Merge pull request iovisor#3691 from chenhengqi/add-mdflush

docs/kernel-versions.md

@@ -69,7 +69,8 @@ BPF attached to cgroups as device controller | 4.15 | [`ebc614f68736`](https://g
 bpf2bpf function calls | 4.16 | [`cc8b0b92a169`](https://github.com/torvalds/linux/commit/cc8b0b92a1699bc32f7fec71daa2bfc90de43a4d)
 BPF used for monitoring socket RX/TX data | 4.17 | [`4f738adba30a`](https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=4f738adba30a7cfc006f605707e7aee847ffefa0)
 BPF attached to raw tracepoints | 4.17 | [`c4f6699dfcb8`](https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=c4f6699dfcb8558d138fe838f741b2c10f416cf9)
-BPF attached to `bind()` system call | 4.17 | [`4fbac77d2d09`](https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=4fbac77d2d092b475dda9eea66da674369665427)
+BPF attached to `bind()` system call | 4.17 | [`4fbac77d2d09`](https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=4fbac77d2d092b475dda9eea66da674369665427) [`aac3fc320d94`](https://github.com/torvalds/linux/commit/aac3fc320d9404f2665a8b1249dc3170d5fa3caf)
+BPF attached to `connect()` system call | 4.17 | [`d74bad4e74ee`](https://github.com/torvalds/linux/commit/d74bad4e74ee373787a9ae24197c17b7cdc428d5)
 BPF Type Format (BTF) | 4.18 | [`69b693f0aefa`](https://git.kernel.org/cgit/linux/kernel/git/davem/net-next.git/commit/?id=69b693f0aefa0ed521e8bd02260523b5ae446ad7)
 AF_XDP | 4.18 | [`fbfc504a24f5`](https://git.kernel.org/cgit/linux/kernel/git/davem/net-next.git/commit/?id=fbfc504a24f53f7ebe128ab55cb5dba634f4ece8)
 bpfilter | 4.18 | [`d2ba09c17a06`](https://git.kernel.org/cgit/linux/kernel/git/davem/net-next.git/commit/?id=d2ba09c17a0647f899d6c20a11bab9e6d3382f07)

libbpf-tools/.gitignore

@@ -29,6 +29,7 @@
 /llcstat
 /nfsdist
 /nfsslower
+/mdflush
 /mountsnoop
 /numamove
 /offcputime

libbpf-tools/Makefile

@@ -42,6 +42,7 @@ APPS = \
  klockstat \
  ksnoop \
  llcstat \
+ mdflush \
  mountsnoop \
  numamove \
  offcputime \
@@ -79,6 +80,15 @@ COMMON_OBJ = \
  $(if $(ENABLE_MIN_CORE_BTFS),$(OUTPUT)/min_core_btf_tar.o) \
  #
 
+define allow-override
+ $(if $(or $(findstring environment,$(origin $(1))),\
+ $(findstring command line,$(origin $(1)))),,\
+ $(eval $(1) = $(2)))
+endef
+
+$(call allow-override,CC,$(CROSS_COMPILE)cc)
+$(call allow-override,LD,$(CROSS_COMPILE)ld)
+
 .PHONY: all
 all: $(APPS) $(APP_ALIASES)
 
@@ -91,6 +101,13 @@ msg = @printf ' %-8s %s%s\n' "$(1)" "$(notdir $(2))" "$(if $(3), $(3))";
 MAKEFLAGS += --no-print-directory
 endif
 
+ifneq ($(EXTRA_CFLAGS),)
+CFLAGS += $(EXTRA_CFLAGS)
+endif
+ifneq ($(EXTRA_LDFLAGS),)
+LDFLAGS += $(EXTRA_LDFLAGS)
+endif
+
 .PHONY: clean
 clean:
  $(call msg,CLEAN)
@@ -103,7 +120,7 @@ $(OUTPUT) $(OUTPUT)/libbpf:
 .PHONY: bpftool
 bpftool:
  $(Q)mkdir -p $(OUTPUT)/bpftool
- $(Q)$(MAKE) OUTPUT=$(OUTPUT)/bpftool/ -C $(BPFTOOL_SRC)
+ $(Q)$(MAKE) ARCH= CROSS_COMPILE= OUTPUT=$(OUTPUT)/bpftool/ -C $(BPFTOOL_SRC)
 
 $(APPS): %: $(OUTPUT)/%.o $(LIBBPF_OBJ) $(COMMON_OBJ) | $(OUTPUT)
  $(call msg,BINARY,$@)

libbpf-tools/biolatency.bpf.c

@@ -35,7 +35,6 @@ struct {
  __uint(max_entries, MAX_ENTRIES);
  __type(key, struct request *);
  __type(value, u64);
- __uint(map_flags, BPF_F_NO_PREALLOC);
 } start SEC(".maps");
 
 static struct hist initial_hist;
@@ -45,7 +44,6 @@ struct {
  __uint(max_entries, MAX_ENTRIES);
  __type(key, struct hist_key);
  __type(value, struct hist);
- __uint(map_flags, BPF_F_NO_PREALLOC);
 } hists SEC(".maps");
 
 static __always_inline

libbpf-tools/biopattern.bpf.c

@@ -14,7 +14,6 @@ struct {
  __uint(max_entries, 64);
  __type(key, u32);
  __type(value, struct counter);
- __uint(map_flags, BPF_F_NO_PREALLOC);
 } counters SEC(".maps");
 
 SEC("tracepoint/block/block_rq_complete")

libbpf-tools/biosnoop.bpf.c

@@ -36,7 +36,6 @@ struct {
  __uint(max_entries, MAX_ENTRIES);
  __type(key, struct request *);
  __type(value, struct piddata);
- __uint(map_flags, BPF_F_NO_PREALLOC);
 } infobyreq SEC(".maps");
 
 struct stage {

libbpf-tools/biosnoop.c

@@ -167,7 +167,7 @@ void handle_event(void *ctx, int cpu, void *data, __u32 data_sz)
  start_ts = e->ts;
  blk_fill_rwbs(rwbs, e->cmd_flags);
  partition = partitions__get_by_dev(partitions, e->dev);
- printf("%-11.6f %-14.14s %-6d %-7s %-4s %-10lld %-7d ",
+ printf("%-11.6f %-14.14s %-7d %-7s %-4s %-10lld %-7d ",
  (e->ts - start_ts) / 1000000000.0,
  e->comm, e->pid, partition ? partition->name : "Unknown", rwbs,
  e->sector, e->len);
@@ -230,6 +230,13 @@ int main(int argc, char **argv)
  obj->rodata->targ_queued = env.queued;
  obj->rodata->filter_cg = env.cg;
 
+ if (fentry_can_attach("blk_account_io_start", NULL))
+ bpf_program__set_attach_target(obj->progs.blk_account_io_start, 0,
+ "blk_account_io_start");
+ else
+ bpf_program__set_attach_target(obj->progs.blk_account_io_start, 0,
+ "__blk_account_io_start");
+
  err = biosnoop_bpf__load(obj);
  if (err) {
  fprintf(stderr, "failed to load BPF object: %d\n", err);
@@ -304,7 +311,7 @@ int main(int argc, char **argv)
  goto cleanup;
  }
 
- printf("%-11s %-14s %-6s %-7s %-4s %-10s %-7s ",
+ printf("%-11s %-14s %-7s %-7s %-4s %-10s %-7s ",
  "TIME(s)", "COMM", "PID", "DISK", "T", "SECTOR", "BYTES");
  if (env.queued)
  printf("%7s ", "QUE(ms)");

libbpf-tools/biostacks.bpf.c

@@ -28,15 +28,13 @@ struct {
  __uint(max_entries, MAX_ENTRIES);
  __type(key, struct request *);
  __type(value, struct internal_rqinfo);
- __uint(map_flags, BPF_F_NO_PREALLOC);
 } rqinfos SEC(".maps");
 
 struct {
  __uint(type, BPF_MAP_TYPE_HASH);
  __uint(max_entries, MAX_ENTRIES);
  __type(key, struct rqinfo);
  __type(value, struct hist);
- __uint(map_flags, BPF_F_NO_PREALLOC);
 } hists SEC(".maps");
 
 static struct hist zero;

libbpf-tools/biostacks.c

@@ -173,6 +173,18 @@ int main(int argc, char **argv)
 
  obj->rodata->targ_ms = env.milliseconds;
 
+ if (fentry_can_attach("blk_account_io_start", NULL)) {
+ bpf_program__set_attach_target(obj->progs.blk_account_io_start, 0,
+ "blk_account_io_start");
+ bpf_program__set_attach_target(obj->progs.blk_account_io_done, 0,
+ "blk_account_io_done");
+ } else {
+ bpf_program__set_attach_target(obj->progs.blk_account_io_start, 0,
+ "__blk_account_io_start");
+ bpf_program__set_attach_target(obj->progs.blk_account_io_done, 0,
+ "__blk_account_io_done");
+ }
+
  err = biostacks_bpf__load(obj);
  if (err) {
  fprintf(stderr, "failed to load BPF object: %d\n", err);

libbpf-tools/bitesize.bpf.c

@@ -22,7 +22,6 @@ struct {
  __uint(max_entries, 10240);
  __type(key, struct hist_key);
  __type(value, struct hist);
- __uint(map_flags, BPF_F_NO_PREALLOC);
 } hists SEC(".maps");
 
 static struct hist initial_hist;

libbpf-tools/core_fixes.bpf.h

@@ -15,15 +15,34 @@
  */
 struct task_struct___x {
  unsigned int __state;
-};
+} __attribute__((preserve_access_index));
 
-static __s64 get_task_state(void *task)
+/**
+ * commit 309dca309fc3 ("block: store a block_device pointer in struct bio")
+ * adds a new member bi_bdev which is a pointer to struct block_device
+ * see:
+ * https://github.com/torvalds/linux/commit/309dca309fc3
+ */
+struct bio___x {
+ struct block_device *bi_bdev;
+} __attribute__((preserve_access_index));
+
+static __always_inline __s64 get_task_state(void *task)
 {
  struct task_struct___x *t = task;
 
  if (bpf_core_field_exists(t->__state))
- return t->__state;
- return ((struct task_struct *)task)->state;
+ return BPF_CORE_READ(t, __state);
+ return BPF_CORE_READ((struct task_struct *)task, state);
+}
+
+static __always_inline struct gendisk *get_gendisk(void *bio)
+{
+ struct bio___x *b = bio;
+
+ if (bpf_core_field_exists(b->bi_bdev))
+ return BPF_CORE_READ(b, bi_bdev, bd_disk);
+ return BPF_CORE_READ((struct bio *)bio, bi_disk);
 }
 
 #endif /* __CORE_FIXES_BPF_H */

libbpf-tools/klockstat.bpf.c

@@ -14,6 +14,7 @@
 const volatile pid_t targ_tgid = 0;
 const volatile pid_t targ_pid = 0;
 void *const volatile targ_lock = NULL;
+const volatile int per_thread = 0;
 
 struct {
  __uint(type, BPF_MAP_TYPE_STACK_TRACE);
@@ -147,6 +148,10 @@ static void account(struct lockholder_info *li)
 {
  struct lock_stat *ls;
  u64 delta;
+ u32 key = li->stack_id;
+
+ if (per_thread)
+ key = li->task_id;
 
  /*
  * Multiple threads may have the same stack_id. Even though we are
@@ -155,15 +160,19 @@ static void account(struct lockholder_info *li)
  * by multiple readers at the same time. They will be accounted as
  * the same lock, which is what we want, but we need to use atomics to
  * avoid corruption, especially for the total_time variables.
+ * But it should be ok for per-thread since it's not racy anymore.
  */
- ls = bpf_map_lookup_elem(&stat_map, &li->stack_id);
+ ls = bpf_map_lookup_elem(&stat_map, &key);
  if (!ls) {
  struct lock_stat fresh = {0};
 
- bpf_map_update_elem(&stat_map, &li->stack_id, &fresh, BPF_ANY);
- ls = bpf_map_lookup_elem(&stat_map, &li->stack_id);
+ bpf_map_update_elem(&stat_map, &key, &fresh, BPF_ANY);
+ ls = bpf_map_lookup_elem(&stat_map, &key);
  if (!ls)
  return;
+
+ if (per_thread)
+ bpf_get_current_comm(ls->acq_max_comm, TASK_COMM_LEN);
  }
 
  delta = li->acq_at - li->try_at;
@@ -176,7 +185,8 @@ static void account(struct lockholder_info *li)
  * Potentially racy, if multiple threads think they are the max,
  * so you may get a clobbered write.
  */
- bpf_get_current_comm(ls->acq_max_comm, TASK_COMM_LEN);
+ if (!per_thread)
+ bpf_get_current_comm(ls->acq_max_comm, TASK_COMM_LEN);
  }
 
  delta = li->rel_at - li->acq_at;
@@ -185,7 +195,8 @@ static void account(struct lockholder_info *li)
  if (delta > READ_ONCE(ls->hld_max_time)) {
  WRITE_ONCE(ls->hld_max_time, delta);
  WRITE_ONCE(ls->hld_max_id, li->task_id);
- bpf_get_current_comm(ls->hld_max_comm, TASK_COMM_LEN);
+ if (!per_thread)
+ bpf_get_current_comm(ls->hld_max_comm, TASK_COMM_LEN);
  }
 }