^^^^^^^^^
This should probably be "ret" instead of "res" and then goto outfile or
got outerr?

   909          }
   910  
   911          if (c->wbuf_len && ino)
   912                  jffs2_wbuf_dirties_inode(c, ino);
   913  
   914          ret = 0;
   915          up_write(&c->wbuf_sem);
   916          return ret;
   917  
   918  outfile:
   919          /*
   920           * At this point we have no problem, c->wbuf is empty. However
   921           * refile nextblock to avoid writing again to same address.
   922           */
   923  
   924          spin_lock(&c->erase_completion_lock);
   925  
   926          jeb = &c->blocks[outvec_to / c->sector_size];
   927          jffs2_block_refile(c, jeb, REFILE_ANYWAY);
   928  
   929          spin_unlock(&c->erase_completion_lock);
   930  
   931  outerr:
   932          *retlen = 0;
   933          up_write(&c->wbuf_sem);
   934          return ret;
   935  }

regards,
dan carpenter

______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/