1 /* 2 * (C) 2001 Clemson University and The University of Chicago 3 * 4 * See COPYING in top-level directory. 5 */ 6 7 /* 8 * The ORANGEFS Linux kernel support allows ORANGEFS volumes to be mounted and 9 * accessed through the Linux VFS (i.e. using standard I/O system calls). 10 * This support is only needed on clients that wish to mount the file system. 11 * 12 */ 13 14 /* 15 * Declarations and macros for the ORANGEFS Linux kernel support. 16 */ 17 18 #ifndef __ORANGEFSKERNEL_H 19 #define __ORANGEFSKERNEL_H 20 21 #include <linux/kernel.h> 22 #include <linux/moduleparam.h> 23 #include <linux/statfs.h> 24 #include <linux/backing-dev.h> 25 #include <linux/device.h> 26 #include <linux/mpage.h> 27 #include <linux/namei.h> 28 #include <linux/errno.h> 29 #include <linux/init.h> 30 #include <linux/module.h> 31 #include <linux/slab.h> 32 #include <linux/types.h> 33 #include <linux/fs.h> 34 #include <linux/vmalloc.h> 35 36 #include <linux/aio.h> 37 #include <linux/posix_acl.h> 38 #include <linux/posix_acl_xattr.h> 39 #include <linux/compat.h> 40 #include <linux/mount.h> 41 #include <linux/uaccess.h> 42 #include <linux/atomic.h> 43 #include <linux/uio.h> 44 #include <linux/sched/signal.h> 45 #include <linux/mm.h> 46 #include <linux/wait.h> 47 #include <linux/dcache.h> 48 #include <linux/pagemap.h> 49 #include <linux/poll.h> 50 #include <linux/rwsem.h> 51 #include <linux/xattr.h> 52 #include <linux/exportfs.h> 53 54 #include <asm/unaligned.h> 55 56 #include "orangefs-dev-proto.h" 57 58 #ifdef ORANGEFS_KERNEL_DEBUG 59 #define ORANGEFS_DEFAULT_OP_TIMEOUT_SECS 10 60 #else 61 #define ORANGEFS_DEFAULT_OP_TIMEOUT_SECS 20 62 #endif 63 64 #define ORANGEFS_BUFMAP_WAIT_TIMEOUT_SECS 30 65 66 #define ORANGEFS_DEFAULT_SLOT_TIMEOUT_SECS 900 /* 15 minutes */ 67 68 #define ORANGEFS_REQDEVICE_NAME "pvfs2-req" 69 70 #define ORANGEFS_DEVREQ_MAGIC 0x20030529 71 #define ORANGEFS_LINK_MAX 0x000000FF 72 #define ORANGEFS_PURGE_RETRY_COUNT 0x00000005 73 #define ORANGEFS_MAX_NUM_OPTIONS 0x00000004 74 #define ORANGEFS_MAX_MOUNT_OPT_LEN 0x00000080 75 #define ORANGEFS_MAX_FSKEY_LEN 64 76 77 #define MAX_DEV_REQ_UPSIZE (2 * sizeof(__s32) + \ 78 sizeof(__u64) + sizeof(struct orangefs_upcall_s)) 79 #define MAX_DEV_REQ_DOWNSIZE (2 * sizeof(__s32) + \ 80 sizeof(__u64) + sizeof(struct orangefs_downcall_s)) 81 82 /* 83 * valid orangefs kernel operation states 84 * 85 * unknown - op was just initialized 86 * waiting - op is on request_list (upward bound) 87 * inprogr - op is in progress (waiting for downcall) 88 * serviced - op has matching downcall; ok 89 * purged - op has to start a timer since client-core 90 * exited uncleanly before servicing op 91 * given up - submitter has given up waiting for it 92 */ 93 enum orangefs_vfs_op_states { 94 OP_VFS_STATE_UNKNOWN = 0, 95 OP_VFS_STATE_WAITING = 1, 96 OP_VFS_STATE_INPROGR = 2, 97 OP_VFS_STATE_SERVICED = 4, 98 OP_VFS_STATE_PURGED = 8, 99 OP_VFS_STATE_GIVEN_UP = 16, 100 }; 101 102 /* 103 * orangefs kernel memory related flags 104 */ 105 106 #if ((defined ORANGEFS_KERNEL_DEBUG) && (defined CONFIG_DEBUG_SLAB)) 107 #define ORANGEFS_CACHE_CREATE_FLAGS SLAB_RED_ZONE 108 #else 109 #define ORANGEFS_CACHE_CREATE_FLAGS 0 110 #endif /* ((defined ORANGEFS_KERNEL_DEBUG) && (defined CONFIG_DEBUG_SLAB)) */ 111 112 extern int orangefs_init_acl(struct inode *inode, struct inode *dir); 113 extern const struct xattr_handler *orangefs_xattr_handlers[]; 114 115 extern struct posix_acl *orangefs_get_acl(struct inode *inode, int type); 116 extern int orangefs_set_acl(struct inode *inode, struct posix_acl *acl, int type); 117 118 /* 119 * Redefine xtvec structure so that we could move helper functions out of 120 * the define 121 */ 122 struct xtvec { 123 __kernel_off_t xtv_off; /* must be off_t */ 124 __kernel_size_t xtv_len; /* must be size_t */ 125 }; 126 127 /* 128 * orangefs data structures 129 */ 130 struct orangefs_kernel_op_s { 131 enum orangefs_vfs_op_states op_state; 132 __u64 tag; 133 134 /* 135 * Set uses_shared_memory to non zero if this operation uses 136 * shared memory. If true, then a retry on the op must also 137 * get a new shared memory buffer and re-populate it. 138 * Cancels don't care - it only matters for service_operation() 139 * retry logics and cancels don't go through it anymore. It 140 * safely stays non-zero when we use it as slot_to_free. 141 */ 142 union { 143 int uses_shared_memory; 144 int slot_to_free; 145 }; 146 147 struct orangefs_upcall_s upcall; 148 struct orangefs_downcall_s downcall; 149 150 struct completion waitq; 151 spinlock_t lock; 152 153 int attempts; 154 155 struct list_head list; 156 }; 157 158 #define set_op_state_waiting(op) ((op)->op_state = OP_VFS_STATE_WAITING) 159 #define set_op_state_inprogress(op) ((op)->op_state = OP_VFS_STATE_INPROGR) 160 #define set_op_state_given_up(op) ((op)->op_state = OP_VFS_STATE_GIVEN_UP) 161 static inline void set_op_state_serviced(struct orangefs_kernel_op_s *op) 162 { 163 op->op_state = OP_VFS_STATE_SERVICED; 164 complete(&op->waitq); 165 } 166 167 #define op_state_waiting(op) ((op)->op_state & OP_VFS_STATE_WAITING) 168 #define op_state_in_progress(op) ((op)->op_state & OP_VFS_STATE_INPROGR) 169 #define op_state_serviced(op) ((op)->op_state & OP_VFS_STATE_SERVICED) 170 #define op_state_purged(op) ((op)->op_state & OP_VFS_STATE_PURGED) 171 #define op_state_given_up(op) ((op)->op_state & OP_VFS_STATE_GIVEN_UP) 172 #define op_is_cancel(op) ((op)->upcall.type == ORANGEFS_VFS_OP_CANCEL) 173 174 void op_release(struct orangefs_kernel_op_s *op); 175 176 extern void orangefs_bufmap_put(int); 177 static inline void put_cancel(struct orangefs_kernel_op_s *op) 178 { 179 orangefs_bufmap_put(op->slot_to_free); 180 op_release(op); 181 } 182 183 static inline void set_op_state_purged(struct orangefs_kernel_op_s *op) 184 { 185 spin_lock(&op->lock); 186 if (unlikely(op_is_cancel(op))) { 187 list_del_init(&op->list); 188 spin_unlock(&op->lock); 189 put_cancel(op); 190 } else { 191 op->op_state |= OP_VFS_STATE_PURGED; 192 complete(&op->waitq); 193 spin_unlock(&op->lock); 194 } 195 } 196 197 /* per inode private orangefs info */ 198 struct orangefs_inode_s { 199 struct orangefs_object_kref refn; 200 char link_target[ORANGEFS_NAME_MAX]; 201 __s64 blksize; 202 /* 203 * Reading/Writing Extended attributes need to acquire the appropriate 204 * reader/writer semaphore on the orangefs_inode_s structure. 205 */ 206 struct rw_semaphore xattr_sem; 207 208 struct inode vfs_inode; 209 sector_t last_failed_block_index_read; 210 211 /* 212 * State of in-memory attributes not yet flushed to disk associated 213 * with this object 214 */ 215 unsigned long pinode_flags; 216 217 unsigned long getattr_time; 218 }; 219 220 #define P_ATIME_FLAG 0 221 #define P_MTIME_FLAG 1 222 #define P_CTIME_FLAG 2 223 #define P_MODE_FLAG 3 224 225 #define ClearAtimeFlag(pinode) clear_bit(P_ATIME_FLAG, &(pinode)->pinode_flags) 226 #define SetAtimeFlag(pinode) set_bit(P_ATIME_FLAG, &(pinode)->pinode_flags) 227 #define AtimeFlag(pinode) test_bit(P_ATIME_FLAG, &(pinode)->pinode_flags) 228 229 #define ClearMtimeFlag(pinode) clear_bit(P_MTIME_FLAG, &(pinode)->pinode_flags) 230 #define SetMtimeFlag(pinode) set_bit(P_MTIME_FLAG, &(pinode)->pinode_flags) 231 #define MtimeFlag(pinode) test_bit(P_MTIME_FLAG, &(pinode)->pinode_flags) 232 233 #define ClearCtimeFlag(pinode) clear_bit(P_CTIME_FLAG, &(pinode)->pinode_flags) 234 #define SetCtimeFlag(pinode) set_bit(P_CTIME_FLAG, &(pinode)->pinode_flags) 235 #define CtimeFlag(pinode) test_bit(P_CTIME_FLAG, &(pinode)->pinode_flags) 236 237 #define ClearModeFlag(pinode) clear_bit(P_MODE_FLAG, &(pinode)->pinode_flags) 238 #define SetModeFlag(pinode) set_bit(P_MODE_FLAG, &(pinode)->pinode_flags) 239 #define ModeFlag(pinode) test_bit(P_MODE_FLAG, &(pinode)->pinode_flags) 240 241 /* per superblock private orangefs info */ 242 struct orangefs_sb_info_s { 243 struct orangefs_khandle root_khandle; 244 __s32 fs_id; 245 int id; 246 int flags; 247 #define ORANGEFS_OPT_INTR 0x01 248 #define ORANGEFS_OPT_LOCAL_LOCK 0x02 249 char devname[ORANGEFS_MAX_SERVER_ADDR_LEN]; 250 struct super_block *sb; 251 int mount_pending; 252 struct list_head list; 253 }; 254 255 /* 256 * structure that holds the state of any async I/O operation issued 257 * through the VFS. Needed especially to handle cancellation requests 258 * or even completion notification so that the VFS client-side daemon 259 * can free up its vfs_request slots. 260 */ 261 struct orangefs_kiocb_s { 262 /* the pointer to the task that initiated the AIO */ 263 struct task_struct *tsk; 264 265 /* pointer to the kiocb that kicked this operation */ 266 struct kiocb *kiocb; 267 268 /* buffer index that was used for the I/O */ 269 struct orangefs_bufmap *bufmap; 270 int buffer_index; 271 272 /* orangefs kernel operation type */ 273 struct orangefs_kernel_op_s *op; 274 275 /* The user space buffers from/to which I/O is being staged */ 276 struct iovec *iov; 277 278 /* number of elements in the iovector */ 279 unsigned long nr_segs; 280 281 /* set to indicate the type of the operation */ 282 int rw; 283 284 /* file offset */ 285 loff_t offset; 286 287 /* and the count in bytes */ 288 size_t bytes_to_be_copied; 289 290 ssize_t bytes_copied; 291 int needs_cleanup; 292 }; 293 294 struct orangefs_stats { 295 unsigned long cache_hits; 296 unsigned long cache_misses; 297 unsigned long reads; 298 unsigned long writes; 299 }; 300 301 extern struct orangefs_stats orangefs_stats; 302 303 /* 304 * NOTE: See Documentation/filesystems/porting for information 305 * on implementing FOO_I and properly accessing fs private data 306 */ 307 static inline struct orangefs_inode_s *ORANGEFS_I(struct inode *inode) 308 { 309 return container_of(inode, struct orangefs_inode_s, vfs_inode); 310 } 311 312 static inline struct orangefs_sb_info_s *ORANGEFS_SB(struct super_block *sb) 313 { 314 return (struct orangefs_sb_info_s *) sb->s_fs_info; 315 } 316 317 /* ino_t descends from "unsigned long", 8 bytes, 64 bits. */ 318 static inline ino_t orangefs_khandle_to_ino(struct orangefs_khandle *khandle) 319 { 320 union { 321 unsigned char u[8]; 322 __u64 ino; 323 } ihandle; 324 325 ihandle.u[0] = khandle->u[0] ^ khandle->u[4]; 326 ihandle.u[1] = khandle->u[1] ^ khandle->u[5]; 327 ihandle.u[2] = khandle->u[2] ^ khandle->u[6]; 328 ihandle.u[3] = khandle->u[3] ^ khandle->u[7]; 329 ihandle.u[4] = khandle->u[12] ^ khandle->u[8]; 330 ihandle.u[5] = khandle->u[13] ^ khandle->u[9]; 331 ihandle.u[6] = khandle->u[14] ^ khandle->u[10]; 332 ihandle.u[7] = khandle->u[15] ^ khandle->u[11]; 333 334 return ihandle.ino; 335 } 336 337 static inline struct orangefs_khandle *get_khandle_from_ino(struct inode *inode) 338 { 339 return &(ORANGEFS_I(inode)->refn.khandle); 340 } 341 342 static inline __s32 get_fsid_from_ino(struct inode *inode) 343 { 344 return ORANGEFS_I(inode)->refn.fs_id; 345 } 346 347 static inline ino_t get_ino_from_khandle(struct inode *inode) 348 { 349 struct orangefs_khandle *khandle; 350 ino_t ino; 351 352 khandle = get_khandle_from_ino(inode); 353 ino = orangefs_khandle_to_ino(khandle); 354 return ino; 355 } 356 357 static inline ino_t get_parent_ino_from_dentry(struct dentry *dentry) 358 { 359 return get_ino_from_khandle(dentry->d_parent->d_inode); 360 } 361 362 static inline int is_root_handle(struct inode *inode) 363 { 364 gossip_debug(GOSSIP_DCACHE_DEBUG, 365 "%s: root handle: %pU, this handle: %pU:\n", 366 __func__, 367 &ORANGEFS_SB(inode->i_sb)->root_khandle, 368 get_khandle_from_ino(inode)); 369 370 if (ORANGEFS_khandle_cmp(&(ORANGEFS_SB(inode->i_sb)->root_khandle), 371 get_khandle_from_ino(inode))) 372 return 0; 373 else 374 return 1; 375 } 376 377 static inline int match_handle(struct orangefs_khandle resp_handle, 378 struct inode *inode) 379 { 380 gossip_debug(GOSSIP_DCACHE_DEBUG, 381 "%s: one handle: %pU, another handle:%pU:\n", 382 __func__, 383 &resp_handle, 384 get_khandle_from_ino(inode)); 385 386 if (ORANGEFS_khandle_cmp(&resp_handle, get_khandle_from_ino(inode))) 387 return 0; 388 else 389 return 1; 390 } 391 392 /* 393 * defined in orangefs-cache.c 394 */ 395 int op_cache_initialize(void); 396 int op_cache_finalize(void); 397 struct orangefs_kernel_op_s *op_alloc(__s32 type); 398 void orangefs_new_tag(struct orangefs_kernel_op_s *op); 399 char *get_opname_string(struct orangefs_kernel_op_s *new_op); 400 401 int orangefs_inode_cache_initialize(void); 402 int orangefs_inode_cache_finalize(void); 403 404 /* 405 * defined in orangefs-mod.c 406 */ 407 void purge_inprogress_ops(void); 408 409 /* 410 * defined in waitqueue.c 411 */ 412 void purge_waiting_ops(void); 413 414 /* 415 * defined in super.c 416 */ 417 extern uint64_t orangefs_features; 418 419 struct dentry *orangefs_mount(struct file_system_type *fst, 420 int flags, 421 const char *devname, 422 void *data); 423 424 void orangefs_kill_sb(struct super_block *sb); 425 int orangefs_remount(struct orangefs_sb_info_s *); 426 427 int fsid_key_table_initialize(void); 428 void fsid_key_table_finalize(void); 429 430 /* 431 * defined in inode.c 432 */ 433 __u32 convert_to_orangefs_mask(unsigned long lite_mask); 434 struct inode *orangefs_new_inode(struct super_block *sb, 435 struct inode *dir, 436 int mode, 437 dev_t dev, 438 struct orangefs_object_kref *ref); 439 440 int orangefs_setattr(struct dentry *dentry, struct iattr *iattr); 441 442 int orangefs_getattr(const struct path *path, struct kstat *stat, 443 u32 request_mask, unsigned int flags); 444 445 int orangefs_permission(struct inode *inode, int mask); 446 447 /* 448 * defined in xattr.c 449 */ 450 int orangefs_setxattr(struct dentry *dentry, 451 const char *name, 452 const void *value, 453 size_t size, 454 int flags); 455 456 ssize_t orangefs_getxattr(struct dentry *dentry, 457 const char *name, 458 void *buffer, 459 size_t size); 460 461 ssize_t orangefs_listxattr(struct dentry *dentry, char *buffer, size_t size); 462 463 /* 464 * defined in namei.c 465 */ 466 struct inode *orangefs_iget(struct super_block *sb, 467 struct orangefs_object_kref *ref); 468 469 ssize_t orangefs_inode_read(struct inode *inode, 470 struct iov_iter *iter, 471 loff_t *offset, 472 loff_t readahead_size); 473 474 /* 475 * defined in devorangefs-req.c 476 */ 477 extern uint32_t orangefs_userspace_version; 478 479 int orangefs_dev_init(void); 480 void orangefs_dev_cleanup(void); 481 int is_daemon_in_service(void); 482 bool __is_daemon_in_service(void); 483 484 /* 485 * defined in orangefs-utils.c 486 */ 487 __s32 fsid_of_op(struct orangefs_kernel_op_s *op); 488 489 int orangefs_flush_inode(struct inode *inode); 490 491 ssize_t orangefs_inode_getxattr(struct inode *inode, 492 const char *name, 493 void *buffer, 494 size_t size); 495 496 int orangefs_inode_setxattr(struct inode *inode, 497 const char *name, 498 const void *value, 499 size_t size, 500 int flags); 501 502 int orangefs_inode_getattr(struct inode *inode, int new, int bypass); 503 504 int orangefs_inode_check_changed(struct inode *inode); 505 506 int orangefs_inode_setattr(struct inode *inode, struct iattr *iattr); 507 508 void orangefs_make_bad_inode(struct inode *inode); 509 510 int orangefs_unmount_sb(struct super_block *sb); 511 512 bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op); 513 514 int orangefs_normalize_to_errno(__s32 error_code); 515 516 extern struct mutex orangefs_request_mutex; 517 extern int op_timeout_secs; 518 extern int slot_timeout_secs; 519 extern int orangefs_dcache_timeout_msecs; 520 extern int orangefs_getattr_timeout_msecs; 521 extern struct list_head orangefs_superblocks; 522 extern spinlock_t orangefs_superblocks_lock; 523 extern struct list_head orangefs_request_list; 524 extern spinlock_t orangefs_request_list_lock; 525 extern wait_queue_head_t orangefs_request_list_waitq; 526 extern struct list_head *orangefs_htable_ops_in_progress; 527 extern spinlock_t orangefs_htable_ops_in_progress_lock; 528 extern int hash_table_size; 529 530 extern const struct address_space_operations orangefs_address_operations; 531 extern const struct inode_operations orangefs_file_inode_operations; 532 extern const struct file_operations orangefs_file_operations; 533 extern const struct inode_operations orangefs_symlink_inode_operations; 534 extern const struct inode_operations orangefs_dir_inode_operations; 535 extern const struct file_operations orangefs_dir_operations; 536 extern const struct dentry_operations orangefs_dentry_operations; 537 extern const struct file_operations orangefs_devreq_file_operations; 538 539 extern wait_queue_head_t orangefs_bufmap_init_waitq; 540 541 /* 542 * misc convenience macros 543 */ 544 545 #define ORANGEFS_OP_INTERRUPTIBLE 1 /* service_operation() is interruptible */ 546 #define ORANGEFS_OP_PRIORITY 2 /* service_operation() is high priority */ 547 #define ORANGEFS_OP_CANCELLATION 4 /* this is a cancellation */ 548 #define ORANGEFS_OP_NO_MUTEX 8 /* don't acquire request_mutex */ 549 #define ORANGEFS_OP_ASYNC 16 /* Queue it, but don't wait */ 550 551 int service_operation(struct orangefs_kernel_op_s *op, 552 const char *op_name, 553 int flags); 554 555 #define get_interruptible_flag(inode) \ 556 ((ORANGEFS_SB(inode->i_sb)->flags & ORANGEFS_OPT_INTR) ? \ 557 ORANGEFS_OP_INTERRUPTIBLE : 0) 558 559 #define fill_default_sys_attrs(sys_attr, type, mode) \ 560 do { \ 561 sys_attr.owner = from_kuid(&init_user_ns, current_fsuid()); \ 562 sys_attr.group = from_kgid(&init_user_ns, current_fsgid()); \ 563 sys_attr.perms = ORANGEFS_util_translate_mode(mode); \ 564 sys_attr.mtime = 0; \ 565 sys_attr.atime = 0; \ 566 sys_attr.ctime = 0; \ 567 sys_attr.mask = ORANGEFS_ATTR_SYS_ALL_SETABLE; \ 568 } while (0) 569 570 static inline void orangefs_i_size_write(struct inode *inode, loff_t i_size) 571 { 572 #if BITS_PER_LONG == 32 && defined(CONFIG_SMP) 573 inode_lock(inode); 574 #endif 575 i_size_write(inode, i_size); 576 #if BITS_PER_LONG == 32 && defined(CONFIG_SMP) 577 inode_unlock(inode); 578 #endif 579 } 580 581 static inline void orangefs_set_timeout(struct dentry *dentry) 582 { 583 unsigned long time = jiffies + orangefs_dcache_timeout_msecs*HZ/1000; 584 585 dentry->d_fsdata = (void *) time; 586 } 587 588 #endif /* __ORANGEFSKERNEL_H */ 589