fs/tracefs/inode.c

   1 /*
   2  *  inode.c - part of tracefs, a pseudo file system for activating tracing
   3  *
   4  * Based on debugfs by: Greg Kroah-Hartman <greg@kroah.com>
   5  *
   6  *  Copyright (C) 2014 Red Hat Inc, author: Steven Rostedt <srostedt@redhat.com>
   7  *
   8  *      This program is free software; you can redistribute it and/or
   9  *      modify it under the terms of the GNU General Public License version
  10  *      2 as published by the Free Software Foundation.
  11  *
  12  * tracefs is the file system that is used by the tracing infrastructure.
  13  *
  14  */
  15
  16 #include <linux/module.h>
  17 #include <linux/fs.h>
  18 #include <linux/mount.h>
  19 #include <linux/kobject.h>
  20 #include <linux/namei.h>
  21 #include <linux/tracefs.h>
  22 #include <linux/fsnotify.h>
  23 #include <linux/seq_file.h>
  24 #include <linux/parser.h>
  25 #include <linux/magic.h>
  26 #include <linux/slab.h>
  27
  28 #define TRACEFS_DEFAULT_MODE    0700
  29
  30 static struct vfsmount *tracefs_mount;
  31 static int tracefs_mount_count;
  32 static bool tracefs_registered;
  33
  34 static ssize_t default_read_file(struct file *file, char __user *buf,
  35                                  size_t count, loff_t *ppos)
  36 {
  37         return 0;
  38 }
  39
  40 static ssize_t default_write_file(struct file *file, const char __user *buf,
  41                                    size_t count, loff_t *ppos)
  42 {
  43         return count;
  44 }
  45
  46 static const struct file_operations tracefs_file_operations = {
  47         .read =         default_read_file,
  48         .write =        default_write_file,
  49         .open =         simple_open,
  50         .llseek =       noop_llseek,
  51 };
  52
  53 static struct tracefs_dir_ops {
  54         int (*mkdir)(const char *name);
  55         int (*rmdir)(const char *name);
  56 } tracefs_ops;
  57
  58 static char *get_dname(struct dentry *dentry)
  59 {
  60         const char *dname;
  61         char *name;
  62         int len = dentry->d_name.len;
  63
  64         dname = dentry->d_name.name;
  65         name = kmalloc(len + 1, GFP_KERNEL);
  66         if (!name)
  67                 return NULL;
  68         memcpy(name, dname, len);
  69         name[len] = 0;
  70         return name;
  71 }
  72
  73 static int tracefs_syscall_mkdir(struct inode *inode, struct dentry *dentry, umode_t mode)
  74 {
  75         char *name;
  76         int ret;
  77
  78         name = get_dname(dentry);
  79         if (!name)
  80                 return -ENOMEM;
  81
  82         /*
  83          * The mkdir call can call the generic functions that create
  84          * the files within the tracefs system. It is up to the individual
  85          * mkdir routine to handle races.
  86          */
  87         inode_unlock(inode);
  88         ret = tracefs_ops.mkdir(name);
  89         inode_lock(inode);
  90
  91         kfree(name);
  92
  93         return ret;
  94 }
  95
  96 static int tracefs_syscall_rmdir(struct inode *inode, struct dentry *dentry)
  97 {
  98         char *name;
  99         int ret;
 100
 101         name = get_dname(dentry);
 102         if (!name)
 103                 return -ENOMEM;
 104
 105         /*
 106          * The rmdir call can call the generic functions that create
 107          * the files within the tracefs system. It is up to the individual
 108          * rmdir routine to handle races.
 109          * This time we need to unlock not only the parent (inode) but
 110          * also the directory that is being deleted.
 111          */
 112         inode_unlock(inode);
 113         inode_unlock(dentry->d_inode);
 114
 115         ret = tracefs_ops.rmdir(name);
 116
 117         inode_lock_nested(inode, I_MUTEX_PARENT);
 118         inode_lock(dentry->d_inode);
 119
 120         kfree(name);
 121
 122         return ret;
 123 }
 124
 125 static const struct inode_operations tracefs_dir_inode_operations = {
 126         .lookup         = simple_lookup,
 127         .mkdir          = tracefs_syscall_mkdir,
 128         .rmdir          = tracefs_syscall_rmdir,
 129 };
 130
 131 static struct inode *tracefs_get_inode(struct super_block *sb)
 132 {
 133         struct inode *inode = new_inode(sb);
 134         if (inode) {
 135                 inode->i_ino = get_next_ino();
 136                 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
 137         }
 138         return inode;
 139 }
 140
 141 struct tracefs_mount_opts {
 142         kuid_t uid;
 143         kgid_t gid;
 144         umode_t mode;
 145 };
 146
 147 enum {
 148         Opt_uid,
 149         Opt_gid,
 150         Opt_mode,
 151         Opt_err
 152 };
 153
 154 static const match_table_t tokens = {
 155         {Opt_uid, "uid=%u"},
 156         {Opt_gid, "gid=%u"},
 157         {Opt_mode, "mode=%o"},
 158         {Opt_err, NULL}
 159 };
 160
 161 struct tracefs_fs_info {
 162         struct tracefs_mount_opts mount_opts;
 163 };
 164
 165 static int tracefs_parse_options(char *data, struct tracefs_mount_opts *opts)
 166 {
 167         substring_t args[MAX_OPT_ARGS];
 168         int option;
 169         int token;
 170         kuid_t uid;
 171         kgid_t gid;
 172         char *p;
 173
 174         opts->mode = TRACEFS_DEFAULT_MODE;
 175
 176         while ((p = strsep(&data, ",")) != NULL) {
 177                 if (!*p)
 178                         continue;
 179
 180                 token = match_token(p, tokens, args);
 181                 switch (token) {
 182                 case Opt_uid:
 183                         if (match_int(&args[0], &option))
 184                                 return -EINVAL;
 185                         uid = make_kuid(current_user_ns(), option);
 186                         if (!uid_valid(uid))
 187                                 return -EINVAL;
 188                         opts->uid = uid;
 189                         break;
 190                 case Opt_gid:
 191                         if (match_int(&args[0], &option))
 192                                 return -EINVAL;
 193                         gid = make_kgid(current_user_ns(), option);
 194                         if (!gid_valid(gid))
 195                                 return -EINVAL;
 196                         opts->gid = gid;
 197                         break;
 198                 case Opt_mode:
 199                         if (match_octal(&args[0], &option))
 200                                 return -EINVAL;
 201                         opts->mode = option & S_IALLUGO;
 202                         break;
 203                 /*
 204                  * We might like to report bad mount options here;
 205                  * but traditionally tracefs has ignored all mount options
 206                  */
 207                 }
 208         }
 209
 210         return 0;
 211 }
 212
 213 static int tracefs_apply_options(struct super_block *sb)
 214 {
 215         struct tracefs_fs_info *fsi = sb->s_fs_info;
 216         struct inode *inode = sb->s_root->d_inode;
 217         struct tracefs_mount_opts *opts = &fsi->mount_opts;
 218
 219         inode->i_mode &= ~S_IALLUGO;
 220         inode->i_mode |= opts->mode;
 221
 222         inode->i_uid = opts->uid;
 223         inode->i_gid = opts->gid;
 224
 225         return 0;
 226 }
 227
 228 static int tracefs_remount(struct super_block *sb, int *flags, char *data)
 229 {
 230         int err;
 231         struct tracefs_fs_info *fsi = sb->s_fs_info;
 232
 233         sync_filesystem(sb);
 234         err = tracefs_parse_options(data, &fsi->mount_opts);
 235         if (err)
 236                 goto fail;
 237
 238         tracefs_apply_options(sb);
 239
 240 fail:
 241         return err;
 242 }
 243
 244 static int tracefs_show_options(struct seq_file *m, struct dentry *root)
 245 {
 246         struct tracefs_fs_info *fsi = root->d_sb->s_fs_info;
 247         struct tracefs_mount_opts *opts = &fsi->mount_opts;
 248
 249         if (!uid_eq(opts->uid, GLOBAL_ROOT_UID))
 250                 seq_printf(m, ",uid=%u",
 251                            from_kuid_munged(&init_user_ns, opts->uid));
 252         if (!gid_eq(opts->gid, GLOBAL_ROOT_GID))
 253                 seq_printf(m, ",gid=%u",
 254                            from_kgid_munged(&init_user_ns, opts->gid));
 255         if (opts->mode != TRACEFS_DEFAULT_MODE)
 256                 seq_printf(m, ",mode=%o", opts->mode);
 257
 258         return 0;
 259 }
 260
 261 static const struct super_operations tracefs_super_operations = {
 262         .statfs         = simple_statfs,
 263         .remount_fs     = tracefs_remount,
 264         .show_options   = tracefs_show_options,
 265 };
 266
 267 static int trace_fill_super(struct super_block *sb, void *data, int silent)
 268 {
 269         static const struct tree_descr trace_files[] = {{""}};
 270         struct tracefs_fs_info *fsi;
 271         int err;
 272
 273         fsi = kzalloc(sizeof(struct tracefs_fs_info), GFP_KERNEL);
 274         sb->s_fs_info = fsi;
 275         if (!fsi) {
 276                 err = -ENOMEM;
 277                 goto fail;
 278         }
 279
 280         err = tracefs_parse_options(data, &fsi->mount_opts);
 281         if (err)
 282                 goto fail;
 283
 284         err  =  simple_fill_super(sb, TRACEFS_MAGIC, trace_files);
 285         if (err)
 286                 goto fail;
 287
 288         sb->s_op = &tracefs_super_operations;
 289
 290         tracefs_apply_options(sb);
 291
 292         return 0;
 293
 294 fail:
 295         kfree(fsi);
 296         sb->s_fs_info = NULL;
 297         return err;
 298 }
 299
 300 static struct dentry *trace_mount(struct file_system_type *fs_type,
 301                         int flags, const char *dev_name,
 302                         void *data)
 303 {
 304         return mount_single(fs_type, flags, data, trace_fill_super);
 305 }
 306
 307 static struct file_system_type trace_fs_type = {
 308         .owner =        THIS_MODULE,
 309         .name =         "tracefs",
 310         .mount =        trace_mount,
 311         .kill_sb =      kill_litter_super,
 312 };
 313 MODULE_ALIAS_FS("tracefs");
 314
 315 static struct dentry *start_creating(const char *name, struct dentry *parent)
 316 {
 317         struct dentry *dentry;
 318         int error;
 319
 320         pr_debug("tracefs: creating file '%s'\n",name);
 321
 322         error = simple_pin_fs(&trace_fs_type, &tracefs_mount,
 323                               &tracefs_mount_count);
 324         if (error)
 325                 return ERR_PTR(error);
 326
 327         /* If the parent is not specified, we create it in the root.
 328          * We need the root dentry to do this, which is in the super
 329          * block. A pointer to that is in the struct vfsmount that we
 330          * have around.
 331          */
 332         if (!parent)
 333                 parent = tracefs_mount->mnt_root;
 334
 335         inode_lock(parent->d_inode);
 336         dentry = lookup_one_len(name, parent, strlen(name));
 337         if (!IS_ERR(dentry) && dentry->d_inode) {
 338                 dput(dentry);
 339                 dentry = ERR_PTR(-EEXIST);
 340         }
 341
 342         if (IS_ERR(dentry)) {
 343                 inode_unlock(parent->d_inode);
 344                 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
 345         }
 346
 347         return dentry;
 348 }
 349
 350 static struct dentry *failed_creating(struct dentry *dentry)
 351 {
 352         inode_unlock(dentry->d_parent->d_inode);
 353         dput(dentry);
 354         simple_release_fs(&tracefs_mount, &tracefs_mount_count);
 355         return NULL;
 356 }
 357
 358 static struct dentry *end_creating(struct dentry *dentry)
 359 {
 360         inode_unlock(dentry->d_parent->d_inode);
 361         return dentry;
 362 }
 363
 364 /**
 365  * tracefs_create_file - create a file in the tracefs filesystem
 366  * @name: a pointer to a string containing the name of the file to create.
 367  * @mode: the permission that the file should have.
 368  * @parent: a pointer to the parent dentry for this file.  This should be a
 369  *          directory dentry if set.  If this parameter is NULL, then the
 370  *          file will be created in the root of the tracefs filesystem.
 371  * @data: a pointer to something that the caller will want to get to later
 372  *        on.  The inode.i_private pointer will point to this value on
 373  *        the open() call.
 374  * @fops: a pointer to a struct file_operations that should be used for
 375  *        this file.
 376  *
 377  * This is the basic "create a file" function for tracefs.  It allows for a
 378  * wide range of flexibility in creating a file, or a directory (if you want
 379  * to create a directory, the tracefs_create_dir() function is
 380  * recommended to be used instead.)
 381  *
 382  * This function will return a pointer to a dentry if it succeeds.  This
 383  * pointer must be passed to the tracefs_remove() function when the file is
 384  * to be removed (no automatic cleanup happens if your module is unloaded,
 385  * you are responsible here.)  If an error occurs, %NULL will be returned.
 386  *
 387  * If tracefs is not enabled in the kernel, the value -%ENODEV will be
 388  * returned.
 389  */
 390 struct dentry *tracefs_create_file(const char *name, umode_t mode,
 391                                    struct dentry *parent, void *data,
 392                                    const struct file_operations *fops)
 393 {
 394         struct dentry *dentry;
 395         struct inode *inode;
 396
 397         if (!(mode & S_IFMT))
 398                 mode |= S_IFREG;
 399         BUG_ON(!S_ISREG(mode));
 400         dentry = start_creating(name, parent);
 401
 402         if (IS_ERR(dentry))
 403                 return NULL;
 404
 405         inode = tracefs_get_inode(dentry->d_sb);
 406         if (unlikely(!inode))
 407                 return failed_creating(dentry);
 408
 409         inode->i_mode = mode;
 410         inode->i_fop = fops ? fops : &tracefs_file_operations;
 411         inode->i_private = data;
 412         d_instantiate(dentry, inode);
 413         fsnotify_create(dentry->d_parent->d_inode, dentry);
 414         return end_creating(dentry);
 415 }
 416
 417 static struct dentry *__create_dir(const char *name, struct dentry *parent,
 418                                    const struct inode_operations *ops)
 419 {
 420         struct dentry *dentry = start_creating(name, parent);
 421         struct inode *inode;
 422
 423         if (IS_ERR(dentry))
 424                 return NULL;
 425
 426         inode = tracefs_get_inode(dentry->d_sb);
 427         if (unlikely(!inode))
 428                 return failed_creating(dentry);
 429
 430         inode->i_mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO;
 431         inode->i_op = ops;
 432         inode->i_fop = &simple_dir_operations;
 433
 434         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 435         inc_nlink(inode);
 436         d_instantiate(dentry, inode);
 437         inc_nlink(dentry->d_parent->d_inode);
 438         fsnotify_mkdir(dentry->d_parent->d_inode, dentry);
 439         return end_creating(dentry);
 440 }
 441
 442 /**
 443  * tracefs_create_dir - create a directory in the tracefs filesystem
 444  * @name: a pointer to a string containing the name of the directory to
 445  *        create.
 446  * @parent: a pointer to the parent dentry for this file.  This should be a
 447  *          directory dentry if set.  If this parameter is NULL, then the
 448  *          directory will be created in the root of the tracefs filesystem.
 449  *
 450  * This function creates a directory in tracefs with the given name.
 451  *
 452  * This function will return a pointer to a dentry if it succeeds.  This
 453  * pointer must be passed to the tracefs_remove() function when the file is
 454  * to be removed. If an error occurs, %NULL will be returned.
 455  *
 456  * If tracing is not enabled in the kernel, the value -%ENODEV will be
 457  * returned.
 458  */
 459 struct dentry *tracefs_create_dir(const char *name, struct dentry *parent)
 460 {
 461         return __create_dir(name, parent, &simple_dir_inode_operations);
 462 }
 463
 464 /**
 465  * tracefs_create_instance_dir - create the tracing instances directory
 466  * @name: The name of the instances directory to create
 467  * @parent: The parent directory that the instances directory will exist
 468  * @mkdir: The function to call when a mkdir is performed.
 469  * @rmdir: The function to call when a rmdir is performed.
 470  *
 471  * Only one instances directory is allowed.
 472  *
 473  * The instances directory is special as it allows for mkdir and rmdir to
 474  * to be done by userspace. When a mkdir or rmdir is performed, the inode
 475  * locks are released and the methhods passed in (@mkdir and @rmdir) are
 476  * called without locks and with the name of the directory being created
 477  * within the instances directory.
 478  *
 479  * Returns the dentry of the instances directory.
 480  */
 481 struct dentry *tracefs_create_instance_dir(const char *name, struct dentry *parent,
 482                                           int (*mkdir)(const char *name),
 483                                           int (*rmdir)(const char *name))
 484 {
 485         struct dentry *dentry;
 486
 487         /* Only allow one instance of the instances directory. */
 488         if (WARN_ON(tracefs_ops.mkdir || tracefs_ops.rmdir))
 489                 return NULL;
 490
 491         dentry = __create_dir(name, parent, &tracefs_dir_inode_operations);
 492         if (!dentry)
 493                 return NULL;
 494
 495         tracefs_ops.mkdir = mkdir;
 496         tracefs_ops.rmdir = rmdir;
 497
 498         return dentry;
 499 }
 500
 501 static int __tracefs_remove(struct dentry *dentry, struct dentry *parent)
 502 {
 503         int ret = 0;
 504
 505         if (simple_positive(dentry)) {
 506                 if (dentry->d_inode) {
 507                         dget(dentry);
 508                         switch (dentry->d_inode->i_mode & S_IFMT) {
 509                         case S_IFDIR:
 510                                 ret = simple_rmdir(parent->d_inode, dentry);
 511                                 break;
 512                         default:
 513                                 simple_unlink(parent->d_inode, dentry);
 514                                 break;
 515                         }
 516                         if (!ret)
 517                                 d_delete(dentry);
 518                         dput(dentry);
 519                 }
 520         }
 521         return ret;
 522 }
 523
 524 /**
 525  * tracefs_remove - removes a file or directory from the tracefs filesystem
 526  * @dentry: a pointer to a the dentry of the file or directory to be
 527  *          removed.
 528  *
 529  * This function removes a file or directory in tracefs that was previously
 530  * created with a call to another tracefs function (like
 531  * tracefs_create_file() or variants thereof.)
 532  */
 533 void tracefs_remove(struct dentry *dentry)
 534 {
 535         struct dentry *parent;
 536         int ret;
 537
 538         if (IS_ERR_OR_NULL(dentry))
 539                 return;
 540
 541         parent = dentry->d_parent;
 542         inode_lock(parent->d_inode);
 543         ret = __tracefs_remove(dentry, parent);
 544         inode_unlock(parent->d_inode);
 545         if (!ret)
 546                 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
 547 }
 548
 549 /**
 550  * tracefs_remove_recursive - recursively removes a directory
 551  * @dentry: a pointer to a the dentry of the directory to be removed.
 552  *
 553  * This function recursively removes a directory tree in tracefs that
 554  * was previously created with a call to another tracefs function
 555  * (like tracefs_create_file() or variants thereof.)
 556  */
 557 void tracefs_remove_recursive(struct dentry *dentry)
 558 {
 559         struct dentry *child, *parent;
 560
 561         if (IS_ERR_OR_NULL(dentry))
 562                 return;
 563
 564         parent = dentry;
 565  down:
 566         inode_lock(parent->d_inode);
 567  loop:
 568         /*
 569          * The parent->d_subdirs is protected by the d_lock. Outside that
 570          * lock, the child can be unlinked and set to be freed which can
 571          * use the d_u.d_child as the rcu head and corrupt this list.
 572          */
 573         spin_lock(&parent->d_lock);
 574         list_for_each_entry(child, &parent->d_subdirs, d_child) {
 575                 if (!simple_positive(child))
 576                         continue;
 577
 578                 /* perhaps simple_empty(child) makes more sense */
 579                 if (!list_empty(&child->d_subdirs)) {
 580                         spin_unlock(&parent->d_lock);
 581                         inode_unlock(parent->d_inode);
 582                         parent = child;
 583                         goto down;
 584                 }
 585
 586                 spin_unlock(&parent->d_lock);
 587
 588                 if (!__tracefs_remove(child, parent))
 589                         simple_release_fs(&tracefs_mount, &tracefs_mount_count);
 590
 591                 /*
 592                  * The parent->d_lock protects agaist child from unlinking
 593                  * from d_subdirs. When releasing the parent->d_lock we can
 594                  * no longer trust that the next pointer is valid.
 595                  * Restart the loop. We'll skip this one with the
 596                  * simple_positive() check.
 597                  */
 598                 goto loop;
 599         }
 600         spin_unlock(&parent->d_lock);
 601
 602         inode_unlock(parent->d_inode);
 603         child = parent;
 604         parent = parent->d_parent;
 605         inode_lock(parent->d_inode);
 606
 607         if (child != dentry)
 608                 /* go up */
 609                 goto loop;
 610
 611         if (!__tracefs_remove(child, parent))
 612                 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
 613         inode_unlock(parent->d_inode);
 614 }
 615
 616 /**
 617  * tracefs_initialized - Tells whether tracefs has been registered
 618  */
 619 bool tracefs_initialized(void)
 620 {
 621         return tracefs_registered;
 622 }
 623
 624 static int __init tracefs_init(void)
 625 {
 626         int retval;
 627
 628         retval = sysfs_create_mount_point(kernel_kobj, "tracing");
 629         if (retval)
 630                 return -EINVAL;
 631
 632         retval = register_filesystem(&trace_fs_type);
 633         if (!retval)
 634                 tracefs_registered = true;
 635
 636         return retval;
 637 }
 638 core_initcall(tracefs_init);