include/linux/kernfs.h

   1 /*
   2  * kernfs.h - pseudo filesystem decoupled from vfs locking
   3  *
   4  * This file is released under the GPLv2.
   5  */
   6
   7 #ifndef __LINUX_KERNFS_H
   8 #define __LINUX_KERNFS_H
   9
  10 #include <linux/kernel.h>
  11 #include <linux/err.h>
  12 #include <linux/list.h>
  13 #include <linux/mutex.h>
  14 #include <linux/idr.h>
  15 #include <linux/lockdep.h>
  16 #include <linux/rbtree.h>
  17 #include <linux/atomic.h>
  18 #include <linux/wait.h>
  19
  20 struct file;
  21 struct dentry;
  22 struct iattr;
  23 struct seq_file;
  24 struct vm_area_struct;
  25 struct super_block;
  26 struct file_system_type;
  27
  28 struct kernfs_open_node;
  29 struct kernfs_iattrs;
  30
  31 enum kernfs_node_type {
  32         KERNFS_DIR              = 0x0001,
  33         KERNFS_FILE             = 0x0002,
  34         KERNFS_LINK             = 0x0004,
  35 };
  36
  37 #define KERNFS_TYPE_MASK        0x000f
  38 #define KERNFS_FLAG_MASK        ~KERNFS_TYPE_MASK
  39
  40 enum kernfs_node_flag {
  41         KERNFS_ACTIVATED        = 0x0010,
  42         KERNFS_NS               = 0x0020,
  43         KERNFS_HAS_SEQ_SHOW     = 0x0040,
  44         KERNFS_HAS_MMAP         = 0x0080,
  45         KERNFS_LOCKDEP          = 0x0100,
  46         KERNFS_SUICIDAL         = 0x0400,
  47         KERNFS_SUICIDED         = 0x0800,
  48         KERNFS_EMPTY_DIR        = 0x1000,
  49 };
  50
  51 /* @flags for kernfs_create_root() */
  52 enum kernfs_root_flag {
  53         /*
  54          * kernfs_nodes are created in the deactivated state and invisible.
  55          * They require explicit kernfs_activate() to become visible.  This
  56          * can be used to make related nodes become visible atomically
  57          * after all nodes are created successfully.
  58          */
  59         KERNFS_ROOT_CREATE_DEACTIVATED          = 0x0001,
  60
  61         /*
  62          * For regular flies, if the opener has CAP_DAC_OVERRIDE, open(2)
  63          * succeeds regardless of the RW permissions.  sysfs had an extra
  64          * layer of enforcement where open(2) fails with -EACCES regardless
  65          * of CAP_DAC_OVERRIDE if the permission doesn't have the
  66          * respective read or write access at all (none of S_IRUGO or
  67          * S_IWUGO) or the respective operation isn't implemented.  The
  68          * following flag enables that behavior.
  69          */
  70         KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK       = 0x0002,
  71 };
  72
  73 /* type-specific structures for kernfs_node union members */
  74 struct kernfs_elem_dir {
  75         unsigned long           subdirs;
  76         /* children rbtree starts here and goes through kn->rb */
  77         struct rb_root          children;
  78
  79         /*
  80          * The kernfs hierarchy this directory belongs to.  This fits
  81          * better directly in kernfs_node but is here to save space.
  82          */
  83         struct kernfs_root      *root;
  84 };
  85
  86 struct kernfs_elem_symlink {
  87         struct kernfs_node      *target_kn;
  88 };
  89
  90 struct kernfs_elem_attr {
  91         const struct kernfs_ops *ops;
  92         struct kernfs_open_node *open;
  93         loff_t                  size;
  94         struct kernfs_node      *notify_next;   /* for kernfs_notify() */
  95 };
  96
  97 /*
  98  * kernfs_node - the building block of kernfs hierarchy.  Each and every
  99  * kernfs node is represented by single kernfs_node.  Most fields are
 100  * private to kernfs and shouldn't be accessed directly by kernfs users.
 101  *
 102  * As long as s_count reference is held, the kernfs_node itself is
 103  * accessible.  Dereferencing elem or any other outer entity requires
 104  * active reference.
 105  */
 106 struct kernfs_node {
 107         atomic_t                count;
 108         atomic_t                active;
 109 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 110         struct lockdep_map      dep_map;
 111 #endif
 112         /*
 113          * Use kernfs_get_parent() and kernfs_name/path() instead of
 114          * accessing the following two fields directly.  If the node is
 115          * never moved to a different parent, it is safe to access the
 116          * parent directly.
 117          */
 118         struct kernfs_node      *parent;
 119         const char              *name;
 120
 121         struct rb_node          rb;
 122
 123         const void              *ns;    /* namespace tag */
 124         unsigned int            hash;   /* ns + name hash */
 125         union {
 126                 struct kernfs_elem_dir          dir;
 127                 struct kernfs_elem_symlink      symlink;
 128                 struct kernfs_elem_attr         attr;
 129         };
 130
 131         void                    *priv;
 132
 133         unsigned short          flags;
 134         umode_t                 mode;
 135         unsigned int            ino;
 136         struct kernfs_iattrs    *iattr;
 137 };
 138
 139 /*
 140  * kernfs_syscall_ops may be specified on kernfs_create_root() to support
 141  * syscalls.  These optional callbacks are invoked on the matching syscalls
 142  * and can perform any kernfs operations which don't necessarily have to be
 143  * the exact operation requested.  An active reference is held for each
 144  * kernfs_node parameter.
 145  */
 146 struct kernfs_syscall_ops {
 147         int (*remount_fs)(struct kernfs_root *root, int *flags, char *data);
 148         int (*show_options)(struct seq_file *sf, struct kernfs_root *root);
 149
 150         int (*mkdir)(struct kernfs_node *parent, const char *name,
 151                      umode_t mode);
 152         int (*rmdir)(struct kernfs_node *kn);
 153         int (*rename)(struct kernfs_node *kn, struct kernfs_node *new_parent,
 154                       const char *new_name);
 155 };
 156
 157 struct kernfs_root {
 158         /* published fields */
 159         struct kernfs_node      *kn;
 160         unsigned int            flags;  /* KERNFS_ROOT_* flags */
 161
 162         /* private fields, do not use outside kernfs proper */
 163         struct ida              ino_ida;
 164         struct kernfs_syscall_ops *syscall_ops;
 165
 166         /* list of kernfs_super_info of this root, protected by kernfs_mutex */
 167         struct list_head        supers;
 168
 169         wait_queue_head_t       deactivate_waitq;
 170 };
 171
 172 struct kernfs_open_file {
 173         /* published fields */
 174         struct kernfs_node      *kn;
 175         struct file             *file;
 176         void                    *priv;
 177
 178         /* private fields, do not use outside kernfs proper */
 179         struct mutex            mutex;
 180         int                     event;
 181         struct list_head        list;
 182         char                    *prealloc_buf;
 183
 184         size_t                  atomic_write_len;
 185         bool                    mmapped;
 186         const struct vm_operations_struct *vm_ops;
 187 };
 188
 189 struct kernfs_ops {
 190         /*
 191          * Read is handled by either seq_file or raw_read().
 192          *
 193          * If seq_show() is present, seq_file path is active.  Other seq
 194          * operations are optional and if not implemented, the behavior is
 195          * equivalent to single_open().  @sf->private points to the
 196          * associated kernfs_open_file.
 197          *
 198          * read() is bounced through kernel buffer and a read larger than
 199          * PAGE_SIZE results in partial operation of PAGE_SIZE.
 200          */
 201         int (*seq_show)(struct seq_file *sf, void *v);
 202
 203         void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
 204         void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
 205         void (*seq_stop)(struct seq_file *sf, void *v);
 206
 207         ssize_t (*read)(struct kernfs_open_file *of, char *buf, size_t bytes,
 208                         loff_t off);
 209
 210         /*
 211          * write() is bounced through kernel buffer.  If atomic_write_len
 212          * is not set, a write larger than PAGE_SIZE results in partial
 213          * operations of PAGE_SIZE chunks.  If atomic_write_len is set,
 214          * writes upto the specified size are executed atomically but
 215          * larger ones are rejected with -E2BIG.
 216          */
 217         size_t atomic_write_len;
 218         /*
 219          * "prealloc" causes a buffer to be allocated at open for
 220          * all read/write requests.  As ->seq_show uses seq_read()
 221          * which does its own allocation, it is incompatible with
 222          * ->prealloc.  Provide ->read and ->write with ->prealloc.
 223          */
 224         bool prealloc;
 225         ssize_t (*write)(struct kernfs_open_file *of, char *buf, size_t bytes,
 226                          loff_t off);
 227
 228         int (*mmap)(struct kernfs_open_file *of, struct vm_area_struct *vma);
 229
 230 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 231         struct lock_class_key   lockdep_key;
 232 #endif
 233 };
 234
 235 #ifdef CONFIG_KERNFS
 236
 237 static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
 238 {
 239         return kn->flags & KERNFS_TYPE_MASK;
 240 }
 241
 242 /**
 243  * kernfs_enable_ns - enable namespace under a directory
 244  * @kn: directory of interest, should be empty
 245  *
 246  * This is to be called right after @kn is created to enable namespace
 247  * under it.  All children of @kn must have non-NULL namespace tags and
 248  * only the ones which match the super_block's tag will be visible.
 249  */
 250 static inline void kernfs_enable_ns(struct kernfs_node *kn)
 251 {
 252         WARN_ON_ONCE(kernfs_type(kn) != KERNFS_DIR);
 253         WARN_ON_ONCE(!RB_EMPTY_ROOT(&kn->dir.children));
 254         kn->flags |= KERNFS_NS;
 255 }
 256
 257 /**
 258  * kernfs_ns_enabled - test whether namespace is enabled
 259  * @kn: the node to test
 260  *
 261  * Test whether namespace filtering is enabled for the children of @ns.
 262  */
 263 static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
 264 {
 265         return kn->flags & KERNFS_NS;
 266 }
 267
 268 int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen);
 269 size_t kernfs_path_len(struct kernfs_node *kn);
 270 int kernfs_path_from_node(struct kernfs_node *root_kn, struct kernfs_node *kn,
 271                           char *buf, size_t buflen);
 272 char *kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen);
 273 void pr_cont_kernfs_name(struct kernfs_node *kn);
 274 void pr_cont_kernfs_path(struct kernfs_node *kn);
 275 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn);
 276 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
 277                                            const char *name, const void *ns);
 278 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
 279                                            const char *path, const void *ns);
 280 void kernfs_get(struct kernfs_node *kn);
 281 void kernfs_put(struct kernfs_node *kn);
 282
 283 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry);
 284 struct kernfs_root *kernfs_root_from_sb(struct super_block *sb);
 285 struct inode *kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn);
 286
 287 struct dentry *kernfs_node_dentry(struct kernfs_node *kn,
 288                                   struct super_block *sb);
 289 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
 290                                        unsigned int flags, void *priv);
 291 void kernfs_destroy_root(struct kernfs_root *root);
 292
 293 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
 294                                          const char *name, umode_t mode,
 295                                          void *priv, const void *ns);
 296 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
 297                                             const char *name);
 298 struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent,
 299                                          const char *name,
 300                                          umode_t mode, loff_t size,
 301                                          const struct kernfs_ops *ops,
 302                                          void *priv, const void *ns,
 303                                          struct lock_class_key *key);
 304 struct kernfs_node *kernfs_create_link(struct kernfs_node *parent,
 305                                        const char *name,
 306                                        struct kernfs_node *target);
 307 void kernfs_activate(struct kernfs_node *kn);
 308 void kernfs_remove(struct kernfs_node *kn);
 309 void kernfs_break_active_protection(struct kernfs_node *kn);
 310 void kernfs_unbreak_active_protection(struct kernfs_node *kn);
 311 bool kernfs_remove_self(struct kernfs_node *kn);
 312 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
 313                              const void *ns);
 314 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
 315                      const char *new_name, const void *new_ns);
 316 int kernfs_setattr(struct kernfs_node *kn, const struct iattr *iattr);
 317 void kernfs_notify(struct kernfs_node *kn);
 318
 319 const void *kernfs_super_ns(struct super_block *sb);
 320 struct dentry *kernfs_mount_ns(struct file_system_type *fs_type, int flags,
 321                                struct kernfs_root *root, unsigned long magic,
 322                                bool *new_sb_created, const void *ns);
 323 void kernfs_kill_sb(struct super_block *sb);
 324 struct super_block *kernfs_pin_sb(struct kernfs_root *root, const void *ns);
 325
 326 void kernfs_init(void);
 327
 328 #else   /* CONFIG_KERNFS */
 329
 330 static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
 331 { return 0; }   /* whatever */
 332
 333 static inline void kernfs_enable_ns(struct kernfs_node *kn) { }
 334
 335 static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
 336 { return false; }
 337
 338 static inline int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
 339 { return -ENOSYS; }
 340
 341 static inline size_t kernfs_path_len(struct kernfs_node *kn)
 342 { return 0; }
 343
 344 static inline char *kernfs_path(struct kernfs_node *kn, char *buf,
 345                                 size_t buflen)
 346 { return NULL; }
 347
 348 static inline void pr_cont_kernfs_name(struct kernfs_node *kn) { }
 349 static inline void pr_cont_kernfs_path(struct kernfs_node *kn) { }
 350
 351 static inline struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
 352 { return NULL; }
 353
 354 static inline struct kernfs_node *
 355 kernfs_find_and_get_ns(struct kernfs_node *parent, const char *name,
 356                        const void *ns)
 357 { return NULL; }
 358 static inline struct kernfs_node *
 359 kernfs_walk_and_get_ns(struct kernfs_node *parent, const char *path,
 360                        const void *ns)
 361 { return NULL; }
 362
 363 static inline void kernfs_get(struct kernfs_node *kn) { }
 364 static inline void kernfs_put(struct kernfs_node *kn) { }
 365
 366 static inline struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
 367 { return NULL; }
 368
 369 static inline struct kernfs_root *kernfs_root_from_sb(struct super_block *sb)
 370 { return NULL; }
 371
 372 static inline struct inode *
 373 kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn)
 374 { return NULL; }
 375
 376 static inline struct kernfs_root *
 377 kernfs_create_root(struct kernfs_syscall_ops *scops, unsigned int flags,
 378                    void *priv)
 379 { return ERR_PTR(-ENOSYS); }
 380
 381 static inline void kernfs_destroy_root(struct kernfs_root *root) { }
 382
 383 static inline struct kernfs_node *
 384 kernfs_create_dir_ns(struct kernfs_node *parent, const char *name,
 385                      umode_t mode, void *priv, const void *ns)
 386 { return ERR_PTR(-ENOSYS); }
 387
 388 static inline struct kernfs_node *
 389 __kernfs_create_file(struct kernfs_node *parent, const char *name,
 390                      umode_t mode, loff_t size, const struct kernfs_ops *ops,
 391                      void *priv, const void *ns, struct lock_class_key *key)
 392 { return ERR_PTR(-ENOSYS); }
 393
 394 static inline struct kernfs_node *
 395 kernfs_create_link(struct kernfs_node *parent, const char *name,
 396                    struct kernfs_node *target)
 397 { return ERR_PTR(-ENOSYS); }
 398
 399 static inline void kernfs_activate(struct kernfs_node *kn) { }
 400
 401 static inline void kernfs_remove(struct kernfs_node *kn) { }
 402
 403 static inline bool kernfs_remove_self(struct kernfs_node *kn)
 404 { return false; }
 405
 406 static inline int kernfs_remove_by_name_ns(struct kernfs_node *kn,
 407                                            const char *name, const void *ns)
 408 { return -ENOSYS; }
 409
 410 static inline int kernfs_rename_ns(struct kernfs_node *kn,
 411                                    struct kernfs_node *new_parent,
 412                                    const char *new_name, const void *new_ns)
 413 { return -ENOSYS; }
 414
 415 static inline int kernfs_setattr(struct kernfs_node *kn,
 416                                  const struct iattr *iattr)
 417 { return -ENOSYS; }
 418
 419 static inline void kernfs_notify(struct kernfs_node *kn) { }
 420
 421 static inline const void *kernfs_super_ns(struct super_block *sb)
 422 { return NULL; }
 423
 424 static inline struct dentry *
 425 kernfs_mount_ns(struct file_system_type *fs_type, int flags,
 426                 struct kernfs_root *root, unsigned long magic,
 427                 bool *new_sb_created, const void *ns)
 428 { return ERR_PTR(-ENOSYS); }
 429
 430 static inline void kernfs_kill_sb(struct super_block *sb) { }
 431
 432 static inline void kernfs_init(void) { }
 433
 434 #endif  /* CONFIG_KERNFS */
 435
 436 static inline struct kernfs_node *
 437 kernfs_find_and_get(struct kernfs_node *kn, const char *name)
 438 {
 439         return kernfs_find_and_get_ns(kn, name, NULL);
 440 }
 441
 442 static inline struct kernfs_node *
 443 kernfs_walk_and_get(struct kernfs_node *kn, const char *path)
 444 {
 445         return kernfs_walk_and_get_ns(kn, path, NULL);
 446 }
 447
 448 static inline struct kernfs_node *
 449 kernfs_create_dir(struct kernfs_node *parent, const char *name, umode_t mode,
 450                   void *priv)
 451 {
 452         return kernfs_create_dir_ns(parent, name, mode, priv, NULL);
 453 }
 454
 455 static inline struct kernfs_node *
 456 kernfs_create_file_ns(struct kernfs_node *parent, const char *name,
 457                       umode_t mode, loff_t size, const struct kernfs_ops *ops,
 458                       void *priv, const void *ns)
 459 {
 460         struct lock_class_key *key = NULL;
 461
 462 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 463         key = (struct lock_class_key *)&ops->lockdep_key;
 464 #endif
 465         return __kernfs_create_file(parent, name, mode, size, ops, priv, ns,
 466                                     key);
 467 }
 468
 469 static inline struct kernfs_node *
 470 kernfs_create_file(struct kernfs_node *parent, const char *name, umode_t mode,
 471                    loff_t size, const struct kernfs_ops *ops, void *priv)
 472 {
 473         return kernfs_create_file_ns(parent, name, mode, size, ops, priv, NULL);
 474 }
 475
 476 static inline int kernfs_remove_by_name(struct kernfs_node *parent,
 477                                         const char *name)
 478 {
 479         return kernfs_remove_by_name_ns(parent, name, NULL);
 480 }
 481
 482 static inline int kernfs_rename(struct kernfs_node *kn,
 483                                 struct kernfs_node *new_parent,
 484                                 const char *new_name)
 485 {
 486         return kernfs_rename_ns(kn, new_parent, new_name, NULL);
 487 }
 488
 489 static inline struct dentry *
 490 kernfs_mount(struct file_system_type *fs_type, int flags,
 491                 struct kernfs_root *root, unsigned long magic,
 492                 bool *new_sb_created)
 493 {
 494         return kernfs_mount_ns(fs_type, flags, root,
 495                                 magic, new_sb_created, NULL);
 496 }
 497
 498 #endif  /* __LINUX_KERNFS_H */