1 /* SPDX-License-Identifier: GPL-2.0 */
3 * (C) 2001 Clemson University and The University of Chicago
5 * See COPYING in top-level directory.
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.
16 * Declarations and macros for the ORANGEFS Linux kernel support.
19 #ifndef __ORANGEFSKERNEL_H
20 #define __ORANGEFSKERNEL_H
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>
35 #include <linux/vmalloc.h>
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>
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>
55 #include <asm/unaligned.h>
57 #include "orangefs-dev-proto.h"
59 #ifdef ORANGEFS_KERNEL_DEBUG
60 #define ORANGEFS_DEFAULT_OP_TIMEOUT_SECS 10
62 #define ORANGEFS_DEFAULT_OP_TIMEOUT_SECS 20
65 #define ORANGEFS_BUFMAP_WAIT_TIMEOUT_SECS 30
67 #define ORANGEFS_DEFAULT_SLOT_TIMEOUT_SECS 900 /* 15 minutes */
69 #define ORANGEFS_REQDEVICE_NAME "pvfs2-req"
71 #define ORANGEFS_DEVREQ_MAGIC 0x20030529
72 #define ORANGEFS_LINK_MAX 0x000000FF
73 #define ORANGEFS_PURGE_RETRY_COUNT 0x00000005
74 #define ORANGEFS_MAX_NUM_OPTIONS 0x00000004
75 #define ORANGEFS_MAX_MOUNT_OPT_LEN 0x00000080
76 #define ORANGEFS_MAX_FSKEY_LEN 64
78 #define MAX_DEV_REQ_UPSIZE (2 * sizeof(__s32) + \
79 sizeof(__u64) + sizeof(struct orangefs_upcall_s))
80 #define MAX_DEV_REQ_DOWNSIZE (2 * sizeof(__s32) + \
81 sizeof(__u64) + sizeof(struct orangefs_downcall_s))
84 * valid orangefs kernel operation states
86 * unknown - op was just initialized
87 * waiting - op is on request_list (upward bound)
88 * inprogr - op is in progress (waiting for downcall)
89 * serviced - op has matching downcall; ok
90 * purged - op has to start a timer since client-core
91 * exited uncleanly before servicing op
92 * given up - submitter has given up waiting for it
94 enum orangefs_vfs_op_states
{
95 OP_VFS_STATE_UNKNOWN
= 0,
96 OP_VFS_STATE_WAITING
= 1,
97 OP_VFS_STATE_INPROGR
= 2,
98 OP_VFS_STATE_SERVICED
= 4,
99 OP_VFS_STATE_PURGED
= 8,
100 OP_VFS_STATE_GIVEN_UP
= 16,
104 * orangefs kernel memory related flags
107 #if ((defined ORANGEFS_KERNEL_DEBUG) && (defined CONFIG_DEBUG_SLAB))
108 #define ORANGEFS_CACHE_CREATE_FLAGS SLAB_RED_ZONE
110 #define ORANGEFS_CACHE_CREATE_FLAGS 0
111 #endif /* ((defined ORANGEFS_KERNEL_DEBUG) && (defined CONFIG_DEBUG_SLAB)) */
113 extern int orangefs_init_acl(struct inode
*inode
, struct inode
*dir
);
114 extern const struct xattr_handler
*orangefs_xattr_handlers
[];
116 extern struct posix_acl
*orangefs_get_acl(struct inode
*inode
, int type
);
117 extern int orangefs_set_acl(struct inode
*inode
, struct posix_acl
*acl
, int type
);
120 * Redefine xtvec structure so that we could move helper functions out of
124 __kernel_off_t xtv_off
; /* must be off_t */
125 __kernel_size_t xtv_len
; /* must be size_t */
129 * orangefs data structures
131 struct orangefs_kernel_op_s
{
132 enum orangefs_vfs_op_states op_state
;
136 * Set uses_shared_memory to non zero if this operation uses
137 * shared memory. If true, then a retry on the op must also
138 * get a new shared memory buffer and re-populate it.
139 * Cancels don't care - it only matters for service_operation()
140 * retry logics and cancels don't go through it anymore. It
141 * safely stays non-zero when we use it as slot_to_free.
144 int uses_shared_memory
;
148 struct orangefs_upcall_s upcall
;
149 struct orangefs_downcall_s downcall
;
151 struct completion waitq
;
156 struct list_head list
;
159 #define set_op_state_waiting(op) ((op)->op_state = OP_VFS_STATE_WAITING)
160 #define set_op_state_inprogress(op) ((op)->op_state = OP_VFS_STATE_INPROGR)
161 #define set_op_state_given_up(op) ((op)->op_state = OP_VFS_STATE_GIVEN_UP)
162 static inline void set_op_state_serviced(struct orangefs_kernel_op_s
*op
)
164 op
->op_state
= OP_VFS_STATE_SERVICED
;
165 complete(&op
->waitq
);
168 #define op_state_waiting(op) ((op)->op_state & OP_VFS_STATE_WAITING)
169 #define op_state_in_progress(op) ((op)->op_state & OP_VFS_STATE_INPROGR)
170 #define op_state_serviced(op) ((op)->op_state & OP_VFS_STATE_SERVICED)
171 #define op_state_purged(op) ((op)->op_state & OP_VFS_STATE_PURGED)
172 #define op_state_given_up(op) ((op)->op_state & OP_VFS_STATE_GIVEN_UP)
173 #define op_is_cancel(op) ((op)->upcall.type == ORANGEFS_VFS_OP_CANCEL)
175 void op_release(struct orangefs_kernel_op_s
*op
);
177 extern void orangefs_bufmap_put(int);
178 static inline void put_cancel(struct orangefs_kernel_op_s
*op
)
180 orangefs_bufmap_put(op
->slot_to_free
);
184 static inline void set_op_state_purged(struct orangefs_kernel_op_s
*op
)
186 spin_lock(&op
->lock
);
187 if (unlikely(op_is_cancel(op
))) {
188 list_del_init(&op
->list
);
189 spin_unlock(&op
->lock
);
192 op
->op_state
|= OP_VFS_STATE_PURGED
;
193 complete(&op
->waitq
);
194 spin_unlock(&op
->lock
);
198 /* per inode private orangefs info */
199 struct orangefs_inode_s
{
200 struct orangefs_object_kref refn
;
201 char link_target
[ORANGEFS_NAME_MAX
];
204 * Reading/Writing Extended attributes need to acquire the appropriate
205 * reader/writer semaphore on the orangefs_inode_s structure.
207 struct rw_semaphore xattr_sem
;
209 struct inode vfs_inode
;
210 sector_t last_failed_block_index_read
;
213 * State of in-memory attributes not yet flushed to disk associated
216 unsigned long pinode_flags
;
218 unsigned long getattr_time
;
222 #define P_ATIME_FLAG 0
223 #define P_MTIME_FLAG 1
224 #define P_CTIME_FLAG 2
225 #define P_MODE_FLAG 3
227 #define ClearAtimeFlag(pinode) clear_bit(P_ATIME_FLAG, &(pinode)->pinode_flags)
228 #define SetAtimeFlag(pinode) set_bit(P_ATIME_FLAG, &(pinode)->pinode_flags)
229 #define AtimeFlag(pinode) test_bit(P_ATIME_FLAG, &(pinode)->pinode_flags)
231 #define ClearMtimeFlag(pinode) clear_bit(P_MTIME_FLAG, &(pinode)->pinode_flags)
232 #define SetMtimeFlag(pinode) set_bit(P_MTIME_FLAG, &(pinode)->pinode_flags)
233 #define MtimeFlag(pinode) test_bit(P_MTIME_FLAG, &(pinode)->pinode_flags)
235 #define ClearCtimeFlag(pinode) clear_bit(P_CTIME_FLAG, &(pinode)->pinode_flags)
236 #define SetCtimeFlag(pinode) set_bit(P_CTIME_FLAG, &(pinode)->pinode_flags)
237 #define CtimeFlag(pinode) test_bit(P_CTIME_FLAG, &(pinode)->pinode_flags)
239 #define ClearModeFlag(pinode) clear_bit(P_MODE_FLAG, &(pinode)->pinode_flags)
240 #define SetModeFlag(pinode) set_bit(P_MODE_FLAG, &(pinode)->pinode_flags)
241 #define ModeFlag(pinode) test_bit(P_MODE_FLAG, &(pinode)->pinode_flags)
243 /* per superblock private orangefs info */
244 struct orangefs_sb_info_s
{
245 struct orangefs_khandle root_khandle
;
249 #define ORANGEFS_OPT_INTR 0x01
250 #define ORANGEFS_OPT_LOCAL_LOCK 0x02
251 char devname
[ORANGEFS_MAX_SERVER_ADDR_LEN
];
252 struct super_block
*sb
;
255 struct list_head list
;
259 * structure that holds the state of any async I/O operation issued
260 * through the VFS. Needed especially to handle cancellation requests
261 * or even completion notification so that the VFS client-side daemon
262 * can free up its vfs_request slots.
264 struct orangefs_kiocb_s
{
265 /* the pointer to the task that initiated the AIO */
266 struct task_struct
*tsk
;
268 /* pointer to the kiocb that kicked this operation */
271 /* buffer index that was used for the I/O */
272 struct orangefs_bufmap
*bufmap
;
275 /* orangefs kernel operation type */
276 struct orangefs_kernel_op_s
*op
;
278 /* The user space buffers from/to which I/O is being staged */
281 /* number of elements in the iovector */
282 unsigned long nr_segs
;
284 /* set to indicate the type of the operation */
290 /* and the count in bytes */
291 size_t bytes_to_be_copied
;
293 ssize_t bytes_copied
;
297 struct orangefs_stats
{
298 unsigned long cache_hits
;
299 unsigned long cache_misses
;
301 unsigned long writes
;
304 extern struct orangefs_stats orangefs_stats
;
307 * NOTE: See Documentation/filesystems/porting for information
308 * on implementing FOO_I and properly accessing fs private data
310 static inline struct orangefs_inode_s
*ORANGEFS_I(struct inode
*inode
)
312 return container_of(inode
, struct orangefs_inode_s
, vfs_inode
);
315 static inline struct orangefs_sb_info_s
*ORANGEFS_SB(struct super_block
*sb
)
317 return (struct orangefs_sb_info_s
*) sb
->s_fs_info
;
320 /* ino_t descends from "unsigned long", 8 bytes, 64 bits. */
321 static inline ino_t
orangefs_khandle_to_ino(struct orangefs_khandle
*khandle
)
328 ihandle
.u
[0] = khandle
->u
[0] ^ khandle
->u
[4];
329 ihandle
.u
[1] = khandle
->u
[1] ^ khandle
->u
[5];
330 ihandle
.u
[2] = khandle
->u
[2] ^ khandle
->u
[6];
331 ihandle
.u
[3] = khandle
->u
[3] ^ khandle
->u
[7];
332 ihandle
.u
[4] = khandle
->u
[12] ^ khandle
->u
[8];
333 ihandle
.u
[5] = khandle
->u
[13] ^ khandle
->u
[9];
334 ihandle
.u
[6] = khandle
->u
[14] ^ khandle
->u
[10];
335 ihandle
.u
[7] = khandle
->u
[15] ^ khandle
->u
[11];
340 static inline struct orangefs_khandle
*get_khandle_from_ino(struct inode
*inode
)
342 return &(ORANGEFS_I(inode
)->refn
.khandle
);
345 static inline ino_t
get_ino_from_khandle(struct inode
*inode
)
347 struct orangefs_khandle
*khandle
;
350 khandle
= get_khandle_from_ino(inode
);
351 ino
= orangefs_khandle_to_ino(khandle
);
355 static inline ino_t
get_parent_ino_from_dentry(struct dentry
*dentry
)
357 return get_ino_from_khandle(dentry
->d_parent
->d_inode
);
360 static inline int is_root_handle(struct inode
*inode
)
362 gossip_debug(GOSSIP_DCACHE_DEBUG
,
363 "%s: root handle: %pU, this handle: %pU:\n",
365 &ORANGEFS_SB(inode
->i_sb
)->root_khandle
,
366 get_khandle_from_ino(inode
));
368 if (ORANGEFS_khandle_cmp(&(ORANGEFS_SB(inode
->i_sb
)->root_khandle
),
369 get_khandle_from_ino(inode
)))
375 static inline int match_handle(struct orangefs_khandle resp_handle
,
378 gossip_debug(GOSSIP_DCACHE_DEBUG
,
379 "%s: one handle: %pU, another handle:%pU:\n",
382 get_khandle_from_ino(inode
));
384 if (ORANGEFS_khandle_cmp(&resp_handle
, get_khandle_from_ino(inode
)))
391 * defined in orangefs-cache.c
393 int op_cache_initialize(void);
394 int op_cache_finalize(void);
395 struct orangefs_kernel_op_s
*op_alloc(__s32 type
);
396 void orangefs_new_tag(struct orangefs_kernel_op_s
*op
);
397 char *get_opname_string(struct orangefs_kernel_op_s
*new_op
);
399 int orangefs_inode_cache_initialize(void);
400 int orangefs_inode_cache_finalize(void);
403 * defined in orangefs-mod.c
405 void purge_inprogress_ops(void);
408 * defined in waitqueue.c
410 void purge_waiting_ops(void);
415 extern uint64_t orangefs_features
;
417 struct dentry
*orangefs_mount(struct file_system_type
*fst
,
422 void orangefs_kill_sb(struct super_block
*sb
);
423 int orangefs_remount(struct orangefs_sb_info_s
*);
425 int fsid_key_table_initialize(void);
426 void fsid_key_table_finalize(void);
431 __u32
convert_to_orangefs_mask(unsigned long lite_mask
);
432 struct inode
*orangefs_new_inode(struct super_block
*sb
,
436 struct orangefs_object_kref
*ref
);
438 int orangefs_setattr(struct dentry
*dentry
, struct iattr
*iattr
);
440 int orangefs_getattr(const struct path
*path
, struct kstat
*stat
,
441 u32 request_mask
, unsigned int flags
);
443 int orangefs_permission(struct inode
*inode
, int mask
);
448 int orangefs_setxattr(struct dentry
*dentry
,
454 ssize_t
orangefs_getxattr(struct dentry
*dentry
,
459 ssize_t
orangefs_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
);
464 struct inode
*orangefs_iget(struct super_block
*sb
,
465 struct orangefs_object_kref
*ref
);
467 ssize_t
orangefs_inode_read(struct inode
*inode
,
468 struct iov_iter
*iter
,
470 loff_t readahead_size
);
473 * defined in devorangefs-req.c
475 extern uint32_t orangefs_userspace_version
;
477 int orangefs_dev_init(void);
478 void orangefs_dev_cleanup(void);
479 int is_daemon_in_service(void);
480 bool __is_daemon_in_service(void);
483 * defined in orangefs-utils.c
485 __s32
fsid_of_op(struct orangefs_kernel_op_s
*op
);
487 int orangefs_flush_inode(struct inode
*inode
);
489 ssize_t
orangefs_inode_getxattr(struct inode
*inode
,
494 int orangefs_inode_setxattr(struct inode
*inode
,
500 int orangefs_inode_getattr(struct inode
*inode
, int new, int bypass
,
503 int orangefs_inode_check_changed(struct inode
*inode
);
505 int orangefs_inode_setattr(struct inode
*inode
, struct iattr
*iattr
);
507 void orangefs_make_bad_inode(struct inode
*inode
);
509 int orangefs_unmount_sb(struct super_block
*sb
);
511 bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s
*op
);
513 int orangefs_normalize_to_errno(__s32 error_code
);
515 extern struct mutex orangefs_request_mutex
;
516 extern int op_timeout_secs
;
517 extern int slot_timeout_secs
;
518 extern int orangefs_dcache_timeout_msecs
;
519 extern int orangefs_getattr_timeout_msecs
;
520 extern struct list_head orangefs_superblocks
;
521 extern spinlock_t orangefs_superblocks_lock
;
522 extern struct list_head orangefs_request_list
;
523 extern spinlock_t orangefs_request_list_lock
;
524 extern wait_queue_head_t orangefs_request_list_waitq
;
525 extern struct list_head
*orangefs_htable_ops_in_progress
;
526 extern spinlock_t orangefs_htable_ops_in_progress_lock
;
527 extern int hash_table_size
;
529 extern const struct address_space_operations orangefs_address_operations
;
530 extern const struct inode_operations orangefs_file_inode_operations
;
531 extern const struct file_operations orangefs_file_operations
;
532 extern const struct inode_operations orangefs_symlink_inode_operations
;
533 extern const struct inode_operations orangefs_dir_inode_operations
;
534 extern const struct file_operations orangefs_dir_operations
;
535 extern const struct dentry_operations orangefs_dentry_operations
;
536 extern const struct file_operations orangefs_devreq_file_operations
;
538 extern wait_queue_head_t orangefs_bufmap_init_waitq
;
541 * misc convenience macros
544 #define ORANGEFS_OP_INTERRUPTIBLE 1 /* service_operation() is interruptible */
545 #define ORANGEFS_OP_PRIORITY 2 /* service_operation() is high priority */
546 #define ORANGEFS_OP_CANCELLATION 4 /* this is a cancellation */
547 #define ORANGEFS_OP_NO_MUTEX 8 /* don't acquire request_mutex */
548 #define ORANGEFS_OP_ASYNC 16 /* Queue it, but don't wait */
550 int service_operation(struct orangefs_kernel_op_s
*op
,
554 #define get_interruptible_flag(inode) \
555 ((ORANGEFS_SB(inode->i_sb)->flags & ORANGEFS_OPT_INTR) ? \
556 ORANGEFS_OP_INTERRUPTIBLE : 0)
558 #define fill_default_sys_attrs(sys_attr, type, mode) \
560 sys_attr.owner = from_kuid(&init_user_ns, current_fsuid()); \
561 sys_attr.group = from_kgid(&init_user_ns, current_fsgid()); \
562 sys_attr.perms = ORANGEFS_util_translate_mode(mode); \
563 sys_attr.mtime = 0; \
564 sys_attr.atime = 0; \
565 sys_attr.ctime = 0; \
566 sys_attr.mask = ORANGEFS_ATTR_SYS_ALL_SETABLE; \
569 static inline void orangefs_i_size_write(struct inode
*inode
, loff_t i_size
)
571 #if BITS_PER_LONG == 32 && defined(CONFIG_SMP)
574 i_size_write(inode
, i_size
);
575 #if BITS_PER_LONG == 32 && defined(CONFIG_SMP)
580 static inline void orangefs_set_timeout(struct dentry
*dentry
)
582 unsigned long time
= jiffies
+ orangefs_dcache_timeout_msecs
*HZ
/1000;
584 dentry
->d_fsdata
= (void *) time
;
587 #endif /* __ORANGEFSKERNEL_H */