Merge branch 'master' into '6.0/stage'

.github/workflows/main.yml

@@ -94,10 +94,7 @@ jobs:
   #     comes with Ubuntu is 0.75 which has a couple of bugs triggered
   #     by our codebase (they've been resolved on trunk though so we
   #     may want to change this soon).
-  # [2] We supply --allow-untyped-calls because even though drgn has
-  #     stubs on typeshed now, there are still some untyped functions
-  #     (mainly in the helper API).
-  # [3] We supply --ignore-missing-imports to the tests package because
+  # [2] We supply --ignore-missing-imports to the tests package because
   #     pytest doesn't provide stubs on typeshed.
   #
   mypy:
@@ -109,7 +106,7 @@ jobs:
           python-version: '3.6'
       - run: ./.github/scripts/install-drgn.sh
       - run: python3 -m pip install mypy==0.730
-      - run: python3 -m mypy --strict --allow-untyped-calls --show-error-codes -p sdb
+      - run: python3 -m mypy --strict --show-error-codes -p sdb
       - run: python3 -m mypy --strict --ignore-missing-imports --show-error-codes -p tests
   #
   # Verify that "shfmt" ran successfully against our shell scripts.

sdb/command.py

@@ -643,13 +643,13 @@ def _call(self, objs: Iterable[drgn.Object]) -> Iterable[drgn.Object]:
         the types as we go.
         """
         assert self.input_type is not None
-        type_ = target.get_type(self.input_type)
+        expected_type = type_canonicalize_name(self.input_type)
         for obj in objs:
-            if obj.type_ != type_:
+            if type_canonical_name(obj.type_) != expected_type:
                 raise CommandError(
                     self.name,
                     'expected input of type {}, but received {}'.format(
-                        type_, obj.type_))
+                        expected_type, type_canonical_name(obj.type_)))
 
             yield from self.walk(obj)
 

sdb/commands/internal/util.py

@@ -24,13 +24,46 @@ def get_valid_type_by_name(cmd: sdb.Command, tname: str) -> drgn.Type:
     """
     Given a type name in string form (`tname`) without any C keyword
     prefixes (e.g. 'struct', 'enum', 'class', 'union'), return the
-    corresponding drgn.Type object.
+    corresponding drgn.Type object. If `tname` starts with a C keyword
+    we just return the type as is.
 
     This function is used primarily by commands that accept a type
     name as an argument and exist only to save keystrokes for the
     user.
     """
-    if tname in ['struct', 'enum', 'union', 'class']:
+    TYPE_KEYWORDS = ['struct', 'enum', 'union', 'class']
+
+    tokens = tname.split()
+    if len(tokens) > 2:
+        #
+        # drgn fails in all kinds of ways when we pass it an
+        # invalid type that consists of more than 2 text tokens.
+        #
+        raise sdb.CommandError(cmd.name,
+                               f"input '{tname}' is not a valid type name")
+
+    if len(tokens) == 2:
+        if tokens[0] not in TYPE_KEYWORDS or tokens[1] in TYPE_KEYWORDS:
+            #
+            # For the same reason mentioned in the above comment
+            # we also ensure that someone may not invalid two-token
+            # input that has the following errors:
+            # 1] Doesn't start with a type keyword - e.g "bogus type"
+            # 2] Has a type keyword as its type name (also see
+            #    comment below) - e.g. struct struct
+            #
+            raise sdb.CommandError(cmd.name,
+                                   f"input '{tname}' is not a valid type name")
+        try:
+            return sdb.get_type(tname)
+        except LookupError as err:
+            raise sdb.CommandError(cmd.name,
+                                   f"couldn't find type '{tname}'") from err
+        except SyntaxError as err:
+            raise sdb.CommandError(
+                cmd.name, f"input '{tname}' is not a valid type name") from err
+
+    if tname in TYPE_KEYWORDS:
         #
         # Note: We have to do this because currently in drgn
         # prog.type('struct') returns a different error than
@@ -82,9 +115,12 @@ def get_valid_type_by_name(cmd: sdb.Command, tname: str) -> drgn.Type:
         # it is a structure, an enum, or a union.
         #
         pass
-    for prefix in ["struct ", "enum ", "union "]:
+    except SyntaxError as err:
+        raise sdb.CommandError(
+            cmd.name, f"input '{tname}' is not a valid type name") from err
+    for prefix in TYPE_KEYWORDS:
         try:
-            return sdb.get_type(f"{prefix}{tname}")
+            return sdb.get_type(f"{prefix} {tname}")
         except LookupError:
             pass
     raise sdb.CommandError(

sdb/commands/ptype.py

@@ -50,16 +50,15 @@ class PType(sdb.Command):
 
         Print enums and unions:
 
-            sdb> ptype zfs_case v_t thread_union
+            sdb> ptype zfs_case 'struct v' thread_union
             enum zfs_case {
                     ZFS_CASE_SENSITIVE = 0,
                     ZFS_CASE_INSENSITIVE = 1,
                     ZFS_CASE_MIXED = 2,
             }
-            typedef union {
-                    iv_t e;
-                    uint8_t b[8];
-            } v_t
+            struct v {
+                    uint8_t b[16];
+            }
             union thread_union {
                     struct task_struct task;
                     unsigned long stack[2048];

sdb/commands/threads.py

@@ -16,16 +16,40 @@
 
 # pylint: disable=missing-docstring
 
+from textwrap import shorten
 from typing import Callable, Dict, Iterable, Union
 
 import drgn
 from drgn.helpers.linux.pid import for_each_task
+from drgn.helpers.linux.mm import cmdline
 
 import sdb
 from sdb.commands.internal.table import Table
 from sdb.commands.stacks import Stacks
 
 
+def _cmdline(obj: drgn.Object) -> str:
+    try:
+        s = " ".join(map(lambda s: s.decode("utf-8"), cmdline(obj)))
+
+        #
+        # The command line for a given thread can be obnoxiously long,
+        # so (by default) we limit it to 50 characters here. This helps
+        # preserve the readability of the command's output, but comes at
+        # the cost of not always showing the full command line of a
+        # thread.
+        #
+        return shorten(s, width=50)
+    except drgn.FaultError:
+        #
+        # The command line information is contained in the user address
+        # space of each thread, rather than in the kernel's address
+        # space. Thus, often, it may not be possible to retreive the
+        # thread's command line; e.g. when reading from a core dump.
+        #
+        return ""
+
+
 class Threads(sdb.Locator, sdb.PrettyPrinter):
     """
     Locate and print information about threads (task_stuct)
@@ -36,16 +60,16 @@ class Threads(sdb.Locator, sdb.PrettyPrinter):
         pid - the pid of the thread's process
         prio - the priority of the thread
         comm - the thread's command
+        cmdline - the thread's command line (when available)
 
     EXAMPLE
         sdb> threads | filter 'obj.comm == "java"' | threads
-        task               state         pid  prio comm
-        ------------------ ------------- ---- ---- ----
-        0xffff95d48b0e8000 INTERRUPTIBLE 4386 120  java
-        0xffff95d48b0e96c0 INTERRUPTIBLE 4388 120  java
-        0xffff95d48b0ead80 INTERRUPTIBLE 4387 120  java
-        0xffff95d48b0edb00 INTERRUPTIBLE 4304 120  java
-        0xffff95d4af20ad80 INTERRUPTIBLE 4395 120  java
+        task               state         pid  prio comm cmdline
+        ------------------ ------------- ---- ---- ---- ----------------------------------------
+        0xffff8c96a7c70000 INTERRUPTIBLE 3029 120  java /usr/bin/java -Ddelphix.debug=true [...]
+        0xffff8c96a7c71740 INTERRUPTIBLE 3028 120  java /usr/bin/java -Ddelphix.debug=true [...]
+        0xffff8c96a7c75d00 INTERRUPTIBLE 3024 120  java /usr/bin/java -Ddelphix.debug=true [...]
+        0xffff8c9715808000 INTERRUPTIBLE 3027 120  java /usr/bin/java -Ddelphix.debug=true [...]
     """
 
     names = ["threads", "thread"]
@@ -58,6 +82,7 @@ class Threads(sdb.Locator, sdb.PrettyPrinter):
         "pid": lambda obj: int(obj.pid),
         "prio": lambda obj: int(obj.prio),
         "comm": lambda obj: str(obj.comm.string_().decode("utf-8")),
+        "cmdline": _cmdline,
     }
 
     def pretty_print(self, objs: Iterable[drgn.Object]) -> None:

tests/integration/data/regression_output/core/ptype $abc

@@ -0,0 +1 @@
+sdb: ptype: input '$abc' is not a valid type name

tests/integration/data/regression_output/core/ptype 'a b c'

@@ -0,0 +1 @@
+sdb: ptype: input 'a b c' is not a valid type name

tests/integration/data/regression_output/core/ptype 'bogus union'

@@ -0,0 +1 @@
+sdb: ptype: input 'bogus union' is not a valid type name

tests/integration/data/regression_output/core/ptype 'struct bogus'

@@ -0,0 +1 @@
+sdb: ptype: couldn't find type 'struct bogus'

tests/integration/data/regression_output/core/ptype 'struct union'

@@ -0,0 +1 @@
+sdb: ptype: input 'struct union' is not a valid type name

tests/integration/data/regression_output/core/ptype 'struct'

@@ -0,0 +1 @@
+sdb: ptype: skip keyword 'struct' or quote your type "struct <typename>"

tests/integration/data/regression_output/core/ptype 'union struct struct'

@@ -0,0 +1 @@
+sdb: ptype: input 'union struct struct' is not a valid type name

tests/integration/data/regression_output/core/ptype 2abc

@@ -0,0 +1 @@
+sdb: ptype: input '2abc' is not a valid type name

tests/integration/data/regression_output/core/ptype @

@@ -0,0 +1 @@
+sdb: ptype: input '@' is not a valid type name

tests/integration/data/regression_output/core/ptype zfs_case v_t thread_union → tests/integration/data/regression_output/core/ptype zfs_case 'struct v' thread_union

@@ -3,10 +3,9 @@ enum zfs_case {
 	ZFS_CASE_INSENSITIVE = 1,
 	ZFS_CASE_MIXED = 2,
 }
-typedef union {
-	iv_t e;
-	uint8_t b[8];
-} v_t
+struct v {
+	uint8_t b[16];
+}
 union thread_union {
 	struct task_struct task;
 	unsigned long stack[2048];

tests/integration/data/regression_output/core/thread | filter "obj.comm == \"bogus\"" | thread

@@ -1,2 +1,2 @@
-task state pid prio comm
----- ----- --- ---- ----
+task state pid prio comm cmdline
+---- ----- --- ---- ---- -------

tests/integration/data/regression_output/linux/echo 0x0 | cpu_counter_sum

@@ -1 +1 @@
-sdb: cpu_counter_sum: invalid memory access: could not read memory from kdump: Cannot get page I/O address: PDPT table not present: pgd[0] = 0x0: 0x8
+sdb: cpu_counter_sum: invalid memory access: could not read memory from kdump: Cannot get page I/O address: PDPT table not present: p4d[0] = 0x0: 0x8