xref: /openbmc/linux/fs/orangefs/orangefs-kernel.h (revision e3b9f1e8)
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 
55 #include <asm/unaligned.h>
56 
57 #include "orangefs-dev-proto.h"
58 
59 #define ORANGEFS_DEFAULT_OP_TIMEOUT_SECS       20
60 
61 #define ORANGEFS_BUFMAP_WAIT_TIMEOUT_SECS   30
62 
63 #define ORANGEFS_DEFAULT_SLOT_TIMEOUT_SECS     900	/* 15 minutes */
64 
65 #define ORANGEFS_REQDEVICE_NAME          "pvfs2-req"
66 
67 #define ORANGEFS_DEVREQ_MAGIC             0x20030529
68 #define ORANGEFS_LINK_MAX                 0x000000FF
69 #define ORANGEFS_PURGE_RETRY_COUNT     0x00000005
70 #define ORANGEFS_MAX_NUM_OPTIONS          0x00000004
71 #define ORANGEFS_MAX_MOUNT_OPT_LEN        0x00000080
72 #define ORANGEFS_MAX_FSKEY_LEN            64
73 
74 #define MAX_DEV_REQ_UPSIZE (2 * sizeof(__s32) +   \
75 sizeof(__u64) + sizeof(struct orangefs_upcall_s))
76 #define MAX_DEV_REQ_DOWNSIZE (2 * sizeof(__s32) + \
77 sizeof(__u64) + sizeof(struct orangefs_downcall_s))
78 
79 /*
80  * valid orangefs kernel operation states
81  *
82  * unknown  - op was just initialized
83  * waiting  - op is on request_list (upward bound)
84  * inprogr  - op is in progress (waiting for downcall)
85  * serviced - op has matching downcall; ok
86  * purged   - op has to start a timer since client-core
87  *            exited uncleanly before servicing op
88  * given up - submitter has given up waiting for it
89  */
90 enum orangefs_vfs_op_states {
91 	OP_VFS_STATE_UNKNOWN = 0,
92 	OP_VFS_STATE_WAITING = 1,
93 	OP_VFS_STATE_INPROGR = 2,
94 	OP_VFS_STATE_SERVICED = 4,
95 	OP_VFS_STATE_PURGED = 8,
96 	OP_VFS_STATE_GIVEN_UP = 16,
97 };
98 
99 /*
100  * orangefs kernel memory related flags
101  */
102 
103 #if (defined CONFIG_DEBUG_SLAB)
104 #define ORANGEFS_CACHE_CREATE_FLAGS SLAB_RED_ZONE
105 #else
106 #define ORANGEFS_CACHE_CREATE_FLAGS 0
107 #endif
108 
109 extern int orangefs_init_acl(struct inode *inode, struct inode *dir);
110 extern const struct xattr_handler *orangefs_xattr_handlers[];
111 
112 extern struct posix_acl *orangefs_get_acl(struct inode *inode, int type);
113 extern int orangefs_set_acl(struct inode *inode, struct posix_acl *acl, int type);
114 
115 /*
116  * Redefine xtvec structure so that we could move helper functions out of
117  * the define
118  */
119 struct xtvec {
120 	__kernel_off_t xtv_off;		/* must be off_t */
121 	__kernel_size_t xtv_len;	/* must be size_t */
122 };
123 
124 /*
125  * orangefs data structures
126  */
127 struct orangefs_kernel_op_s {
128 	enum orangefs_vfs_op_states op_state;
129 	__u64 tag;
130 
131 	/*
132 	 * Set uses_shared_memory to non zero if this operation uses
133 	 * shared memory. If true, then a retry on the op must also
134 	 * get a new shared memory buffer and re-populate it.
135 	 * Cancels don't care - it only matters for service_operation()
136 	 * retry logics and cancels don't go through it anymore. It
137 	 * safely stays non-zero when we use it as slot_to_free.
138 	 */
139 	union {
140 		int uses_shared_memory;
141 		int slot_to_free;
142 	};
143 
144 	struct orangefs_upcall_s upcall;
145 	struct orangefs_downcall_s downcall;
146 
147 	struct completion waitq;
148 	spinlock_t lock;
149 
150 	int attempts;
151 
152 	struct list_head list;
153 };
154 
155 #define set_op_state_waiting(op)     ((op)->op_state = OP_VFS_STATE_WAITING)
156 #define set_op_state_inprogress(op)  ((op)->op_state = OP_VFS_STATE_INPROGR)
157 #define set_op_state_given_up(op)  ((op)->op_state = OP_VFS_STATE_GIVEN_UP)
158 static inline void set_op_state_serviced(struct orangefs_kernel_op_s *op)
159 {
160 	op->op_state = OP_VFS_STATE_SERVICED;
161 	complete(&op->waitq);
162 }
163 
164 #define op_state_waiting(op)     ((op)->op_state & OP_VFS_STATE_WAITING)
165 #define op_state_in_progress(op) ((op)->op_state & OP_VFS_STATE_INPROGR)
166 #define op_state_serviced(op)    ((op)->op_state & OP_VFS_STATE_SERVICED)
167 #define op_state_purged(op)      ((op)->op_state & OP_VFS_STATE_PURGED)
168 #define op_state_given_up(op)    ((op)->op_state & OP_VFS_STATE_GIVEN_UP)
169 #define op_is_cancel(op)         ((op)->upcall.type == ORANGEFS_VFS_OP_CANCEL)
170 
171 void op_release(struct orangefs_kernel_op_s *op);
172 
173 extern void orangefs_bufmap_put(int);
174 static inline void put_cancel(struct orangefs_kernel_op_s *op)
175 {
176 	orangefs_bufmap_put(op->slot_to_free);
177 	op_release(op);
178 }
179 
180 static inline void set_op_state_purged(struct orangefs_kernel_op_s *op)
181 {
182 	spin_lock(&op->lock);
183 	if (unlikely(op_is_cancel(op))) {
184 		list_del_init(&op->list);
185 		spin_unlock(&op->lock);
186 		put_cancel(op);
187 	} else {
188 		op->op_state |= OP_VFS_STATE_PURGED;
189 		complete(&op->waitq);
190 		spin_unlock(&op->lock);
191 	}
192 }
193 
194 /* per inode private orangefs info */
195 struct orangefs_inode_s {
196 	struct orangefs_object_kref refn;
197 	char link_target[ORANGEFS_NAME_MAX];
198 	__s64 blksize;
199 	/*
200 	 * Reading/Writing Extended attributes need to acquire the appropriate
201 	 * reader/writer semaphore on the orangefs_inode_s structure.
202 	 */
203 	struct rw_semaphore xattr_sem;
204 
205 	struct inode vfs_inode;
206 	sector_t last_failed_block_index_read;
207 
208 	unsigned long getattr_time;
209 	u32 getattr_mask;
210 };
211 
212 /* per superblock private orangefs info */
213 struct orangefs_sb_info_s {
214 	struct orangefs_khandle root_khandle;
215 	__s32 fs_id;
216 	int id;
217 	int flags;
218 #define ORANGEFS_OPT_INTR	0x01
219 #define ORANGEFS_OPT_LOCAL_LOCK	0x02
220 	char devname[ORANGEFS_MAX_SERVER_ADDR_LEN];
221 	struct super_block *sb;
222 	int mount_pending;
223 	int no_list;
224 	struct list_head list;
225 };
226 
227 /*
228  * structure that holds the state of any async I/O operation issued
229  * through the VFS. Needed especially to handle cancellation requests
230  * or even completion notification so that the VFS client-side daemon
231  * can free up its vfs_request slots.
232  */
233 struct orangefs_kiocb_s {
234 	/* the pointer to the task that initiated the AIO */
235 	struct task_struct *tsk;
236 
237 	/* pointer to the kiocb that kicked this operation */
238 	struct kiocb *kiocb;
239 
240 	/* buffer index that was used for the I/O */
241 	struct orangefs_bufmap *bufmap;
242 	int buffer_index;
243 
244 	/* orangefs kernel operation type */
245 	struct orangefs_kernel_op_s *op;
246 
247 	/* set to indicate the type of the operation */
248 	int rw;
249 
250 	/* file offset */
251 	loff_t offset;
252 
253 	/* and the count in bytes */
254 	size_t bytes_to_be_copied;
255 
256 	ssize_t bytes_copied;
257 	int needs_cleanup;
258 };
259 
260 struct orangefs_stats {
261 	unsigned long cache_hits;
262 	unsigned long cache_misses;
263 	unsigned long reads;
264 	unsigned long writes;
265 };
266 
267 extern struct orangefs_stats orangefs_stats;
268 
269 /*
270  * NOTE: See Documentation/filesystems/porting for information
271  * on implementing FOO_I and properly accessing fs private data
272  */
273 static inline struct orangefs_inode_s *ORANGEFS_I(struct inode *inode)
274 {
275 	return container_of(inode, struct orangefs_inode_s, vfs_inode);
276 }
277 
278 static inline struct orangefs_sb_info_s *ORANGEFS_SB(struct super_block *sb)
279 {
280 	return (struct orangefs_sb_info_s *) sb->s_fs_info;
281 }
282 
283 /* ino_t descends from "unsigned long", 8 bytes, 64 bits. */
284 static inline ino_t orangefs_khandle_to_ino(struct orangefs_khandle *khandle)
285 {
286 	union {
287 		unsigned char u[8];
288 		__u64 ino;
289 	} ihandle;
290 
291 	ihandle.u[0] = khandle->u[0] ^ khandle->u[4];
292 	ihandle.u[1] = khandle->u[1] ^ khandle->u[5];
293 	ihandle.u[2] = khandle->u[2] ^ khandle->u[6];
294 	ihandle.u[3] = khandle->u[3] ^ khandle->u[7];
295 	ihandle.u[4] = khandle->u[12] ^ khandle->u[8];
296 	ihandle.u[5] = khandle->u[13] ^ khandle->u[9];
297 	ihandle.u[6] = khandle->u[14] ^ khandle->u[10];
298 	ihandle.u[7] = khandle->u[15] ^ khandle->u[11];
299 
300 	return ihandle.ino;
301 }
302 
303 static inline struct orangefs_khandle *get_khandle_from_ino(struct inode *inode)
304 {
305 	return &(ORANGEFS_I(inode)->refn.khandle);
306 }
307 
308 static inline ino_t get_ino_from_khandle(struct inode *inode)
309 {
310 	struct orangefs_khandle *khandle;
311 	ino_t ino;
312 
313 	khandle = get_khandle_from_ino(inode);
314 	ino = orangefs_khandle_to_ino(khandle);
315 	return ino;
316 }
317 
318 static inline ino_t get_parent_ino_from_dentry(struct dentry *dentry)
319 {
320 	return get_ino_from_khandle(dentry->d_parent->d_inode);
321 }
322 
323 static inline int is_root_handle(struct inode *inode)
324 {
325 	gossip_debug(GOSSIP_DCACHE_DEBUG,
326 		     "%s: root handle: %pU, this handle: %pU:\n",
327 		     __func__,
328 		     &ORANGEFS_SB(inode->i_sb)->root_khandle,
329 		     get_khandle_from_ino(inode));
330 
331 	if (ORANGEFS_khandle_cmp(&(ORANGEFS_SB(inode->i_sb)->root_khandle),
332 			     get_khandle_from_ino(inode)))
333 		return 0;
334 	else
335 		return 1;
336 }
337 
338 static inline int match_handle(struct orangefs_khandle resp_handle,
339 			       struct inode *inode)
340 {
341 	gossip_debug(GOSSIP_DCACHE_DEBUG,
342 		     "%s: one handle: %pU, another handle:%pU:\n",
343 		     __func__,
344 		     &resp_handle,
345 		     get_khandle_from_ino(inode));
346 
347 	if (ORANGEFS_khandle_cmp(&resp_handle, get_khandle_from_ino(inode)))
348 		return 0;
349 	else
350 		return 1;
351 }
352 
353 /*
354  * defined in orangefs-cache.c
355  */
356 int op_cache_initialize(void);
357 int op_cache_finalize(void);
358 struct orangefs_kernel_op_s *op_alloc(__s32 type);
359 void orangefs_new_tag(struct orangefs_kernel_op_s *op);
360 char *get_opname_string(struct orangefs_kernel_op_s *new_op);
361 
362 int orangefs_inode_cache_initialize(void);
363 int orangefs_inode_cache_finalize(void);
364 
365 /*
366  * defined in orangefs-mod.c
367  */
368 void purge_inprogress_ops(void);
369 
370 /*
371  * defined in waitqueue.c
372  */
373 void purge_waiting_ops(void);
374 
375 /*
376  * defined in super.c
377  */
378 extern uint64_t orangefs_features;
379 
380 struct dentry *orangefs_mount(struct file_system_type *fst,
381 			   int flags,
382 			   const char *devname,
383 			   void *data);
384 
385 void orangefs_kill_sb(struct super_block *sb);
386 int orangefs_remount(struct orangefs_sb_info_s *);
387 
388 int fsid_key_table_initialize(void);
389 void fsid_key_table_finalize(void);
390 
391 /*
392  * defined in inode.c
393  */
394 __u32 convert_to_orangefs_mask(unsigned long lite_mask);
395 struct inode *orangefs_new_inode(struct super_block *sb,
396 			      struct inode *dir,
397 			      int mode,
398 			      dev_t dev,
399 			      struct orangefs_object_kref *ref);
400 
401 int orangefs_setattr(struct dentry *dentry, struct iattr *iattr);
402 
403 int orangefs_getattr(const struct path *path, struct kstat *stat,
404 		     u32 request_mask, unsigned int flags);
405 
406 int orangefs_permission(struct inode *inode, int mask);
407 
408 int orangefs_update_time(struct inode *, struct timespec *, int);
409 
410 /*
411  * defined in xattr.c
412  */
413 int orangefs_setxattr(struct dentry *dentry,
414 		   const char *name,
415 		   const void *value,
416 		   size_t size,
417 		   int flags);
418 
419 ssize_t orangefs_getxattr(struct dentry *dentry,
420 		       const char *name,
421 		       void *buffer,
422 		       size_t size);
423 
424 ssize_t orangefs_listxattr(struct dentry *dentry, char *buffer, size_t size);
425 
426 /*
427  * defined in namei.c
428  */
429 struct inode *orangefs_iget(struct super_block *sb,
430 			 struct orangefs_object_kref *ref);
431 
432 ssize_t orangefs_inode_read(struct inode *inode,
433 			    struct iov_iter *iter,
434 			    loff_t *offset,
435 			    loff_t readahead_size);
436 
437 /*
438  * defined in devorangefs-req.c
439  */
440 extern uint32_t orangefs_userspace_version;
441 
442 int orangefs_dev_init(void);
443 void orangefs_dev_cleanup(void);
444 int is_daemon_in_service(void);
445 bool __is_daemon_in_service(void);
446 
447 /*
448  * defined in orangefs-utils.c
449  */
450 __s32 fsid_of_op(struct orangefs_kernel_op_s *op);
451 
452 ssize_t orangefs_inode_getxattr(struct inode *inode,
453 			     const char *name,
454 			     void *buffer,
455 			     size_t size);
456 
457 int orangefs_inode_setxattr(struct inode *inode,
458 			 const char *name,
459 			 const void *value,
460 			 size_t size,
461 			 int flags);
462 
463 int orangefs_inode_getattr(struct inode *inode, int new, int bypass,
464     u32 request_mask);
465 
466 int orangefs_inode_check_changed(struct inode *inode);
467 
468 int orangefs_inode_setattr(struct inode *inode, struct iattr *iattr);
469 
470 int orangefs_unmount_sb(struct super_block *sb);
471 
472 bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op);
473 
474 int orangefs_normalize_to_errno(__s32 error_code);
475 
476 extern struct mutex orangefs_request_mutex;
477 extern int op_timeout_secs;
478 extern int slot_timeout_secs;
479 extern int orangefs_dcache_timeout_msecs;
480 extern int orangefs_getattr_timeout_msecs;
481 extern struct list_head orangefs_superblocks;
482 extern spinlock_t orangefs_superblocks_lock;
483 extern struct list_head orangefs_request_list;
484 extern spinlock_t orangefs_request_list_lock;
485 extern wait_queue_head_t orangefs_request_list_waitq;
486 extern struct list_head *orangefs_htable_ops_in_progress;
487 extern spinlock_t orangefs_htable_ops_in_progress_lock;
488 extern int hash_table_size;
489 
490 extern const struct address_space_operations orangefs_address_operations;
491 extern const struct inode_operations orangefs_file_inode_operations;
492 extern const struct file_operations orangefs_file_operations;
493 extern const struct inode_operations orangefs_symlink_inode_operations;
494 extern const struct inode_operations orangefs_dir_inode_operations;
495 extern const struct file_operations orangefs_dir_operations;
496 extern const struct dentry_operations orangefs_dentry_operations;
497 extern const struct file_operations orangefs_devreq_file_operations;
498 
499 extern wait_queue_head_t orangefs_bufmap_init_waitq;
500 
501 /*
502  * misc convenience macros
503  */
504 
505 #define ORANGEFS_OP_INTERRUPTIBLE 1   /* service_operation() is interruptible */
506 #define ORANGEFS_OP_PRIORITY      2   /* service_operation() is high priority */
507 #define ORANGEFS_OP_CANCELLATION  4   /* this is a cancellation */
508 #define ORANGEFS_OP_NO_MUTEX      8   /* don't acquire request_mutex */
509 #define ORANGEFS_OP_ASYNC         16  /* Queue it, but don't wait */
510 
511 int service_operation(struct orangefs_kernel_op_s *op,
512 		      const char *op_name,
513 		      int flags);
514 
515 #define get_interruptible_flag(inode) \
516 	((ORANGEFS_SB(inode->i_sb)->flags & ORANGEFS_OPT_INTR) ? \
517 		ORANGEFS_OP_INTERRUPTIBLE : 0)
518 
519 #define fill_default_sys_attrs(sys_attr, type, mode)			\
520 do {									\
521 	sys_attr.owner = from_kuid(&init_user_ns, current_fsuid()); \
522 	sys_attr.group = from_kgid(&init_user_ns, current_fsgid()); \
523 	sys_attr.perms = ORANGEFS_util_translate_mode(mode);		\
524 	sys_attr.mtime = 0;						\
525 	sys_attr.atime = 0;						\
526 	sys_attr.ctime = 0;						\
527 	sys_attr.mask = ORANGEFS_ATTR_SYS_ALL_SETABLE;			\
528 } while (0)
529 
530 static inline void orangefs_set_timeout(struct dentry *dentry)
531 {
532 	unsigned long time = jiffies + orangefs_dcache_timeout_msecs*HZ/1000;
533 
534 	dentry->d_fsdata = (void *) time;
535 }
536 
537 #endif /* __ORANGEFSKERNEL_H */
538