| 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
| 2 | /* |
| 3 | * (C) 2001 Clemson University and The University of Chicago |
| 4 | * |
| 5 | * See COPYING in top-level directory. |
| 6 | */ |
| 7 | |
| 8 | /* |
| 9 | * The ORANGEFS Linux kernel support allows ORANGEFS volumes to be mounted and |
| 10 | * accessed through the Linux VFS (i.e. using standard I/O system calls). |
| 11 | * This support is only needed on clients that wish to mount the file system. |
| 12 | * |
| 13 | */ |
| 14 | |
| 15 | /* |
| 16 | * Declarations and macros for the ORANGEFS Linux kernel support. |
| 17 | */ |
| 18 | |
| 19 | #ifndef __ORANGEFSKERNEL_H |
| 20 | #define __ORANGEFSKERNEL_H |
| 21 | |
| 22 | #include <linux/kernel.h> |
| 23 | #include <linux/moduleparam.h> |
| 24 | #include <linux/statfs.h> |
| 25 | #include <linux/backing-dev.h> |
| 26 | #include <linux/device.h> |
| 27 | #include <linux/mpage.h> |
| 28 | #include <linux/namei.h> |
| 29 | #include <linux/errno.h> |
| 30 | #include <linux/init.h> |
| 31 | #include <linux/module.h> |
| 32 | #include <linux/slab.h> |
| 33 | #include <linux/types.h> |
| 34 | #include <linux/fs.h> |
| 35 | #include <linux/vmalloc.h> |
| 36 | |
| 37 | #include <linux/aio.h> |
| 38 | #include <linux/posix_acl.h> |
| 39 | #include <linux/posix_acl_xattr.h> |
| 40 | #include <linux/compat.h> |
| 41 | #include <linux/mount.h> |
| 42 | #include <linux/uaccess.h> |
| 43 | #include <linux/atomic.h> |
| 44 | #include <linux/uio.h> |
| 45 | #include <linux/sched/signal.h> |
| 46 | #include <linux/mm.h> |
| 47 | #include <linux/wait.h> |
| 48 | #include <linux/dcache.h> |
| 49 | #include <linux/pagemap.h> |
| 50 | #include <linux/poll.h> |
| 51 | #include <linux/rwsem.h> |
| 52 | #include <linux/xattr.h> |
| 53 | #include <linux/exportfs.h> |
| 54 | #include <linux/hashtable.h> |
| 55 | |
| 56 | #include <asm/unaligned.h> |
| 57 | |
| 58 | #include "orangefs-dev-proto.h" |
| 59 | |
| 60 | #define ORANGEFS_DEFAULT_OP_TIMEOUT_SECS 20 |
| 61 | |
| 62 | #define ORANGEFS_BUFMAP_WAIT_TIMEOUT_SECS 30 |
| 63 | |
| 64 | #define ORANGEFS_DEFAULT_SLOT_TIMEOUT_SECS 900 /* 15 minutes */ |
| 65 | |
| 66 | #define ORANGEFS_REQDEVICE_NAME "pvfs2-req" |
| 67 | |
| 68 | #define ORANGEFS_DEVREQ_MAGIC 0x20030529 |
| 69 | #define ORANGEFS_PURGE_RETRY_COUNT 0x00000005 |
| 70 | |
| 71 | #define MAX_DEV_REQ_UPSIZE (2 * sizeof(__s32) + \ |
| 72 | sizeof(__u64) + sizeof(struct orangefs_upcall_s)) |
| 73 | #define MAX_DEV_REQ_DOWNSIZE (2 * sizeof(__s32) + \ |
| 74 | sizeof(__u64) + sizeof(struct orangefs_downcall_s)) |
| 75 | |
| 76 | /* |
| 77 | * valid orangefs kernel operation states |
| 78 | * |
| 79 | * unknown - op was just initialized |
| 80 | * waiting - op is on request_list (upward bound) |
| 81 | * inprogr - op is in progress (waiting for downcall) |
| 82 | * serviced - op has matching downcall; ok |
| 83 | * purged - op has to start a timer since client-core |
| 84 | * exited uncleanly before servicing op |
| 85 | * given up - submitter has given up waiting for it |
| 86 | */ |
| 87 | enum orangefs_vfs_op_states { |
| 88 | OP_VFS_STATE_UNKNOWN = 0, |
| 89 | OP_VFS_STATE_WAITING = 1, |
| 90 | OP_VFS_STATE_INPROGR = 2, |
| 91 | OP_VFS_STATE_SERVICED = 4, |
| 92 | OP_VFS_STATE_PURGED = 8, |
| 93 | OP_VFS_STATE_GIVEN_UP = 16, |
| 94 | }; |
| 95 | |
| 96 | /* |
| 97 | * orangefs kernel memory related flags |
| 98 | */ |
| 99 | |
| 100 | #if (defined CONFIG_DEBUG_SLAB) |
| 101 | #define ORANGEFS_CACHE_CREATE_FLAGS SLAB_RED_ZONE |
| 102 | #else |
| 103 | #define ORANGEFS_CACHE_CREATE_FLAGS 0 |
| 104 | #endif |
| 105 | |
| 106 | extern const struct xattr_handler *orangefs_xattr_handlers[]; |
| 107 | |
| 108 | extern struct posix_acl *orangefs_get_acl(struct inode *inode, int type, bool rcu); |
| 109 | extern int orangefs_set_acl(struct mnt_idmap *idmap, |
| 110 | struct dentry *dentry, struct posix_acl *acl, |
| 111 | int type); |
| 112 | int __orangefs_set_acl(struct inode *inode, struct posix_acl *acl, int type); |
| 113 | |
| 114 | /* |
| 115 | * orangefs data structures |
| 116 | */ |
| 117 | struct orangefs_kernel_op_s { |
| 118 | enum orangefs_vfs_op_states op_state; |
| 119 | __u64 tag; |
| 120 | |
| 121 | /* |
| 122 | * Set uses_shared_memory to non zero if this operation uses |
| 123 | * shared memory. If true, then a retry on the op must also |
| 124 | * get a new shared memory buffer and re-populate it. |
| 125 | * Cancels don't care - it only matters for service_operation() |
| 126 | * retry logics and cancels don't go through it anymore. It |
| 127 | * safely stays non-zero when we use it as slot_to_free. |
| 128 | */ |
| 129 | union { |
| 130 | int uses_shared_memory; |
| 131 | int slot_to_free; |
| 132 | }; |
| 133 | |
| 134 | struct orangefs_upcall_s upcall; |
| 135 | struct orangefs_downcall_s downcall; |
| 136 | |
| 137 | struct completion waitq; |
| 138 | spinlock_t lock; |
| 139 | |
| 140 | int attempts; |
| 141 | |
| 142 | struct list_head list; |
| 143 | }; |
| 144 | |
| 145 | #define set_op_state_waiting(op) ((op)->op_state = OP_VFS_STATE_WAITING) |
| 146 | #define set_op_state_inprogress(op) ((op)->op_state = OP_VFS_STATE_INPROGR) |
| 147 | #define set_op_state_given_up(op) ((op)->op_state = OP_VFS_STATE_GIVEN_UP) |
| 148 | static inline void set_op_state_serviced(struct orangefs_kernel_op_s *op) |
| 149 | { |
| 150 | op->op_state = OP_VFS_STATE_SERVICED; |
| 151 | complete(&op->waitq); |
| 152 | } |
| 153 | |
| 154 | #define op_state_waiting(op) ((op)->op_state & OP_VFS_STATE_WAITING) |
| 155 | #define op_state_in_progress(op) ((op)->op_state & OP_VFS_STATE_INPROGR) |
| 156 | #define op_state_serviced(op) ((op)->op_state & OP_VFS_STATE_SERVICED) |
| 157 | #define op_state_purged(op) ((op)->op_state & OP_VFS_STATE_PURGED) |
| 158 | #define op_state_given_up(op) ((op)->op_state & OP_VFS_STATE_GIVEN_UP) |
| 159 | #define op_is_cancel(op) ((op)->upcall.type == ORANGEFS_VFS_OP_CANCEL) |
| 160 | |
| 161 | void op_release(struct orangefs_kernel_op_s *op); |
| 162 | |
| 163 | extern void orangefs_bufmap_put(int); |
| 164 | static inline void put_cancel(struct orangefs_kernel_op_s *op) |
| 165 | { |
| 166 | orangefs_bufmap_put(op->slot_to_free); |
| 167 | op_release(op); |
| 168 | } |
| 169 | |
| 170 | static inline void set_op_state_purged(struct orangefs_kernel_op_s *op) |
| 171 | { |
| 172 | spin_lock(&op->lock); |
| 173 | if (unlikely(op_is_cancel(op))) { |
| 174 | list_del_init(&op->list); |
| 175 | spin_unlock(&op->lock); |
| 176 | put_cancel(op); |
| 177 | } else { |
| 178 | op->op_state |= OP_VFS_STATE_PURGED; |
| 179 | complete(&op->waitq); |
| 180 | spin_unlock(&op->lock); |
| 181 | } |
| 182 | } |
| 183 | |
| 184 | /* per inode private orangefs info */ |
| 185 | struct orangefs_inode_s { |
| 186 | struct orangefs_object_kref refn; |
| 187 | char link_target[ORANGEFS_NAME_MAX]; |
| 188 | /* |
| 189 | * Reading/Writing Extended attributes need to acquire the appropriate |
| 190 | * reader/writer semaphore on the orangefs_inode_s structure. |
| 191 | */ |
| 192 | struct rw_semaphore xattr_sem; |
| 193 | |
| 194 | struct inode vfs_inode; |
| 195 | sector_t last_failed_block_index_read; |
| 196 | |
| 197 | unsigned long getattr_time; |
| 198 | unsigned long mapping_time; |
| 199 | int attr_valid; |
| 200 | kuid_t attr_uid; |
| 201 | kgid_t attr_gid; |
| 202 | unsigned long bitlock; |
| 203 | |
| 204 | DECLARE_HASHTABLE(xattr_cache, 4); |
| 205 | }; |
| 206 | |
| 207 | /* per superblock private orangefs info */ |
| 208 | struct orangefs_sb_info_s { |
| 209 | struct orangefs_khandle root_khandle; |
| 210 | __s32 fs_id; |
| 211 | int id; |
| 212 | int flags; |
| 213 | #define ORANGEFS_OPT_INTR 0x01 |
| 214 | #define ORANGEFS_OPT_LOCAL_LOCK 0x02 |
| 215 | char devname[ORANGEFS_MAX_SERVER_ADDR_LEN]; |
| 216 | struct super_block *sb; |
| 217 | int mount_pending; |
| 218 | int no_list; |
| 219 | struct list_head list; |
| 220 | }; |
| 221 | |
| 222 | struct orangefs_stats { |
| 223 | unsigned long cache_hits; |
| 224 | unsigned long cache_misses; |
| 225 | unsigned long reads; |
| 226 | unsigned long writes; |
| 227 | }; |
| 228 | |
| 229 | struct orangefs_cached_xattr { |
| 230 | struct hlist_node node; |
| 231 | char key[ORANGEFS_MAX_XATTR_NAMELEN]; |
| 232 | char val[ORANGEFS_MAX_XATTR_VALUELEN]; |
| 233 | ssize_t length; |
| 234 | unsigned long timeout; |
| 235 | }; |
| 236 | |
| 237 | struct orangefs_write_range { |
| 238 | loff_t pos; |
| 239 | size_t len; |
| 240 | kuid_t uid; |
| 241 | kgid_t gid; |
| 242 | }; |
| 243 | |
| 244 | extern struct orangefs_stats orangefs_stats; |
| 245 | |
| 246 | /* |
| 247 | * NOTE: See Documentation/filesystems/porting.rst for information |
| 248 | * on implementing FOO_I and properly accessing fs private data |
| 249 | */ |
| 250 | static inline struct orangefs_inode_s *ORANGEFS_I(struct inode *inode) |
| 251 | { |
| 252 | return container_of(inode, struct orangefs_inode_s, vfs_inode); |
| 253 | } |
| 254 | |
| 255 | static inline struct orangefs_sb_info_s *ORANGEFS_SB(struct super_block *sb) |
| 256 | { |
| 257 | return (struct orangefs_sb_info_s *) sb->s_fs_info; |
| 258 | } |
| 259 | |
| 260 | /* ino_t descends from "unsigned long", 8 bytes, 64 bits. */ |
| 261 | static inline ino_t orangefs_khandle_to_ino(struct orangefs_khandle *khandle) |
| 262 | { |
| 263 | union { |
| 264 | unsigned char u[8]; |
| 265 | __u64 ino; |
| 266 | } ihandle; |
| 267 | |
| 268 | ihandle.u[0] = khandle->u[0] ^ khandle->u[4]; |
| 269 | ihandle.u[1] = khandle->u[1] ^ khandle->u[5]; |
| 270 | ihandle.u[2] = khandle->u[2] ^ khandle->u[6]; |
| 271 | ihandle.u[3] = khandle->u[3] ^ khandle->u[7]; |
| 272 | ihandle.u[4] = khandle->u[12] ^ khandle->u[8]; |
| 273 | ihandle.u[5] = khandle->u[13] ^ khandle->u[9]; |
| 274 | ihandle.u[6] = khandle->u[14] ^ khandle->u[10]; |
| 275 | ihandle.u[7] = khandle->u[15] ^ khandle->u[11]; |
| 276 | |
| 277 | return ihandle.ino; |
| 278 | } |
| 279 | |
| 280 | static inline struct orangefs_khandle *get_khandle_from_ino(struct inode *inode) |
| 281 | { |
| 282 | return &(ORANGEFS_I(inode)->refn.khandle); |
| 283 | } |
| 284 | |
| 285 | static inline int is_root_handle(struct inode *inode) |
| 286 | { |
| 287 | gossip_debug(GOSSIP_DCACHE_DEBUG, |
| 288 | "%s: root handle: %pU, this handle: %pU:\n", |
| 289 | __func__, |
| 290 | &ORANGEFS_SB(inode->i_sb)->root_khandle, |
| 291 | get_khandle_from_ino(inode)); |
| 292 | |
| 293 | if (ORANGEFS_khandle_cmp(&(ORANGEFS_SB(inode->i_sb)->root_khandle), |
| 294 | get_khandle_from_ino(inode))) |
| 295 | return 0; |
| 296 | else |
| 297 | return 1; |
| 298 | } |
| 299 | |
| 300 | static inline int match_handle(struct orangefs_khandle resp_handle, |
| 301 | struct inode *inode) |
| 302 | { |
| 303 | gossip_debug(GOSSIP_DCACHE_DEBUG, |
| 304 | "%s: one handle: %pU, another handle:%pU:\n", |
| 305 | __func__, |
| 306 | &resp_handle, |
| 307 | get_khandle_from_ino(inode)); |
| 308 | |
| 309 | if (ORANGEFS_khandle_cmp(&resp_handle, get_khandle_from_ino(inode))) |
| 310 | return 0; |
| 311 | else |
| 312 | return 1; |
| 313 | } |
| 314 | |
| 315 | /* |
| 316 | * defined in orangefs-cache.c |
| 317 | */ |
| 318 | int op_cache_initialize(void); |
| 319 | int op_cache_finalize(void); |
| 320 | struct orangefs_kernel_op_s *op_alloc(__s32 type); |
| 321 | void orangefs_new_tag(struct orangefs_kernel_op_s *op); |
| 322 | char *get_opname_string(struct orangefs_kernel_op_s *new_op); |
| 323 | |
| 324 | int orangefs_inode_cache_initialize(void); |
| 325 | int orangefs_inode_cache_finalize(void); |
| 326 | |
| 327 | /* |
| 328 | * defined in orangefs-mod.c |
| 329 | */ |
| 330 | void purge_inprogress_ops(void); |
| 331 | |
| 332 | /* |
| 333 | * defined in waitqueue.c |
| 334 | */ |
| 335 | void purge_waiting_ops(void); |
| 336 | |
| 337 | /* |
| 338 | * defined in super.c |
| 339 | */ |
| 340 | extern uint64_t orangefs_features; |
| 341 | |
| 342 | struct dentry *orangefs_mount(struct file_system_type *fst, |
| 343 | int flags, |
| 344 | const char *devname, |
| 345 | void *data); |
| 346 | |
| 347 | void orangefs_kill_sb(struct super_block *sb); |
| 348 | int orangefs_remount(struct orangefs_sb_info_s *); |
| 349 | |
| 350 | int fsid_key_table_initialize(void); |
| 351 | void fsid_key_table_finalize(void); |
| 352 | |
| 353 | /* |
| 354 | * defined in inode.c |
| 355 | */ |
| 356 | vm_fault_t orangefs_page_mkwrite(struct vm_fault *); |
| 357 | struct inode *orangefs_new_inode(struct super_block *sb, |
| 358 | struct inode *dir, |
| 359 | umode_t mode, |
| 360 | dev_t dev, |
| 361 | struct orangefs_object_kref *ref); |
| 362 | |
| 363 | int __orangefs_setattr(struct inode *, struct iattr *); |
| 364 | int __orangefs_setattr_mode(struct dentry *dentry, struct iattr *iattr); |
| 365 | int orangefs_setattr(struct mnt_idmap *, struct dentry *, struct iattr *); |
| 366 | |
| 367 | int orangefs_getattr(struct mnt_idmap *idmap, const struct path *path, |
| 368 | struct kstat *stat, u32 request_mask, unsigned int flags); |
| 369 | |
| 370 | int orangefs_permission(struct mnt_idmap *idmap, |
| 371 | struct inode *inode, int mask); |
| 372 | |
| 373 | int orangefs_update_time(struct inode *, struct timespec64 *, int); |
| 374 | |
| 375 | /* |
| 376 | * defined in xattr.c |
| 377 | */ |
| 378 | ssize_t orangefs_listxattr(struct dentry *dentry, char *buffer, size_t size); |
| 379 | |
| 380 | /* |
| 381 | * defined in namei.c |
| 382 | */ |
| 383 | struct inode *orangefs_iget(struct super_block *sb, |
| 384 | struct orangefs_object_kref *ref); |
| 385 | |
| 386 | /* |
| 387 | * defined in devorangefs-req.c |
| 388 | */ |
| 389 | extern uint32_t orangefs_userspace_version; |
| 390 | |
| 391 | int orangefs_dev_init(void); |
| 392 | void orangefs_dev_cleanup(void); |
| 393 | int is_daemon_in_service(void); |
| 394 | bool __is_daemon_in_service(void); |
| 395 | |
| 396 | /* |
| 397 | * defined in file.c |
| 398 | */ |
| 399 | int orangefs_revalidate_mapping(struct inode *); |
| 400 | ssize_t wait_for_direct_io(enum ORANGEFS_io_type, struct inode *, loff_t *, |
| 401 | struct iov_iter *, size_t, loff_t, struct orangefs_write_range *, int *, |
| 402 | struct file *); |
| 403 | ssize_t do_readv_writev(enum ORANGEFS_io_type, struct file *, loff_t *, |
| 404 | struct iov_iter *); |
| 405 | |
| 406 | /* |
| 407 | * defined in orangefs-utils.c |
| 408 | */ |
| 409 | __s32 fsid_of_op(struct orangefs_kernel_op_s *op); |
| 410 | |
| 411 | ssize_t orangefs_inode_getxattr(struct inode *inode, |
| 412 | const char *name, |
| 413 | void *buffer, |
| 414 | size_t size); |
| 415 | |
| 416 | int orangefs_inode_setxattr(struct inode *inode, |
| 417 | const char *name, |
| 418 | const void *value, |
| 419 | size_t size, |
| 420 | int flags); |
| 421 | |
| 422 | #define ORANGEFS_GETATTR_NEW 1 |
| 423 | #define ORANGEFS_GETATTR_SIZE 2 |
| 424 | |
| 425 | int orangefs_inode_getattr(struct inode *, int); |
| 426 | |
| 427 | int orangefs_inode_check_changed(struct inode *inode); |
| 428 | |
| 429 | int orangefs_inode_setattr(struct inode *inode); |
| 430 | |
| 431 | bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op); |
| 432 | |
| 433 | int orangefs_normalize_to_errno(__s32 error_code); |
| 434 | |
| 435 | extern struct mutex orangefs_request_mutex; |
| 436 | extern int op_timeout_secs; |
| 437 | extern int slot_timeout_secs; |
| 438 | extern int orangefs_cache_timeout_msecs; |
| 439 | extern int orangefs_dcache_timeout_msecs; |
| 440 | extern int orangefs_getattr_timeout_msecs; |
| 441 | extern struct list_head orangefs_superblocks; |
| 442 | extern spinlock_t orangefs_superblocks_lock; |
| 443 | extern struct list_head orangefs_request_list; |
| 444 | extern spinlock_t orangefs_request_list_lock; |
| 445 | extern wait_queue_head_t orangefs_request_list_waitq; |
| 446 | extern struct list_head *orangefs_htable_ops_in_progress; |
| 447 | extern spinlock_t orangefs_htable_ops_in_progress_lock; |
| 448 | extern int hash_table_size; |
| 449 | |
| 450 | extern const struct file_operations orangefs_file_operations; |
| 451 | extern const struct inode_operations orangefs_symlink_inode_operations; |
| 452 | extern const struct inode_operations orangefs_dir_inode_operations; |
| 453 | extern const struct file_operations orangefs_dir_operations; |
| 454 | extern const struct dentry_operations orangefs_dentry_operations; |
| 455 | |
| 456 | /* |
| 457 | * misc convenience macros |
| 458 | */ |
| 459 | |
| 460 | #define ORANGEFS_OP_INTERRUPTIBLE 1 /* service_operation() is interruptible */ |
| 461 | #define ORANGEFS_OP_PRIORITY 2 /* service_operation() is high priority */ |
| 462 | #define ORANGEFS_OP_CANCELLATION 4 /* this is a cancellation */ |
| 463 | #define ORANGEFS_OP_NO_MUTEX 8 /* don't acquire request_mutex */ |
| 464 | #define ORANGEFS_OP_ASYNC 16 /* Queue it, but don't wait */ |
| 465 | #define ORANGEFS_OP_WRITEBACK 32 |
| 466 | |
| 467 | int service_operation(struct orangefs_kernel_op_s *op, |
| 468 | const char *op_name, |
| 469 | int flags); |
| 470 | |
| 471 | #define get_interruptible_flag(inode) \ |
| 472 | ((ORANGEFS_SB(inode->i_sb)->flags & ORANGEFS_OPT_INTR) ? \ |
| 473 | ORANGEFS_OP_INTERRUPTIBLE : 0) |
| 474 | |
| 475 | #define fill_default_sys_attrs(sys_attr, type, mode) \ |
| 476 | do { \ |
| 477 | sys_attr.owner = from_kuid(&init_user_ns, current_fsuid()); \ |
| 478 | sys_attr.group = from_kgid(&init_user_ns, current_fsgid()); \ |
| 479 | sys_attr.perms = ORANGEFS_util_translate_mode(mode); \ |
| 480 | sys_attr.mtime = 0; \ |
| 481 | sys_attr.atime = 0; \ |
| 482 | sys_attr.ctime = 0; \ |
| 483 | sys_attr.mask = ORANGEFS_ATTR_SYS_ALL_SETABLE; \ |
| 484 | } while (0) |
| 485 | |
| 486 | static inline void orangefs_set_timeout(struct dentry *dentry) |
| 487 | { |
| 488 | unsigned long time = jiffies + orangefs_dcache_timeout_msecs*HZ/1000; |
| 489 | |
| 490 | dentry->d_fsdata = (void *) time; |
| 491 | } |
| 492 | |
| 493 | #endif /* __ORANGEFSKERNEL_H */ |