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         int (*show_path)(struct seq_file *sf, struct kernfs_node *kn,
 156                          struct kernfs_root *root);
 157 };
 158
 159 struct kernfs_root {
 160         /* published fields */
 161         struct kernfs_node      *kn;
 162         unsigned int            flags;  /* KERNFS_ROOT_* flags */
 163
 164         /* private fields, do not use outside kernfs proper */
 165         struct ida              ino_ida;
 166         struct kernfs_syscall_ops *syscall_ops;
 167
 168         /* list of kernfs_super_info of this root, protected by kernfs_mutex */
 169         struct list_head        supers;
 170
 171         wait_queue_head_t       deactivate_waitq;
 172 };
 173
 174 struct kernfs_open_file {
 175         /* published fields */
 176         struct kernfs_node      *kn;
 177         struct file             *file;
 178         void                    *priv;
 179
 180         /* private fields, do not use outside kernfs proper */
 181         struct mutex            mutex;
 182         int                     event;
 183         struct list_head        list;
 184         char                    *prealloc_buf;
 185
 186         size_t                  atomic_write_len;
 187         bool                    mmapped;
 188         const struct vm_operations_struct *vm_ops;
 189 };
 190
 191 struct kernfs_ops {
 192         /*
 193          * Read is handled by either seq_file or raw_read().
 194          *
 195          * If seq_show() is present, seq_file path is active.  Other seq
 196          * operations are optional and if not implemented, the behavior is
 197          * equivalent to single_open().  @sf->private points to the
 198          * associated kernfs_open_file.
 199          *
 200          * read() is bounced through kernel buffer and a read larger than
 201          * PAGE_SIZE results in partial operation of PAGE_SIZE.
 202          */
 203         int (*seq_show)(struct seq_file *sf, void *v);
 204
 205         void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
 206         void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
 207         void (*seq_stop)(struct seq_file *sf, void *v);
 208
 209         ssize_t (*read)(struct kernfs_open_file *of, char *buf, size_t bytes,
 210                         loff_t off);
 211
 212         /*
 213          * write() is bounced through kernel buffer.  If atomic_write_len
 214          * is not set, a write larger than PAGE_SIZE results in partial
 215          * operations of PAGE_SIZE chunks.  If atomic_write_len is set,
 216          * writes upto the specified size are executed atomically but
 217          * larger ones are rejected with -E2BIG.
 218          */
 219         size_t atomic_write_len;
 220         /*
 221          * "prealloc" causes a buffer to be allocated at open for
 222          * all read/write requests.  As ->seq_show uses seq_read()
 223          * which does its own allocation, it is incompatible with
 224          * ->prealloc.  Provide ->read and ->write with ->prealloc.
 225          */
 226         bool prealloc;
 227         ssize_t (*write)(struct kernfs_open_file *of, char *buf, size_t bytes,
 228                          loff_t off);
 229
 230         int (*mmap)(struct kernfs_open_file *of, struct vm_area_struct *vma);
 231
 232 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 233         struct lock_class_key   lockdep_key;
 234 #endif
 235 };
 236
 237 #ifdef CONFIG_KERNFS
 238
 239 static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
 240 {
 241         return kn->flags & KERNFS_TYPE_MASK;
 242 }
 243
 244 /**
 245  * kernfs_enable_ns - enable namespace under a directory
 246  * @kn: directory of interest, should be empty
 247  *
 248  * This is to be called right after @kn is created to enable namespace
 249  * under it.  All children of @kn must have non-NULL namespace tags and
 250  * only the ones which match the super_block's tag will be visible.
 251  */
 252 static inline void kernfs_enable_ns(struct kernfs_node *kn)
 253 {
 254         WARN_ON_ONCE(kernfs_type(kn) != KERNFS_DIR);
 255         WARN_ON_ONCE(!RB_EMPTY_ROOT(&kn->dir.children));
 256         kn->flags |= KERNFS_NS;
 257 }
 258
 259 /**
 260  * kernfs_ns_enabled - test whether namespace is enabled
 261  * @kn: the node to test
 262  *
 263  * Test whether namespace filtering is enabled for the children of @ns.
 264  */
 265 static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
 266 {
 267         return kn->flags & KERNFS_NS;
 268 }
 269
 270 int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen);
 271 size_t kernfs_path_len(struct kernfs_node *kn);
 272 int kernfs_path_from_node(struct kernfs_node *root_kn, struct kernfs_node *kn,
 273                           char *buf, size_t buflen);
 274 char *kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen);
 275 void pr_cont_kernfs_name(struct kernfs_node *kn);
 276 void pr_cont_kernfs_path(struct kernfs_node *kn);
 277 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn);
 278 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
 279                                            const char *name, const void *ns);
 280 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
 281                                            const char *path, const void *ns);
 282 void kernfs_get(struct kernfs_node *kn);
 283 void kernfs_put(struct kernfs_node *kn);
 284
 285 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry);
 286 struct kernfs_root *kernfs_root_from_sb(struct super_block *sb);
 287 struct inode *kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn);
 288
 289 struct dentry *kernfs_node_dentry(struct kernfs_node *kn,
 290                                   struct super_block *sb);
 291 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
 292                                        unsigned int flags, void *priv);
 293 void kernfs_destroy_root(struct kernfs_root *root);
 294
 295 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
 296                                          const char *name, umode_t mode,
 297                                          void *priv, const void *ns);
 298 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
 299                                             const char *name);
 300 struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent,
 301                                          const char *name,
 302                                          umode_t mode, loff_t size,
 303                                          const struct kernfs_ops *ops,
 304                                          void *priv, const void *ns,
 305                                          struct lock_class_key *key);
 306 struct kernfs_node *kernfs_create_link(struct kernfs_node *parent,
 307                                        const char *name,
 308                                        struct kernfs_node *target);
 309 void kernfs_activate(struct kernfs_node *kn);
 310 void kernfs_remove(struct kernfs_node *kn);
 311 void kernfs_break_active_protection(struct kernfs_node *kn);
 312 void kernfs_unbreak_active_protection(struct kernfs_node *kn);
 313 bool kernfs_remove_self(struct kernfs_node *kn);
 314 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
 315                              const void *ns);
 316 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
 317                      const char *new_name, const void *new_ns);
 318 int kernfs_setattr(struct kernfs_node *kn, const struct iattr *iattr);
 319 void kernfs_notify(struct kernfs_node *kn);
 320
 321 const void *kernfs_super_ns(struct super_block *sb);
 322 struct dentry *kernfs_mount_ns(struct file_system_type *fs_type, int flags,
 323                                struct kernfs_root *root, unsigned long magic,
 324                                bool *new_sb_created, const void *ns);
 325 void kernfs_kill_sb(struct super_block *sb);
 326 struct super_block *kernfs_pin_sb(struct kernfs_root *root, const void *ns);
 327
 328 void kernfs_init(void);
 329
 330 #else   /* CONFIG_KERNFS */
 331
 332 static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
 333 { return 0; }   /* whatever */
 334
 335 static inline void kernfs_enable_ns(struct kernfs_node *kn) { }
 336
 337 static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
 338 { return false; }
 339
 340 static inline int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
 341 { return -ENOSYS; }
 342
 343 static inline size_t kernfs_path_len(struct kernfs_node *kn)
 344 { return 0; }
 345
 346 static inline char *kernfs_path(struct kernfs_node *kn, char *buf,
 347                                 size_t buflen)
 348 { return NULL; }
 349
 350 static inline void pr_cont_kernfs_name(struct kernfs_node *kn) { }
 351 static inline void pr_cont_kernfs_path(struct kernfs_node *kn) { }
 352
 353 static inline struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
 354 { return NULL; }
 355
 356 static inline struct kernfs_node *
 357 kernfs_find_and_get_ns(struct kernfs_node *parent, const char *name,
 358                        const void *ns)
 359 { return NULL; }
 360 static inline struct kernfs_node *
 361 kernfs_walk_and_get_ns(struct kernfs_node *parent, const char *path,
 362                        const void *ns)
 363 { return NULL; }
 364
 365 static inline void kernfs_get(struct kernfs_node *kn) { }
 366 static inline void kernfs_put(struct kernfs_node *kn) { }
 367
 368 static inline struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
 369 { return NULL; }
 370
 371 static inline struct kernfs_root *kernfs_root_from_sb(struct super_block *sb)
 372 { return NULL; }
 373
 374 static inline struct inode *
 375 kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn)
 376 { return NULL; }
 377
 378 static inline struct kernfs_root *
 379 kernfs_create_root(struct kernfs_syscall_ops *scops, unsigned int flags,
 380                    void *priv)
 381 { return ERR_PTR(-ENOSYS); }
 382
 383 static inline void kernfs_destroy_root(struct kernfs_root *root) { }
 384
 385 static inline struct kernfs_node *
 386 kernfs_create_dir_ns(struct kernfs_node *parent, const char *name,
 387                      umode_t mode, void *priv, const void *ns)
 388 { return ERR_PTR(-ENOSYS); }
 389
 390 static inline struct kernfs_node *
 391 __kernfs_create_file(struct kernfs_node *parent, const char *name,
 392                      umode_t mode, loff_t size, const struct kernfs_ops *ops,
 393                      void *priv, const void *ns, struct lock_class_key *key)
 394 { return ERR_PTR(-ENOSYS); }
 395
 396 static inline struct kernfs_node *
 397 kernfs_create_link(struct kernfs_node *parent, const char *name,
 398                    struct kernfs_node *target)
 399 { return ERR_PTR(-ENOSYS); }
 400
 401 static inline void kernfs_activate(struct kernfs_node *kn) { }
 402
 403 static inline void kernfs_remove(struct kernfs_node *kn) { }
 404
 405 static inline bool kernfs_remove_self(struct kernfs_node *kn)
 406 { return false; }
 407
 408 static inline int kernfs_remove_by_name_ns(struct kernfs_node *kn,
 409                                            const char *name, const void *ns)
 410 { return -ENOSYS; }
 411
 412 static inline int kernfs_rename_ns(struct kernfs_node *kn,
 413                                    struct kernfs_node *new_parent,
 414                                    const char *new_name, const void *new_ns)
 415 { return -ENOSYS; }
 416
 417 static inline int kernfs_setattr(struct kernfs_node *kn,
 418                                  const struct iattr *iattr)
 419 { return -ENOSYS; }
 420
 421 static inline void kernfs_notify(struct kernfs_node *kn) { }
 422
 423 static inline const void *kernfs_super_ns(struct super_block *sb)
 424 { return NULL; }
 425
 426 static inline struct dentry *
 427 kernfs_mount_ns(struct file_system_type *fs_type, int flags,
 428                 struct kernfs_root *root, unsigned long magic,
 429                 bool *new_sb_created, const void *ns)
 430 { return ERR_PTR(-ENOSYS); }
 431
 432 static inline void kernfs_kill_sb(struct super_block *sb) { }
 433
 434 static inline void kernfs_init(void) { }
 435
 436 #endif  /* CONFIG_KERNFS */
 437
 438 static inline struct kernfs_node *
 439 kernfs_find_and_get(struct kernfs_node *kn, const char *name)
 440 {
 441         return kernfs_find_and_get_ns(kn, name, NULL);
 442 }
 443
 444 static inline struct kernfs_node *
 445 kernfs_walk_and_get(struct kernfs_node *kn, const char *path)
 446 {
 447         return kernfs_walk_and_get_ns(kn, path, NULL);
 448 }
 449
 450 static inline struct kernfs_node *
 451 kernfs_create_dir(struct kernfs_node *parent, const char *name, umode_t mode,
 452                   void *priv)
 453 {
 454         return kernfs_create_dir_ns(parent, name, mode, priv, NULL);
 455 }
 456
 457 static inline struct kernfs_node *
 458 kernfs_create_file_ns(struct kernfs_node *parent, const char *name,
 459                       umode_t mode, loff_t size, const struct kernfs_ops *ops,
 460                       void *priv, const void *ns)
 461 {
 462         struct lock_class_key *key = NULL;
 463
 464 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 465         key = (struct lock_class_key *)&ops->lockdep_key;
 466 #endif
 467         return __kernfs_create_file(parent, name, mode, size, ops, priv, ns,
 468                                     key);
 469 }
 470
 471 static inline struct kernfs_node *
 472 kernfs_create_file(struct kernfs_node *parent, const char *name, umode_t mode,
 473                    loff_t size, const struct kernfs_ops *ops, void *priv)
 474 {
 475         return kernfs_create_file_ns(parent, name, mode, size, ops, priv, NULL);
 476 }
 477
 478 static inline int kernfs_remove_by_name(struct kernfs_node *parent,
 479                                         const char *name)
 480 {
 481         return kernfs_remove_by_name_ns(parent, name, NULL);
 482 }
 483
 484 static inline int kernfs_rename(struct kernfs_node *kn,
 485                                 struct kernfs_node *new_parent,
 486                                 const char *new_name)
 487 {
 488         return kernfs_rename_ns(kn, new_parent, new_name, NULL);
 489 }
 490
 491 static inline struct dentry *
 492 kernfs_mount(struct file_system_type *fs_type, int flags,
 493                 struct kernfs_root *root, unsigned long magic,
 494                 bool *new_sb_created)
 495 {
 496         return kernfs_mount_ns(fs_type, flags, root,
 497                                 magic, new_sb_created, NULL);
 498 }
 499
 500 #endif  /* __LINUX_KERNFS_H */