minicoredumper: retry elf parsing as long as needed

Maximum number of tries, in rare cases, is insufficient for
elf parse. Backport patch that fixes the issue.

Signed-off-by: Sakib Sajal <sakib.sajal@windriver.com>
Signed-off-by: Khem Raj <raj.khem@gmail.com>

2 files changed, 129 insertions, 0 deletions

diff --git a/meta-oe/recipes-kernel/minicoredumper/files/0001-minicoredumper-retry-elf-parsing-as-long-as-needed.patch b/meta-oe/recipes-kernel/minicoredumper/files/0001-minicoredumper-retry-elf-parsing-as-long-as-needed.patch
new file mode 100644
index 000000000..8d5b8b6cb
--- /dev/null
+++ b/meta-oe/recipes-kernel/minicoredumper/files/0001-minicoredumper-retry-elf-parsing-as-long-as-needed.patch
@@ -0,0 +1,128 @@
+From 7a8c6a06c86e133e4346b1dc66483bd8d0d3c716 Mon Sep 17 00:00:00 2001
+From: John Ogness <john.ogness@linutronix.de>
+Date: Tue, 24 Aug 2021 21:10:43 +0200
+Subject: [PATCH] minicoredumper: retry elf parsing as long as needed
+
+As was reported in github issue #2 ("maximum number of tries
+insufficient, in rare cases, for elf parse"), the number of retries
+for parsing a process may be insufficient. Rather than setting an
+upper limit on the maximum number of retries, track the number of
+headers seen. As long as the number of seen headers is greater than
+the previous try, try again.
+
+In order to avoid introducing any new issues, preserve the behavior
+of retrying at least 10 times, even if no new headers are seen.
+
+Reported-by: github.com/ssajal-wr
+Signed-off-by: John Ogness <john.ogness@linutronix.de>
+
+Upstream-Status: Backport [7a8c6a06c86e133e4346b1dc66483bd8d0d3c716]
+
+Signed-off-by: Sakib Sajal <sakib.sajal@windriver.com>
+---
+ src/minicoredumper/corestripper.c | 30 +++++++++++++++++++++++-------
+ 1 file changed, 23 insertions(+), 7 deletions(-)
+
+diff --git a/src/minicoredumper/corestripper.c b/src/minicoredumper/corestripper.c
+index d96d1df..c96b350 100644
+--- a/src/minicoredumper/corestripper.c
++++ b/src/minicoredumper/corestripper.c
+@@ -761,7 +761,7 @@ static int init_log(struct dump_info *di)
+ typedef int elf_parse_cb(struct dump_info *di, Elf *elf, GElf_Phdr *phdr);
+ 
+ static int do_elf_ph_parse(struct dump_info *di, GElf_Phdr *type,
+-                          elf_parse_cb *callback)
++                          elf_parse_cb *callback, size_t *phnum_found)
+ {
+        GElf_Ehdr ehdr_mem;
+        GElf_Ehdr *ehdr;
+@@ -770,6 +770,9 @@ static int do_elf_ph_parse(struct dump_info *di, GElf_Phdr *type,
+        size_t phnum;
+        size_t cnt;
+ 
++       if (phnum_found)
++               *phnum_found = 0;
++
+        /* start from beginning of core */
+        if (lseek64(di->elf_fd, 0, SEEK_SET) == -1) {
+                info("lseek failed: %s", strerror(errno));
+@@ -809,6 +812,9 @@ static int do_elf_ph_parse(struct dump_info *di, GElf_Phdr *type,
+                goto out;
+        }
+ 
++       if (phnum_found)
++               *phnum_found = phnum;
++
+        for (cnt = 0; cnt < phnum; cnt++) {
+                GElf_Phdr phdr_mem;
+                GElf_Phdr *phdr;
+@@ -891,7 +897,7 @@ static int vma_cb(struct dump_info *di, Elf *elf, GElf_Phdr *phdr)
+ /*
+  * Tries to parse the found ELF headers and reads all vmas from it.
+  */
+-static int parse_vma_info(struct dump_info *di)
++static int parse_vma_info(struct dump_info *di, size_t *phnum_found)
+ {
+        unsigned long min_off = ULONG_MAX;
+        unsigned long max_len = 0;
+@@ -911,7 +917,7 @@ static int parse_vma_info(struct dump_info *di)
+        memset(&type, 0, sizeof(type));
+        type.p_type = PT_LOAD;
+        type.p_flags = PF_R;
+-       if (do_elf_ph_parse(di, &type, vma_cb) != 0)
++       if (do_elf_ph_parse(di, &type, vma_cb, phnum_found) != 0)
+                return -1;
+ 
+        for (v = di->vma; v; v = v->next) {
+@@ -1614,8 +1620,10 @@ int add_core_data(struct dump_info *di, off64_t dest_offset, size_t len,
+  */
+ static int init_src_core(struct dump_info *di, int src)
+ {
++       size_t last_phnum = 0;
+        int tries = 0;
+        int ret = -1;
++       size_t phnum;
+        size_t len;
+        char *buf;
+        long pos;
+@@ -1642,7 +1650,7 @@ again:
+                goto out;
+ 
+        /* try to elf-parse the core to read vma info */
+-       ret = parse_vma_info(di);
++       ret = parse_vma_info(di, &phnum);
+ 
+        /* restore our position */
+        if (lseek64(di->elf_fd, pos, SEEK_SET) == -1)
+@@ -1653,9 +1661,17 @@ again:
+ 
+                tries++;
+ 
+-               /* maybe try again */
+-               if (tries < 10)
++               if (phnum > last_phnum) {
++                       /* new headers found, keep trying */
++                       last_phnum = phnum;
+                        goto again;
++               } else if (tries < 10) {
++                       /*
++                        * even if no new headers are found,
++                        * retry at least 10 times
++                        */
++                       goto again;
++               }
+ 
+                goto out;
+        }
+@@ -2106,7 +2122,7 @@ static int dump_stacks(struct dump_info *di)
+                /* find and set the first task */
+                memset(&type, 0, sizeof(type));
+                type.p_type = PT_NOTE;
+-               do_elf_ph_parse(di, &type, note_cb);
++               do_elf_ph_parse(di, &type, note_cb, NULL);
+        }
+ 
+        if (di->first_pid)
+-- 
+2.25.1
+
diff --git a/meta-oe/recipes-kernel/minicoredumper/minicoredumper_2.0.1.bb b/meta-oe/recipes-kernel/minicoredumper/minicoredumper_2.0.1.bb
index bf9915294..0b934ee2d 100644
--- a/meta-oe/recipes-kernel/minicoredumper/minicoredumper_2.0.1.bb
+++ b/meta-oe/recipes-kernel/minicoredumper/minicoredumper_2.0.1.bb
@@ -17,6 +17,7 @@ SRC_URI = "git://github.com/diamon/minicoredumper;protocol=https;branch=master \
            file://0001-replace-pthread_mutexattr_setrobust_np-with-pthread_.patch \
            file://minicoredumper.service \
            file://minicoredumper.init \
+           file://0001-minicoredumper-retry-elf-parsing-as-long-as-needed.patch \
            "
 
 S = "${WORKDIR}/git"