fs/tracefs/inode.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  inode.c - part of tracefs, a pseudo file system for activating tracing
   4  *
   5  * Based on debugfs by: Greg Kroah-Hartman <greg@kroah.com>
   6  *
   7  *  Copyright (C) 2014 Red Hat Inc, author: Steven Rostedt <srostedt@redhat.com>
   8  *
   9  * tracefs is the file system that is used by the tracing infrastructure.
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/fs.h>
  14 #include <linux/mount.h>
  15 #include <linux/kobject.h>
  16 #include <linux/namei.h>
  17 #include <linux/tracefs.h>
  18 #include <linux/fsnotify.h>
  19 #include <linux/security.h>
  20 #include <linux/seq_file.h>
  21 #include <linux/parser.h>
  22 #include <linux/magic.h>
  23 #include <linux/slab.h>
  24
  25 #define TRACEFS_DEFAULT_MODE    0700
  26
  27 static struct vfsmount *tracefs_mount;
  28 static int tracefs_mount_count;
  29 static bool tracefs_registered;
  30
  31 static ssize_t default_read_file(struct file *file, char __user *buf,
  32                                  size_t count, loff_t *ppos)
  33 {
  34         return 0;
  35 }
  36
  37 static ssize_t default_write_file(struct file *file, const char __user *buf,
  38                                    size_t count, loff_t *ppos)
  39 {
  40         return count;
  41 }
  42
  43 static const struct file_operations tracefs_file_operations = {
  44         .read =         default_read_file,
  45         .write =        default_write_file,
  46         .open =         simple_open,
  47         .llseek =       noop_llseek,
  48 };
  49
  50 static struct tracefs_dir_ops {
  51         int (*mkdir)(const char *name);
  52         int (*rmdir)(const char *name);
  53 } tracefs_ops __ro_after_init;
  54
  55 static char *get_dname(struct dentry *dentry)
  56 {
  57         const char *dname;
  58         char *name;
  59         int len = dentry->d_name.len;
  60
  61         dname = dentry->d_name.name;
  62         name = kmalloc(len + 1, GFP_KERNEL);
  63         if (!name)
  64                 return NULL;
  65         memcpy(name, dname, len);
  66         name[len] = 0;
  67         return name;
  68 }
  69
  70 static int tracefs_syscall_mkdir(struct user_namespace *mnt_userns,
  71                                  struct inode *inode, struct dentry *dentry,
  72                                  umode_t mode)
  73 {
  74         char *name;
  75         int ret;
  76
  77         name = get_dname(dentry);
  78         if (!name)
  79                 return -ENOMEM;
  80
  81         /*
  82          * The mkdir call can call the generic functions that create
  83          * the files within the tracefs system. It is up to the individual
  84          * mkdir routine to handle races.
  85          */
  86         inode_unlock(inode);
  87         ret = tracefs_ops.mkdir(name);
  88         inode_lock(inode);
  89
  90         kfree(name);
  91
  92         return ret;
  93 }
  94
  95 static int tracefs_syscall_rmdir(struct inode *inode, struct dentry *dentry)
  96 {
  97         char *name;
  98         int ret;
  99
 100         name = get_dname(dentry);
 101         if (!name)
 102                 return -ENOMEM;
 103
 104         /*
 105          * The rmdir call can call the generic functions that create
 106          * the files within the tracefs system. It is up to the individual
 107          * rmdir routine to handle races.
 108          * This time we need to unlock not only the parent (inode) but
 109          * also the directory that is being deleted.
 110          */
 111         inode_unlock(inode);
 112         inode_unlock(d_inode(dentry));
 113
 114         ret = tracefs_ops.rmdir(name);
 115
 116         inode_lock_nested(inode, I_MUTEX_PARENT);
 117         inode_lock(d_inode(dentry));
 118
 119         kfree(name);
 120
 121         return ret;
 122 }
 123
 124 static const struct inode_operations tracefs_dir_inode_operations = {
 125         .lookup         = simple_lookup,
 126         .mkdir          = tracefs_syscall_mkdir,
 127         .rmdir          = tracefs_syscall_rmdir,
 128 };
 129
 130 static struct inode *tracefs_get_inode(struct super_block *sb)
 131 {
 132         struct inode *inode = new_inode(sb);
 133         if (inode) {
 134                 inode->i_ino = get_next_ino();
 135                 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
 136         }
 137         return inode;
 138 }
 139
 140 struct tracefs_mount_opts {
 141         kuid_t uid;
 142         kgid_t gid;
 143         umode_t mode;
 144 };
 145
 146 enum {
 147         Opt_uid,
 148         Opt_gid,
 149         Opt_mode,
 150         Opt_err
 151 };
 152
 153 static const match_table_t tokens = {
 154         {Opt_uid, "uid=%u"},
 155         {Opt_gid, "gid=%u"},
 156         {Opt_mode, "mode=%o"},
 157         {Opt_err, NULL}
 158 };
 159
 160 struct tracefs_fs_info {
 161         struct tracefs_mount_opts mount_opts;
 162 };
 163
 164 static void change_gid(struct dentry *dentry, kgid_t gid)
 165 {
 166         if (!dentry->d_inode)
 167                 return;
 168         dentry->d_inode->i_gid = gid;
 169 }
 170
 171 /*
 172  * Taken from d_walk, but without he need for handling renames.
 173  * Nothing can be renamed while walking the list, as tracefs
 174  * does not support renames. This is only called when mounting
 175  * or remounting the file system, to set all the files to
 176  * the given gid.
 177  */
 178 static void set_gid(struct dentry *parent, kgid_t gid)
 179 {
 180         struct dentry *this_parent;
 181         struct list_head *next;
 182
 183         this_parent = parent;
 184         spin_lock(&this_parent->d_lock);
 185
 186         change_gid(this_parent, gid);
 187 repeat:
 188         next = this_parent->d_subdirs.next;
 189 resume:
 190         while (next != &this_parent->d_subdirs) {
 191                 struct list_head *tmp = next;
 192                 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
 193                 next = tmp->next;
 194
 195                 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
 196
 197                 change_gid(dentry, gid);
 198
 199                 if (!list_empty(&dentry->d_subdirs)) {
 200                         spin_unlock(&this_parent->d_lock);
 201                         spin_release(&dentry->d_lock.dep_map, _RET_IP_);
 202                         this_parent = dentry;
 203                         spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
 204                         goto repeat;
 205                 }
 206                 spin_unlock(&dentry->d_lock);
 207         }
 208         /*
 209          * All done at this level ... ascend and resume the search.
 210          */
 211         rcu_read_lock();
 212 ascend:
 213         if (this_parent != parent) {
 214                 struct dentry *child = this_parent;
 215                 this_parent = child->d_parent;
 216
 217                 spin_unlock(&child->d_lock);
 218                 spin_lock(&this_parent->d_lock);
 219
 220                 /* go into the first sibling still alive */
 221                 do {
 222                         next = child->d_child.next;
 223                         if (next == &this_parent->d_subdirs)
 224                                 goto ascend;
 225                         child = list_entry(next, struct dentry, d_child);
 226                 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
 227                 rcu_read_unlock();
 228                 goto resume;
 229         }
 230         rcu_read_unlock();
 231         spin_unlock(&this_parent->d_lock);
 232         return;
 233 }
 234
 235 static int tracefs_parse_options(char *data, struct tracefs_mount_opts *opts)
 236 {
 237         substring_t args[MAX_OPT_ARGS];
 238         int option;
 239         int token;
 240         kuid_t uid;
 241         kgid_t gid;
 242         char *p;
 243
 244         opts->mode = TRACEFS_DEFAULT_MODE;
 245
 246         while ((p = strsep(&data, ",")) != NULL) {
 247                 if (!*p)
 248                         continue;
 249
 250                 token = match_token(p, tokens, args);
 251                 switch (token) {
 252                 case Opt_uid:
 253                         if (match_int(&args[0], &option))
 254                                 return -EINVAL;
 255                         uid = make_kuid(current_user_ns(), option);
 256                         if (!uid_valid(uid))
 257                                 return -EINVAL;
 258                         opts->uid = uid;
 259                         break;
 260                 case Opt_gid:
 261                         if (match_int(&args[0], &option))
 262                                 return -EINVAL;
 263                         gid = make_kgid(current_user_ns(), option);
 264                         if (!gid_valid(gid))
 265                                 return -EINVAL;
 266                         opts->gid = gid;
 267                         break;
 268                 case Opt_mode:
 269                         if (match_octal(&args[0], &option))
 270                                 return -EINVAL;
 271                         opts->mode = option & S_IALLUGO;
 272                         break;
 273                 /*
 274                  * We might like to report bad mount options here;
 275                  * but traditionally tracefs has ignored all mount options
 276                  */
 277                 }
 278         }
 279
 280         return 0;
 281 }
 282
 283 static int tracefs_apply_options(struct super_block *sb)
 284 {
 285         struct tracefs_fs_info *fsi = sb->s_fs_info;
 286         struct inode *inode = d_inode(sb->s_root);
 287         struct tracefs_mount_opts *opts = &fsi->mount_opts;
 288
 289         inode->i_mode &= ~S_IALLUGO;
 290         inode->i_mode |= opts->mode;
 291
 292         inode->i_uid = opts->uid;
 293
 294         /* Set all the group ids to the mount option */
 295         set_gid(sb->s_root, opts->gid);
 296
 297         return 0;
 298 }
 299
 300 static int tracefs_remount(struct super_block *sb, int *flags, char *data)
 301 {
 302         int err;
 303         struct tracefs_fs_info *fsi = sb->s_fs_info;
 304
 305         sync_filesystem(sb);
 306         err = tracefs_parse_options(data, &fsi->mount_opts);
 307         if (err)
 308                 goto fail;
 309
 310         tracefs_apply_options(sb);
 311
 312 fail:
 313         return err;
 314 }
 315
 316 static int tracefs_show_options(struct seq_file *m, struct dentry *root)
 317 {
 318         struct tracefs_fs_info *fsi = root->d_sb->s_fs_info;
 319         struct tracefs_mount_opts *opts = &fsi->mount_opts;
 320
 321         if (!uid_eq(opts->uid, GLOBAL_ROOT_UID))
 322                 seq_printf(m, ",uid=%u",
 323                            from_kuid_munged(&init_user_ns, opts->uid));
 324         if (!gid_eq(opts->gid, GLOBAL_ROOT_GID))
 325                 seq_printf(m, ",gid=%u",
 326                            from_kgid_munged(&init_user_ns, opts->gid));
 327         if (opts->mode != TRACEFS_DEFAULT_MODE)
 328                 seq_printf(m, ",mode=%o", opts->mode);
 329
 330         return 0;
 331 }
 332
 333 static const struct super_operations tracefs_super_operations = {
 334         .statfs         = simple_statfs,
 335         .remount_fs     = tracefs_remount,
 336         .show_options   = tracefs_show_options,
 337 };
 338
 339 static int trace_fill_super(struct super_block *sb, void *data, int silent)
 340 {
 341         static const struct tree_descr trace_files[] = {{""}};
 342         struct tracefs_fs_info *fsi;
 343         int err;
 344
 345         fsi = kzalloc(sizeof(struct tracefs_fs_info), GFP_KERNEL);
 346         sb->s_fs_info = fsi;
 347         if (!fsi) {
 348                 err = -ENOMEM;
 349                 goto fail;
 350         }
 351
 352         err = tracefs_parse_options(data, &fsi->mount_opts);
 353         if (err)
 354                 goto fail;
 355
 356         err  =  simple_fill_super(sb, TRACEFS_MAGIC, trace_files);
 357         if (err)
 358                 goto fail;
 359
 360         sb->s_op = &tracefs_super_operations;
 361
 362         tracefs_apply_options(sb);
 363
 364         return 0;
 365
 366 fail:
 367         kfree(fsi);
 368         sb->s_fs_info = NULL;
 369         return err;
 370 }
 371
 372 static struct dentry *trace_mount(struct file_system_type *fs_type,
 373                         int flags, const char *dev_name,
 374                         void *data)
 375 {
 376         return mount_single(fs_type, flags, data, trace_fill_super);
 377 }
 378
 379 static struct file_system_type trace_fs_type = {
 380         .owner =        THIS_MODULE,
 381         .name =         "tracefs",
 382         .mount =        trace_mount,
 383         .kill_sb =      kill_litter_super,
 384 };
 385 MODULE_ALIAS_FS("tracefs");
 386
 387 static struct dentry *start_creating(const char *name, struct dentry *parent)
 388 {
 389         struct dentry *dentry;
 390         int error;
 391
 392         pr_debug("tracefs: creating file '%s'\n",name);
 393
 394         error = simple_pin_fs(&trace_fs_type, &tracefs_mount,
 395                               &tracefs_mount_count);
 396         if (error)
 397                 return ERR_PTR(error);
 398
 399         /* If the parent is not specified, we create it in the root.
 400          * We need the root dentry to do this, which is in the super
 401          * block. A pointer to that is in the struct vfsmount that we
 402          * have around.
 403          */
 404         if (!parent)
 405                 parent = tracefs_mount->mnt_root;
 406
 407         inode_lock(d_inode(parent));
 408         if (unlikely(IS_DEADDIR(d_inode(parent))))
 409                 dentry = ERR_PTR(-ENOENT);
 410         else
 411                 dentry = lookup_one_len(name, parent, strlen(name));
 412         if (!IS_ERR(dentry) && d_inode(dentry)) {
 413                 dput(dentry);
 414                 dentry = ERR_PTR(-EEXIST);
 415         }
 416
 417         if (IS_ERR(dentry)) {
 418                 inode_unlock(d_inode(parent));
 419                 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
 420         }
 421
 422         return dentry;
 423 }
 424
 425 static struct dentry *failed_creating(struct dentry *dentry)
 426 {
 427         inode_unlock(d_inode(dentry->d_parent));
 428         dput(dentry);
 429         simple_release_fs(&tracefs_mount, &tracefs_mount_count);
 430         return NULL;
 431 }
 432
 433 static struct dentry *end_creating(struct dentry *dentry)
 434 {
 435         inode_unlock(d_inode(dentry->d_parent));
 436         return dentry;
 437 }
 438
 439 /**
 440  * tracefs_create_file - create a file in the tracefs filesystem
 441  * @name: a pointer to a string containing the name of the file to create.
 442  * @mode: the permission that the file should have.
 443  * @parent: a pointer to the parent dentry for this file.  This should be a
 444  *          directory dentry if set.  If this parameter is NULL, then the
 445  *          file will be created in the root of the tracefs filesystem.
 446  * @data: a pointer to something that the caller will want to get to later
 447  *        on.  The inode.i_private pointer will point to this value on
 448  *        the open() call.
 449  * @fops: a pointer to a struct file_operations that should be used for
 450  *        this file.
 451  *
 452  * This is the basic "create a file" function for tracefs.  It allows for a
 453  * wide range of flexibility in creating a file, or a directory (if you want
 454  * to create a directory, the tracefs_create_dir() function is
 455  * recommended to be used instead.)
 456  *
 457  * This function will return a pointer to a dentry if it succeeds.  This
 458  * pointer must be passed to the tracefs_remove() function when the file is
 459  * to be removed (no automatic cleanup happens if your module is unloaded,
 460  * you are responsible here.)  If an error occurs, %NULL will be returned.
 461  *
 462  * If tracefs is not enabled in the kernel, the value -%ENODEV will be
 463  * returned.
 464  */
 465 struct dentry *tracefs_create_file(const char *name, umode_t mode,
 466                                    struct dentry *parent, void *data,
 467                                    const struct file_operations *fops)
 468 {
 469         struct dentry *dentry;
 470         struct inode *inode;
 471
 472         if (security_locked_down(LOCKDOWN_TRACEFS))
 473                 return NULL;
 474
 475         if (!(mode & S_IFMT))
 476                 mode |= S_IFREG;
 477         BUG_ON(!S_ISREG(mode));
 478         dentry = start_creating(name, parent);
 479
 480         if (IS_ERR(dentry))
 481                 return NULL;
 482
 483         inode = tracefs_get_inode(dentry->d_sb);
 484         if (unlikely(!inode))
 485                 return failed_creating(dentry);
 486
 487         inode->i_mode = mode;
 488         inode->i_fop = fops ? fops : &tracefs_file_operations;
 489         inode->i_private = data;
 490         inode->i_uid = d_inode(dentry->d_parent)->i_uid;
 491         inode->i_gid = d_inode(dentry->d_parent)->i_gid;
 492         d_instantiate(dentry, inode);
 493         fsnotify_create(d_inode(dentry->d_parent), dentry);
 494         return end_creating(dentry);
 495 }
 496
 497 static struct dentry *__create_dir(const char *name, struct dentry *parent,
 498                                    const struct inode_operations *ops)
 499 {
 500         struct dentry *dentry = start_creating(name, parent);
 501         struct inode *inode;
 502
 503         if (IS_ERR(dentry))
 504                 return NULL;
 505
 506         inode = tracefs_get_inode(dentry->d_sb);
 507         if (unlikely(!inode))
 508                 return failed_creating(dentry);
 509
 510         /* Do not set bits for OTH */
 511         inode->i_mode = S_IFDIR | S_IRWXU | S_IRUSR| S_IRGRP | S_IXUSR | S_IXGRP;
 512         inode->i_op = ops;
 513         inode->i_fop = &simple_dir_operations;
 514         inode->i_uid = d_inode(dentry->d_parent)->i_uid;
 515         inode->i_gid = d_inode(dentry->d_parent)->i_gid;
 516
 517         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 518         inc_nlink(inode);
 519         d_instantiate(dentry, inode);
 520         inc_nlink(d_inode(dentry->d_parent));
 521         fsnotify_mkdir(d_inode(dentry->d_parent), dentry);
 522         return end_creating(dentry);
 523 }
 524
 525 /**
 526  * tracefs_create_dir - create a directory in the tracefs filesystem
 527  * @name: a pointer to a string containing the name of the directory to
 528  *        create.
 529  * @parent: a pointer to the parent dentry for this file.  This should be a
 530  *          directory dentry if set.  If this parameter is NULL, then the
 531  *          directory will be created in the root of the tracefs filesystem.
 532  *
 533  * This function creates a directory in tracefs with the given name.
 534  *
 535  * This function will return a pointer to a dentry if it succeeds.  This
 536  * pointer must be passed to the tracefs_remove() function when the file is
 537  * to be removed. If an error occurs, %NULL will be returned.
 538  *
 539  * If tracing is not enabled in the kernel, the value -%ENODEV will be
 540  * returned.
 541  */
 542 struct dentry *tracefs_create_dir(const char *name, struct dentry *parent)
 543 {
 544         return __create_dir(name, parent, &simple_dir_inode_operations);
 545 }
 546
 547 /**
 548  * tracefs_create_instance_dir - create the tracing instances directory
 549  * @name: The name of the instances directory to create
 550  * @parent: The parent directory that the instances directory will exist
 551  * @mkdir: The function to call when a mkdir is performed.
 552  * @rmdir: The function to call when a rmdir is performed.
 553  *
 554  * Only one instances directory is allowed.
 555  *
 556  * The instances directory is special as it allows for mkdir and rmdir
 557  * to be done by userspace. When a mkdir or rmdir is performed, the inode
 558  * locks are released and the methods passed in (@mkdir and @rmdir) are
 559  * called without locks and with the name of the directory being created
 560  * within the instances directory.
 561  *
 562  * Returns the dentry of the instances directory.
 563  */
 564 __init struct dentry *tracefs_create_instance_dir(const char *name,
 565                                           struct dentry *parent,
 566                                           int (*mkdir)(const char *name),
 567                                           int (*rmdir)(const char *name))
 568 {
 569         struct dentry *dentry;
 570
 571         /* Only allow one instance of the instances directory. */
 572         if (WARN_ON(tracefs_ops.mkdir || tracefs_ops.rmdir))
 573                 return NULL;
 574
 575         dentry = __create_dir(name, parent, &tracefs_dir_inode_operations);
 576         if (!dentry)
 577                 return NULL;
 578
 579         tracefs_ops.mkdir = mkdir;
 580         tracefs_ops.rmdir = rmdir;
 581
 582         return dentry;
 583 }
 584
 585 static void remove_one(struct dentry *victim)
 586 {
 587         simple_release_fs(&tracefs_mount, &tracefs_mount_count);
 588 }
 589
 590 /**
 591  * tracefs_remove - recursively removes a directory
 592  * @dentry: a pointer to a the dentry of the directory to be removed.
 593  *
 594  * This function recursively removes a directory tree in tracefs that
 595  * was previously created with a call to another tracefs function
 596  * (like tracefs_create_file() or variants thereof.)
 597  */
 598 void tracefs_remove(struct dentry *dentry)
 599 {
 600         if (IS_ERR_OR_NULL(dentry))
 601                 return;
 602
 603         simple_pin_fs(&trace_fs_type, &tracefs_mount, &tracefs_mount_count);
 604         simple_recursive_removal(dentry, remove_one);
 605         simple_release_fs(&tracefs_mount, &tracefs_mount_count);
 606 }
 607
 608 /**
 609  * tracefs_initialized - Tells whether tracefs has been registered
 610  */
 611 bool tracefs_initialized(void)
 612 {
 613         return tracefs_registered;
 614 }
 615
 616 static int __init tracefs_init(void)
 617 {
 618         int retval;
 619
 620         retval = sysfs_create_mount_point(kernel_kobj, "tracing");
 621         if (retval)
 622                 return -EINVAL;
 623
 624         retval = register_filesystem(&trace_fs_type);
 625         if (!retval)
 626                 tracefs_registered = true;
 627
 628         return retval;
 629 }
 630 core_initcall(tracefs_init);