kernel/bpf/task_iter.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright (c) 2020 Facebook */
   3
   4 #include <linux/init.h>
   5 #include <linux/namei.h>
   6 #include <linux/pid_namespace.h>
   7 #include <linux/fs.h>
   8 #include <linux/fdtable.h>
   9 #include <linux/filter.h>
  10 #include <linux/btf_ids.h>
  11
  12 struct bpf_iter_seq_task_common {
  13         struct pid_namespace *ns;
  14 };
  15
  16 struct bpf_iter_seq_task_info {
  17         /* The first field must be struct bpf_iter_seq_task_common.
  18          * this is assumed by {init, fini}_seq_pidns() callback functions.
  19          */
  20         struct bpf_iter_seq_task_common common;
  21         u32 tid;
  22 };
  23
  24 static struct task_struct *task_seq_get_next(struct pid_namespace *ns,
  25                                              u32 *tid,
  26                                              bool skip_if_dup_files)
  27 {
  28         struct task_struct *task = NULL;
  29         struct pid *pid;
  30
  31         rcu_read_lock();
  32 retry:
  33         pid = find_ge_pid(*tid, ns);
  34         if (pid) {
  35                 *tid = pid_nr_ns(pid, ns);
  36                 task = get_pid_task(pid, PIDTYPE_PID);
  37                 if (!task) {
  38                         ++*tid;
  39                         goto retry;
  40                 } else if (skip_if_dup_files && !thread_group_leader(task) &&
  41                            task->files == task->group_leader->files) {
  42                         put_task_struct(task);
  43                         task = NULL;
  44                         ++*tid;
  45                         goto retry;
  46                 }
  47         }
  48         rcu_read_unlock();
  49
  50         return task;
  51 }
  52
  53 static void *task_seq_start(struct seq_file *seq, loff_t *pos)
  54 {
  55         struct bpf_iter_seq_task_info *info = seq->private;
  56         struct task_struct *task;
  57
  58         task = task_seq_get_next(info->common.ns, &info->tid, false);
  59         if (!task)
  60                 return NULL;
  61
  62         if (*pos == 0)
  63                 ++*pos;
  64         return task;
  65 }
  66
  67 static void *task_seq_next(struct seq_file *seq, void *v, loff_t *pos)
  68 {
  69         struct bpf_iter_seq_task_info *info = seq->private;
  70         struct task_struct *task;
  71
  72         ++*pos;
  73         ++info->tid;
  74         put_task_struct((struct task_struct *)v);
  75         task = task_seq_get_next(info->common.ns, &info->tid, false);
  76         if (!task)
  77                 return NULL;
  78
  79         return task;
  80 }
  81
  82 struct bpf_iter__task {
  83         __bpf_md_ptr(struct bpf_iter_meta *, meta);
  84         __bpf_md_ptr(struct task_struct *, task);
  85 };
  86
  87 DEFINE_BPF_ITER_FUNC(task, struct bpf_iter_meta *meta, struct task_struct *task)
  88
  89 static int __task_seq_show(struct seq_file *seq, struct task_struct *task,
  90                            bool in_stop)
  91 {
  92         struct bpf_iter_meta meta;
  93         struct bpf_iter__task ctx;
  94         struct bpf_prog *prog;
  95
  96         meta.seq = seq;
  97         prog = bpf_iter_get_info(&meta, in_stop);
  98         if (!prog)
  99                 return 0;
 100
 101         meta.seq = seq;
 102         ctx.meta = &meta;
 103         ctx.task = task;
 104         return bpf_iter_run_prog(prog, &ctx);
 105 }
 106
 107 static int task_seq_show(struct seq_file *seq, void *v)
 108 {
 109         return __task_seq_show(seq, v, false);
 110 }
 111
 112 static void task_seq_stop(struct seq_file *seq, void *v)
 113 {
 114         if (!v)
 115                 (void)__task_seq_show(seq, v, true);
 116         else
 117                 put_task_struct((struct task_struct *)v);
 118 }
 119
 120 static const struct seq_operations task_seq_ops = {
 121         .start  = task_seq_start,
 122         .next   = task_seq_next,
 123         .stop   = task_seq_stop,
 124         .show   = task_seq_show,
 125 };
 126
 127 struct bpf_iter_seq_task_file_info {
 128         /* The first field must be struct bpf_iter_seq_task_common.
 129          * this is assumed by {init, fini}_seq_pidns() callback functions.
 130          */
 131         struct bpf_iter_seq_task_common common;
 132         struct task_struct *task;
 133         u32 tid;
 134         u32 fd;
 135 };
 136
 137 static struct file *
 138 task_file_seq_get_next(struct bpf_iter_seq_task_file_info *info)
 139 {
 140         struct pid_namespace *ns = info->common.ns;
 141         u32 curr_tid = info->tid;
 142         struct task_struct *curr_task;
 143         unsigned int curr_fd = info->fd;
 144
 145         /* If this function returns a non-NULL file object,
 146          * it held a reference to the task/file.
 147          * Otherwise, it does not hold any reference.
 148          */
 149 again:
 150         if (info->task) {
 151                 curr_task = info->task;
 152                 curr_fd = info->fd;
 153         } else {
 154                 curr_task = task_seq_get_next(ns, &curr_tid, true);
 155                 if (!curr_task) {
 156                         info->task = NULL;
 157                         info->tid = curr_tid;
 158                         return NULL;
 159                 }
 160
 161                 /* set info->task and info->tid */
 162                 if (curr_tid == info->tid) {
 163                         curr_fd = info->fd;
 164                 } else {
 165                         info->tid = curr_tid;
 166                         curr_fd = 0;
 167                 }
 168         }
 169
 170         rcu_read_lock();
 171         for (;; curr_fd++) {
 172                 struct file *f;
 173                 f = task_lookup_next_fd_rcu(curr_task, &curr_fd);
 174                 if (!f)
 175                         break;
 176                 if (!get_file_rcu(f))
 177                         continue;
 178
 179                 /* set info->fd */
 180                 info->fd = curr_fd;
 181                 rcu_read_unlock();
 182                 return f;
 183         }
 184
 185         /* the current task is done, go to the next task */
 186         rcu_read_unlock();
 187         put_task_struct(curr_task);
 188         info->task = NULL;
 189         info->fd = 0;
 190         curr_tid = ++(info->tid);
 191         goto again;
 192 }
 193
 194 static void *task_file_seq_start(struct seq_file *seq, loff_t *pos)
 195 {
 196         struct bpf_iter_seq_task_file_info *info = seq->private;
 197         struct file *file;
 198
 199         info->task = NULL;
 200         file = task_file_seq_get_next(info);
 201         if (file && *pos == 0)
 202                 ++*pos;
 203
 204         return file;
 205 }
 206
 207 static void *task_file_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 208 {
 209         struct bpf_iter_seq_task_file_info *info = seq->private;
 210
 211         ++*pos;
 212         ++info->fd;
 213         fput((struct file *)v);
 214         return task_file_seq_get_next(info);
 215 }
 216
 217 struct bpf_iter__task_file {
 218         __bpf_md_ptr(struct bpf_iter_meta *, meta);
 219         __bpf_md_ptr(struct task_struct *, task);
 220         u32 fd __aligned(8);
 221         __bpf_md_ptr(struct file *, file);
 222 };
 223
 224 DEFINE_BPF_ITER_FUNC(task_file, struct bpf_iter_meta *meta,
 225                      struct task_struct *task, u32 fd,
 226                      struct file *file)
 227
 228 static int __task_file_seq_show(struct seq_file *seq, struct file *file,
 229                                 bool in_stop)
 230 {
 231         struct bpf_iter_seq_task_file_info *info = seq->private;
 232         struct bpf_iter__task_file ctx;
 233         struct bpf_iter_meta meta;
 234         struct bpf_prog *prog;
 235
 236         meta.seq = seq;
 237         prog = bpf_iter_get_info(&meta, in_stop);
 238         if (!prog)
 239                 return 0;
 240
 241         ctx.meta = &meta;
 242         ctx.task = info->task;
 243         ctx.fd = info->fd;
 244         ctx.file = file;
 245         return bpf_iter_run_prog(prog, &ctx);
 246 }
 247
 248 static int task_file_seq_show(struct seq_file *seq, void *v)
 249 {
 250         return __task_file_seq_show(seq, v, false);
 251 }
 252
 253 static void task_file_seq_stop(struct seq_file *seq, void *v)
 254 {
 255         struct bpf_iter_seq_task_file_info *info = seq->private;
 256
 257         if (!v) {
 258                 (void)__task_file_seq_show(seq, v, true);
 259         } else {
 260                 fput((struct file *)v);
 261                 put_task_struct(info->task);
 262                 info->task = NULL;
 263         }
 264 }
 265
 266 static int init_seq_pidns(void *priv_data, struct bpf_iter_aux_info *aux)
 267 {
 268         struct bpf_iter_seq_task_common *common = priv_data;
 269
 270         common->ns = get_pid_ns(task_active_pid_ns(current));
 271         return 0;
 272 }
 273
 274 static void fini_seq_pidns(void *priv_data)
 275 {
 276         struct bpf_iter_seq_task_common *common = priv_data;
 277
 278         put_pid_ns(common->ns);
 279 }
 280
 281 static const struct seq_operations task_file_seq_ops = {
 282         .start  = task_file_seq_start,
 283         .next   = task_file_seq_next,
 284         .stop   = task_file_seq_stop,
 285         .show   = task_file_seq_show,
 286 };
 287
 288 BTF_ID_LIST(btf_task_file_ids)
 289 BTF_ID(struct, task_struct)
 290 BTF_ID(struct, file)
 291
 292 static const struct bpf_iter_seq_info task_seq_info = {
 293         .seq_ops                = &task_seq_ops,
 294         .init_seq_private       = init_seq_pidns,
 295         .fini_seq_private       = fini_seq_pidns,
 296         .seq_priv_size          = sizeof(struct bpf_iter_seq_task_info),
 297 };
 298
 299 static struct bpf_iter_reg task_reg_info = {
 300         .target                 = "task",
 301         .feature                = BPF_ITER_RESCHED,
 302         .ctx_arg_info_size      = 1,
 303         .ctx_arg_info           = {
 304                 { offsetof(struct bpf_iter__task, task),
 305                   PTR_TO_BTF_ID_OR_NULL },
 306         },
 307         .seq_info               = &task_seq_info,
 308 };
 309
 310 static const struct bpf_iter_seq_info task_file_seq_info = {
 311         .seq_ops                = &task_file_seq_ops,
 312         .init_seq_private       = init_seq_pidns,
 313         .fini_seq_private       = fini_seq_pidns,
 314         .seq_priv_size          = sizeof(struct bpf_iter_seq_task_file_info),
 315 };
 316
 317 static struct bpf_iter_reg task_file_reg_info = {
 318         .target                 = "task_file",
 319         .feature                = BPF_ITER_RESCHED,
 320         .ctx_arg_info_size      = 2,
 321         .ctx_arg_info           = {
 322                 { offsetof(struct bpf_iter__task_file, task),
 323                   PTR_TO_BTF_ID_OR_NULL },
 324                 { offsetof(struct bpf_iter__task_file, file),
 325                   PTR_TO_BTF_ID_OR_NULL },
 326         },
 327         .seq_info               = &task_file_seq_info,
 328 };
 329
 330 static int __init task_iter_init(void)
 331 {
 332         int ret;
 333
 334         task_reg_info.ctx_arg_info[0].btf_id = btf_task_file_ids[0];
 335         ret = bpf_iter_reg_target(&task_reg_info);
 336         if (ret)
 337                 return ret;
 338
 339         task_file_reg_info.ctx_arg_info[0].btf_id = btf_task_file_ids[0];
 340         task_file_reg_info.ctx_arg_info[1].btf_id = btf_task_file_ids[1];
 341         return bpf_iter_reg_target(&task_file_reg_info);
 342 }
 343 late_initcall(task_iter_init);