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 u32 getattr_mask; 219 }; 220 221 #define P_ATIME_FLAG 0 222 #define P_MTIME_FLAG 1 223 #define P_CTIME_FLAG 2 224 #define P_MODE_FLAG 3 225 226 #define ClearAtimeFlag(pinode) clear_bit(P_ATIME_FLAG, &(pinode)->pinode_flags) 227 #define SetAtimeFlag(pinode) set_bit(P_ATIME_FLAG, &(pinode)->pinode_flags) 228 #define AtimeFlag(pinode) test_bit(P_ATIME_FLAG, &(pinode)->pinode_flags) 229 230 #define ClearMtimeFlag(pinode) clear_bit(P_MTIME_FLAG, &(pinode)->pinode_flags) 231 #define SetMtimeFlag(pinode) set_bit(P_MTIME_FLAG, &(pinode)->pinode_flags) 232 #define MtimeFlag(pinode) test_bit(P_MTIME_FLAG, &(pinode)->pinode_flags) 233 234 #define ClearCtimeFlag(pinode) clear_bit(P_CTIME_FLAG, &(pinode)->pinode_flags) 235 #define SetCtimeFlag(pinode) set_bit(P_CTIME_FLAG, &(pinode)->pinode_flags) 236 #define CtimeFlag(pinode) test_bit(P_CTIME_FLAG, &(pinode)->pinode_flags) 237 238 #define ClearModeFlag(pinode) clear_bit(P_MODE_FLAG, &(pinode)->pinode_flags) 239 #define SetModeFlag(pinode) set_bit(P_MODE_FLAG, &(pinode)->pinode_flags) 240 #define ModeFlag(pinode) test_bit(P_MODE_FLAG, &(pinode)->pinode_flags) 241 242 /* per superblock private orangefs info */ 243 struct orangefs_sb_info_s { 244 struct orangefs_khandle root_khandle; 245 __s32 fs_id; 246 int id; 247 int flags; 248 #define ORANGEFS_OPT_INTR 0x01 249 #define ORANGEFS_OPT_LOCAL_LOCK 0x02 250 char devname[ORANGEFS_MAX_SERVER_ADDR_LEN]; 251 struct super_block *sb; 252 int mount_pending; 253 int no_list; 254 struct list_head list; 255 }; 256 257 /* 258 * structure that holds the state of any async I/O operation issued 259 * through the VFS. Needed especially to handle cancellation requests 260 * or even completion notification so that the VFS client-side daemon 261 * can free up its vfs_request slots. 262 */ 263 struct orangefs_kiocb_s { 264 /* the pointer to the task that initiated the AIO */ 265 struct task_struct *tsk; 266 267 /* pointer to the kiocb that kicked this operation */ 268 struct kiocb *kiocb; 269 270 /* buffer index that was used for the I/O */ 271 struct orangefs_bufmap *bufmap; 272 int buffer_index; 273 274 /* orangefs kernel operation type */ 275 struct orangefs_kernel_op_s *op; 276 277 /* The user space buffers from/to which I/O is being staged */ 278 struct iovec *iov; 279 280 /* number of elements in the iovector */ 281 unsigned long nr_segs; 282 283 /* set to indicate the type of the operation */ 284 int rw; 285 286 /* file offset */ 287 loff_t offset; 288 289 /* and the count in bytes */ 290 size_t bytes_to_be_copied; 291 292 ssize_t bytes_copied; 293 int needs_cleanup; 294 }; 295 296 struct orangefs_stats { 297 unsigned long cache_hits; 298 unsigned long cache_misses; 299 unsigned long reads; 300 unsigned long writes; 301 }; 302 303 extern struct orangefs_stats orangefs_stats; 304 305 /* 306 * NOTE: See Documentation/filesystems/porting for information 307 * on implementing FOO_I and properly accessing fs private data 308 */ 309 static inline struct orangefs_inode_s *ORANGEFS_I(struct inode *inode) 310 { 311 return container_of(inode, struct orangefs_inode_s, vfs_inode); 312 } 313 314 static inline struct orangefs_sb_info_s *ORANGEFS_SB(struct super_block *sb) 315 { 316 return (struct orangefs_sb_info_s *) sb->s_fs_info; 317 } 318 319 /* ino_t descends from "unsigned long", 8 bytes, 64 bits. */ 320 static inline ino_t orangefs_khandle_to_ino(struct orangefs_khandle *khandle) 321 { 322 union { 323 unsigned char u[8]; 324 __u64 ino; 325 } ihandle; 326 327 ihandle.u[0] = khandle->u[0] ^ khandle->u[4]; 328 ihandle.u[1] = khandle->u[1] ^ khandle->u[5]; 329 ihandle.u[2] = khandle->u[2] ^ khandle->u[6]; 330 ihandle.u[3] = khandle->u[3] ^ khandle->u[7]; 331 ihandle.u[4] = khandle->u[12] ^ khandle->u[8]; 332 ihandle.u[5] = khandle->u[13] ^ khandle->u[9]; 333 ihandle.u[6] = khandle->u[14] ^ khandle->u[10]; 334 ihandle.u[7] = khandle->u[15] ^ khandle->u[11]; 335 336 return ihandle.ino; 337 } 338 339 static inline struct orangefs_khandle *get_khandle_from_ino(struct inode *inode) 340 { 341 return &(ORANGEFS_I(inode)->refn.khandle); 342 } 343 344 static inline ino_t get_ino_from_khandle(struct inode *inode) 345 { 346 struct orangefs_khandle *khandle; 347 ino_t ino; 348 349 khandle = get_khandle_from_ino(inode); 350 ino = orangefs_khandle_to_ino(khandle); 351 return ino; 352 } 353 354 static inline ino_t get_parent_ino_from_dentry(struct dentry *dentry) 355 { 356 return get_ino_from_khandle(dentry->d_parent->d_inode); 357 } 358 359 static inline int is_root_handle(struct inode *inode) 360 { 361 gossip_debug(GOSSIP_DCACHE_DEBUG, 362 "%s: root handle: %pU, this handle: %pU:\n", 363 __func__, 364 &ORANGEFS_SB(inode->i_sb)->root_khandle, 365 get_khandle_from_ino(inode)); 366 367 if (ORANGEFS_khandle_cmp(&(ORANGEFS_SB(inode->i_sb)->root_khandle), 368 get_khandle_from_ino(inode))) 369 return 0; 370 else 371 return 1; 372 } 373 374 static inline int match_handle(struct orangefs_khandle resp_handle, 375 struct inode *inode) 376 { 377 gossip_debug(GOSSIP_DCACHE_DEBUG, 378 "%s: one handle: %pU, another handle:%pU:\n", 379 __func__, 380 &resp_handle, 381 get_khandle_from_ino(inode)); 382 383 if (ORANGEFS_khandle_cmp(&resp_handle, get_khandle_from_ino(inode))) 384 return 0; 385 else 386 return 1; 387 } 388 389 /* 390 * defined in orangefs-cache.c 391 */ 392 int op_cache_initialize(void); 393 int op_cache_finalize(void); 394 struct orangefs_kernel_op_s *op_alloc(__s32 type); 395 void orangefs_new_tag(struct orangefs_kernel_op_s *op); 396 char *get_opname_string(struct orangefs_kernel_op_s *new_op); 397 398 int orangefs_inode_cache_initialize(void); 399 int orangefs_inode_cache_finalize(void); 400 401 /* 402 * defined in orangefs-mod.c 403 */ 404 void purge_inprogress_ops(void); 405 406 /* 407 * defined in waitqueue.c 408 */ 409 void purge_waiting_ops(void); 410 411 /* 412 * defined in super.c 413 */ 414 extern uint64_t orangefs_features; 415 416 struct dentry *orangefs_mount(struct file_system_type *fst, 417 int flags, 418 const char *devname, 419 void *data); 420 421 void orangefs_kill_sb(struct super_block *sb); 422 int orangefs_remount(struct orangefs_sb_info_s *); 423 424 int fsid_key_table_initialize(void); 425 void fsid_key_table_finalize(void); 426 427 /* 428 * defined in inode.c 429 */ 430 __u32 convert_to_orangefs_mask(unsigned long lite_mask); 431 struct inode *orangefs_new_inode(struct super_block *sb, 432 struct inode *dir, 433 int mode, 434 dev_t dev, 435 struct orangefs_object_kref *ref); 436 437 int orangefs_setattr(struct dentry *dentry, struct iattr *iattr); 438 439 int orangefs_getattr(const struct path *path, struct kstat *stat, 440 u32 request_mask, unsigned int flags); 441 442 int orangefs_permission(struct inode *inode, int mask); 443 444 /* 445 * defined in xattr.c 446 */ 447 int orangefs_setxattr(struct dentry *dentry, 448 const char *name, 449 const void *value, 450 size_t size, 451 int flags); 452 453 ssize_t orangefs_getxattr(struct dentry *dentry, 454 const char *name, 455 void *buffer, 456 size_t size); 457 458 ssize_t orangefs_listxattr(struct dentry *dentry, char *buffer, size_t size); 459 460 /* 461 * defined in namei.c 462 */ 463 struct inode *orangefs_iget(struct super_block *sb, 464 struct orangefs_object_kref *ref); 465 466 ssize_t orangefs_inode_read(struct inode *inode, 467 struct iov_iter *iter, 468 loff_t *offset, 469 loff_t readahead_size); 470 471 /* 472 * defined in devorangefs-req.c 473 */ 474 extern uint32_t orangefs_userspace_version; 475 476 int orangefs_dev_init(void); 477 void orangefs_dev_cleanup(void); 478 int is_daemon_in_service(void); 479 bool __is_daemon_in_service(void); 480 481 /* 482 * defined in orangefs-utils.c 483 */ 484 __s32 fsid_of_op(struct orangefs_kernel_op_s *op); 485 486 int orangefs_flush_inode(struct inode *inode); 487 488 ssize_t orangefs_inode_getxattr(struct inode *inode, 489 const char *name, 490 void *buffer, 491 size_t size); 492 493 int orangefs_inode_setxattr(struct inode *inode, 494 const char *name, 495 const void *value, 496 size_t size, 497 int flags); 498 499 int orangefs_inode_getattr(struct inode *inode, int new, int bypass, 500 u32 request_mask); 501 502 int orangefs_inode_check_changed(struct inode *inode); 503 504 int orangefs_inode_setattr(struct inode *inode, struct iattr *iattr); 505 506 void orangefs_make_bad_inode(struct inode *inode); 507 508 int orangefs_unmount_sb(struct super_block *sb); 509 510 bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op); 511 512 int orangefs_normalize_to_errno(__s32 error_code); 513 514 extern struct mutex orangefs_request_mutex; 515 extern int op_timeout_secs; 516 extern int slot_timeout_secs; 517 extern int orangefs_dcache_timeout_msecs; 518 extern int orangefs_getattr_timeout_msecs; 519 extern struct list_head orangefs_superblocks; 520 extern spinlock_t orangefs_superblocks_lock; 521 extern struct list_head orangefs_request_list; 522 extern spinlock_t orangefs_request_list_lock; 523 extern wait_queue_head_t orangefs_request_list_waitq; 524 extern struct list_head *orangefs_htable_ops_in_progress; 525 extern spinlock_t orangefs_htable_ops_in_progress_lock; 526 extern int hash_table_size; 527 528 extern const struct address_space_operations orangefs_address_operations; 529 extern const struct inode_operations orangefs_file_inode_operations; 530 extern const struct file_operations orangefs_file_operations; 531 extern const struct inode_operations orangefs_symlink_inode_operations; 532 extern const struct inode_operations orangefs_dir_inode_operations; 533 extern const struct file_operations orangefs_dir_operations; 534 extern const struct dentry_operations orangefs_dentry_operations; 535 extern const struct file_operations orangefs_devreq_file_operations; 536 537 extern wait_queue_head_t orangefs_bufmap_init_waitq; 538 539 /* 540 * misc convenience macros 541 */ 542 543 #define ORANGEFS_OP_INTERRUPTIBLE 1 /* service_operation() is interruptible */ 544 #define ORANGEFS_OP_PRIORITY 2 /* service_operation() is high priority */ 545 #define ORANGEFS_OP_CANCELLATION 4 /* this is a cancellation */ 546 #define ORANGEFS_OP_NO_MUTEX 8 /* don't acquire request_mutex */ 547 #define ORANGEFS_OP_ASYNC 16 /* Queue it, but don't wait */ 548 549 int service_operation(struct orangefs_kernel_op_s *op, 550 const char *op_name, 551 int flags); 552 553 #define get_interruptible_flag(inode) \ 554 ((ORANGEFS_SB(inode->i_sb)->flags & ORANGEFS_OPT_INTR) ? \ 555 ORANGEFS_OP_INTERRUPTIBLE : 0) 556 557 #define fill_default_sys_attrs(sys_attr, type, mode) \ 558 do { \ 559 sys_attr.owner = from_kuid(&init_user_ns, current_fsuid()); \ 560 sys_attr.group = from_kgid(&init_user_ns, current_fsgid()); \ 561 sys_attr.perms = ORANGEFS_util_translate_mode(mode); \ 562 sys_attr.mtime = 0; \ 563 sys_attr.atime = 0; \ 564 sys_attr.ctime = 0; \ 565 sys_attr.mask = ORANGEFS_ATTR_SYS_ALL_SETABLE; \ 566 } while (0) 567 568 static inline void orangefs_i_size_write(struct inode *inode, loff_t i_size) 569 { 570 #if BITS_PER_LONG == 32 && defined(CONFIG_SMP) 571 inode_lock(inode); 572 #endif 573 i_size_write(inode, i_size); 574 #if BITS_PER_LONG == 32 && defined(CONFIG_SMP) 575 inode_unlock(inode); 576 #endif 577 } 578 579 static inline void orangefs_set_timeout(struct dentry *dentry) 580 { 581 unsigned long time = jiffies + orangefs_dcache_timeout_msecs*HZ/1000; 582 583 dentry->d_fsdata = (void *) time; 584 } 585 586 #endif /* __ORANGEFSKERNEL_H */ 587