xref: /openbmc/linux/fs/orangefs/orangefs-kernel.h (revision fb960bd2)
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 #ifdef ORANGEFS_KERNEL_DEBUG
60 #define ORANGEFS_DEFAULT_OP_TIMEOUT_SECS       10
61 #else
62 #define ORANGEFS_DEFAULT_OP_TIMEOUT_SECS       20
63 #endif
64 
65 #define ORANGEFS_BUFMAP_WAIT_TIMEOUT_SECS   30
66 
67 #define ORANGEFS_DEFAULT_SLOT_TIMEOUT_SECS     900	/* 15 minutes */
68 
69 #define ORANGEFS_REQDEVICE_NAME          "pvfs2-req"
70 
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
77 
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))
82 
83 /*
84  * valid orangefs kernel operation states
85  *
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
93  */
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,
101 };
102 
103 /*
104  * orangefs kernel memory related flags
105  */
106 
107 #if ((defined ORANGEFS_KERNEL_DEBUG) && (defined CONFIG_DEBUG_SLAB))
108 #define ORANGEFS_CACHE_CREATE_FLAGS SLAB_RED_ZONE
109 #else
110 #define ORANGEFS_CACHE_CREATE_FLAGS 0
111 #endif /* ((defined ORANGEFS_KERNEL_DEBUG) && (defined CONFIG_DEBUG_SLAB)) */
112 
113 extern int orangefs_init_acl(struct inode *inode, struct inode *dir);
114 extern const struct xattr_handler *orangefs_xattr_handlers[];
115 
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);
118 
119 /*
120  * Redefine xtvec structure so that we could move helper functions out of
121  * the define
122  */
123 struct xtvec {
124 	__kernel_off_t xtv_off;		/* must be off_t */
125 	__kernel_size_t xtv_len;	/* must be size_t */
126 };
127 
128 /*
129  * orangefs data structures
130  */
131 struct orangefs_kernel_op_s {
132 	enum orangefs_vfs_op_states op_state;
133 	__u64 tag;
134 
135 	/*
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.
142 	 */
143 	union {
144 		int uses_shared_memory;
145 		int slot_to_free;
146 	};
147 
148 	struct orangefs_upcall_s upcall;
149 	struct orangefs_downcall_s downcall;
150 
151 	struct completion waitq;
152 	spinlock_t lock;
153 
154 	int attempts;
155 
156 	struct list_head list;
157 };
158 
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)
163 {
164 	op->op_state = OP_VFS_STATE_SERVICED;
165 	complete(&op->waitq);
166 }
167 
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)
174 
175 void op_release(struct orangefs_kernel_op_s *op);
176 
177 extern void orangefs_bufmap_put(int);
178 static inline void put_cancel(struct orangefs_kernel_op_s *op)
179 {
180 	orangefs_bufmap_put(op->slot_to_free);
181 	op_release(op);
182 }
183 
184 static inline void set_op_state_purged(struct orangefs_kernel_op_s *op)
185 {
186 	spin_lock(&op->lock);
187 	if (unlikely(op_is_cancel(op))) {
188 		list_del_init(&op->list);
189 		spin_unlock(&op->lock);
190 		put_cancel(op);
191 	} else {
192 		op->op_state |= OP_VFS_STATE_PURGED;
193 		complete(&op->waitq);
194 		spin_unlock(&op->lock);
195 	}
196 }
197 
198 /* per inode private orangefs info */
199 struct orangefs_inode_s {
200 	struct orangefs_object_kref refn;
201 	char link_target[ORANGEFS_NAME_MAX];
202 	__s64 blksize;
203 	/*
204 	 * Reading/Writing Extended attributes need to acquire the appropriate
205 	 * reader/writer semaphore on the orangefs_inode_s structure.
206 	 */
207 	struct rw_semaphore xattr_sem;
208 
209 	struct inode vfs_inode;
210 	sector_t last_failed_block_index_read;
211 
212 	unsigned long getattr_time;
213 	u32 getattr_mask;
214 };
215 
216 /* per superblock private orangefs info */
217 struct orangefs_sb_info_s {
218 	struct orangefs_khandle root_khandle;
219 	__s32 fs_id;
220 	int id;
221 	int flags;
222 #define ORANGEFS_OPT_INTR	0x01
223 #define ORANGEFS_OPT_LOCAL_LOCK	0x02
224 	char devname[ORANGEFS_MAX_SERVER_ADDR_LEN];
225 	struct super_block *sb;
226 	int mount_pending;
227 	int no_list;
228 	struct list_head list;
229 };
230 
231 /*
232  * structure that holds the state of any async I/O operation issued
233  * through the VFS. Needed especially to handle cancellation requests
234  * or even completion notification so that the VFS client-side daemon
235  * can free up its vfs_request slots.
236  */
237 struct orangefs_kiocb_s {
238 	/* the pointer to the task that initiated the AIO */
239 	struct task_struct *tsk;
240 
241 	/* pointer to the kiocb that kicked this operation */
242 	struct kiocb *kiocb;
243 
244 	/* buffer index that was used for the I/O */
245 	struct orangefs_bufmap *bufmap;
246 	int buffer_index;
247 
248 	/* orangefs kernel operation type */
249 	struct orangefs_kernel_op_s *op;
250 
251 	/* set to indicate the type of the operation */
252 	int rw;
253 
254 	/* file offset */
255 	loff_t offset;
256 
257 	/* and the count in bytes */
258 	size_t bytes_to_be_copied;
259 
260 	ssize_t bytes_copied;
261 	int needs_cleanup;
262 };
263 
264 struct orangefs_stats {
265 	unsigned long cache_hits;
266 	unsigned long cache_misses;
267 	unsigned long reads;
268 	unsigned long writes;
269 };
270 
271 extern struct orangefs_stats orangefs_stats;
272 
273 /*
274  * NOTE: See Documentation/filesystems/porting for information
275  * on implementing FOO_I and properly accessing fs private data
276  */
277 static inline struct orangefs_inode_s *ORANGEFS_I(struct inode *inode)
278 {
279 	return container_of(inode, struct orangefs_inode_s, vfs_inode);
280 }
281 
282 static inline struct orangefs_sb_info_s *ORANGEFS_SB(struct super_block *sb)
283 {
284 	return (struct orangefs_sb_info_s *) sb->s_fs_info;
285 }
286 
287 /* ino_t descends from "unsigned long", 8 bytes, 64 bits. */
288 static inline ino_t orangefs_khandle_to_ino(struct orangefs_khandle *khandle)
289 {
290 	union {
291 		unsigned char u[8];
292 		__u64 ino;
293 	} ihandle;
294 
295 	ihandle.u[0] = khandle->u[0] ^ khandle->u[4];
296 	ihandle.u[1] = khandle->u[1] ^ khandle->u[5];
297 	ihandle.u[2] = khandle->u[2] ^ khandle->u[6];
298 	ihandle.u[3] = khandle->u[3] ^ khandle->u[7];
299 	ihandle.u[4] = khandle->u[12] ^ khandle->u[8];
300 	ihandle.u[5] = khandle->u[13] ^ khandle->u[9];
301 	ihandle.u[6] = khandle->u[14] ^ khandle->u[10];
302 	ihandle.u[7] = khandle->u[15] ^ khandle->u[11];
303 
304 	return ihandle.ino;
305 }
306 
307 static inline struct orangefs_khandle *get_khandle_from_ino(struct inode *inode)
308 {
309 	return &(ORANGEFS_I(inode)->refn.khandle);
310 }
311 
312 static inline ino_t get_ino_from_khandle(struct inode *inode)
313 {
314 	struct orangefs_khandle *khandle;
315 	ino_t ino;
316 
317 	khandle = get_khandle_from_ino(inode);
318 	ino = orangefs_khandle_to_ino(khandle);
319 	return ino;
320 }
321 
322 static inline ino_t get_parent_ino_from_dentry(struct dentry *dentry)
323 {
324 	return get_ino_from_khandle(dentry->d_parent->d_inode);
325 }
326 
327 static inline int is_root_handle(struct inode *inode)
328 {
329 	gossip_debug(GOSSIP_DCACHE_DEBUG,
330 		     "%s: root handle: %pU, this handle: %pU:\n",
331 		     __func__,
332 		     &ORANGEFS_SB(inode->i_sb)->root_khandle,
333 		     get_khandle_from_ino(inode));
334 
335 	if (ORANGEFS_khandle_cmp(&(ORANGEFS_SB(inode->i_sb)->root_khandle),
336 			     get_khandle_from_ino(inode)))
337 		return 0;
338 	else
339 		return 1;
340 }
341 
342 static inline int match_handle(struct orangefs_khandle resp_handle,
343 			       struct inode *inode)
344 {
345 	gossip_debug(GOSSIP_DCACHE_DEBUG,
346 		     "%s: one handle: %pU, another handle:%pU:\n",
347 		     __func__,
348 		     &resp_handle,
349 		     get_khandle_from_ino(inode));
350 
351 	if (ORANGEFS_khandle_cmp(&resp_handle, get_khandle_from_ino(inode)))
352 		return 0;
353 	else
354 		return 1;
355 }
356 
357 /*
358  * defined in orangefs-cache.c
359  */
360 int op_cache_initialize(void);
361 int op_cache_finalize(void);
362 struct orangefs_kernel_op_s *op_alloc(__s32 type);
363 void orangefs_new_tag(struct orangefs_kernel_op_s *op);
364 char *get_opname_string(struct orangefs_kernel_op_s *new_op);
365 
366 int orangefs_inode_cache_initialize(void);
367 int orangefs_inode_cache_finalize(void);
368 
369 /*
370  * defined in orangefs-mod.c
371  */
372 void purge_inprogress_ops(void);
373 
374 /*
375  * defined in waitqueue.c
376  */
377 void purge_waiting_ops(void);
378 
379 /*
380  * defined in super.c
381  */
382 extern uint64_t orangefs_features;
383 
384 struct dentry *orangefs_mount(struct file_system_type *fst,
385 			   int flags,
386 			   const char *devname,
387 			   void *data);
388 
389 void orangefs_kill_sb(struct super_block *sb);
390 int orangefs_remount(struct orangefs_sb_info_s *);
391 
392 int fsid_key_table_initialize(void);
393 void fsid_key_table_finalize(void);
394 
395 /*
396  * defined in inode.c
397  */
398 __u32 convert_to_orangefs_mask(unsigned long lite_mask);
399 struct inode *orangefs_new_inode(struct super_block *sb,
400 			      struct inode *dir,
401 			      int mode,
402 			      dev_t dev,
403 			      struct orangefs_object_kref *ref);
404 
405 int orangefs_setattr(struct dentry *dentry, struct iattr *iattr);
406 
407 int orangefs_getattr(const struct path *path, struct kstat *stat,
408 		     u32 request_mask, unsigned int flags);
409 
410 int orangefs_permission(struct inode *inode, int mask);
411 
412 int orangefs_update_time(struct inode *, struct timespec *, int);
413 
414 /*
415  * defined in xattr.c
416  */
417 int orangefs_setxattr(struct dentry *dentry,
418 		   const char *name,
419 		   const void *value,
420 		   size_t size,
421 		   int flags);
422 
423 ssize_t orangefs_getxattr(struct dentry *dentry,
424 		       const char *name,
425 		       void *buffer,
426 		       size_t size);
427 
428 ssize_t orangefs_listxattr(struct dentry *dentry, char *buffer, size_t size);
429 
430 /*
431  * defined in namei.c
432  */
433 struct inode *orangefs_iget(struct super_block *sb,
434 			 struct orangefs_object_kref *ref);
435 
436 ssize_t orangefs_inode_read(struct inode *inode,
437 			    struct iov_iter *iter,
438 			    loff_t *offset,
439 			    loff_t readahead_size);
440 
441 /*
442  * defined in devorangefs-req.c
443  */
444 extern uint32_t orangefs_userspace_version;
445 
446 int orangefs_dev_init(void);
447 void orangefs_dev_cleanup(void);
448 int is_daemon_in_service(void);
449 bool __is_daemon_in_service(void);
450 
451 /*
452  * defined in orangefs-utils.c
453  */
454 __s32 fsid_of_op(struct orangefs_kernel_op_s *op);
455 
456 ssize_t orangefs_inode_getxattr(struct inode *inode,
457 			     const char *name,
458 			     void *buffer,
459 			     size_t size);
460 
461 int orangefs_inode_setxattr(struct inode *inode,
462 			 const char *name,
463 			 const void *value,
464 			 size_t size,
465 			 int flags);
466 
467 int orangefs_inode_getattr(struct inode *inode, int new, int bypass,
468     u32 request_mask);
469 
470 int orangefs_inode_check_changed(struct inode *inode);
471 
472 int orangefs_inode_setattr(struct inode *inode, struct iattr *iattr);
473 
474 void orangefs_make_bad_inode(struct inode *inode);
475 
476 int orangefs_unmount_sb(struct super_block *sb);
477 
478 bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op);
479 
480 int orangefs_normalize_to_errno(__s32 error_code);
481 
482 extern struct mutex orangefs_request_mutex;
483 extern int op_timeout_secs;
484 extern int slot_timeout_secs;
485 extern int orangefs_dcache_timeout_msecs;
486 extern int orangefs_getattr_timeout_msecs;
487 extern struct list_head orangefs_superblocks;
488 extern spinlock_t orangefs_superblocks_lock;
489 extern struct list_head orangefs_request_list;
490 extern spinlock_t orangefs_request_list_lock;
491 extern wait_queue_head_t orangefs_request_list_waitq;
492 extern struct list_head *orangefs_htable_ops_in_progress;
493 extern spinlock_t orangefs_htable_ops_in_progress_lock;
494 extern int hash_table_size;
495 
496 extern const struct address_space_operations orangefs_address_operations;
497 extern const struct inode_operations orangefs_file_inode_operations;
498 extern const struct file_operations orangefs_file_operations;
499 extern const struct inode_operations orangefs_symlink_inode_operations;
500 extern const struct inode_operations orangefs_dir_inode_operations;
501 extern const struct file_operations orangefs_dir_operations;
502 extern const struct dentry_operations orangefs_dentry_operations;
503 extern const struct file_operations orangefs_devreq_file_operations;
504 
505 extern wait_queue_head_t orangefs_bufmap_init_waitq;
506 
507 /*
508  * misc convenience macros
509  */
510 
511 #define ORANGEFS_OP_INTERRUPTIBLE 1   /* service_operation() is interruptible */
512 #define ORANGEFS_OP_PRIORITY      2   /* service_operation() is high priority */
513 #define ORANGEFS_OP_CANCELLATION  4   /* this is a cancellation */
514 #define ORANGEFS_OP_NO_MUTEX      8   /* don't acquire request_mutex */
515 #define ORANGEFS_OP_ASYNC         16  /* Queue it, but don't wait */
516 
517 int service_operation(struct orangefs_kernel_op_s *op,
518 		      const char *op_name,
519 		      int flags);
520 
521 #define get_interruptible_flag(inode) \
522 	((ORANGEFS_SB(inode->i_sb)->flags & ORANGEFS_OPT_INTR) ? \
523 		ORANGEFS_OP_INTERRUPTIBLE : 0)
524 
525 #define fill_default_sys_attrs(sys_attr, type, mode)			\
526 do {									\
527 	sys_attr.owner = from_kuid(&init_user_ns, current_fsuid()); \
528 	sys_attr.group = from_kgid(&init_user_ns, current_fsgid()); \
529 	sys_attr.perms = ORANGEFS_util_translate_mode(mode);		\
530 	sys_attr.mtime = 0;						\
531 	sys_attr.atime = 0;						\
532 	sys_attr.ctime = 0;						\
533 	sys_attr.mask = ORANGEFS_ATTR_SYS_ALL_SETABLE;			\
534 } while (0)
535 
536 static inline void orangefs_i_size_write(struct inode *inode, loff_t i_size)
537 {
538 #if BITS_PER_LONG == 32 && defined(CONFIG_SMP)
539 	inode_lock(inode);
540 #endif
541 	i_size_write(inode, i_size);
542 #if BITS_PER_LONG == 32 && defined(CONFIG_SMP)
543 	inode_unlock(inode);
544 #endif
545 }
546 
547 static inline void orangefs_set_timeout(struct dentry *dentry)
548 {
549 	unsigned long time = jiffies + orangefs_dcache_timeout_msecs*HZ/1000;
550 
551 	dentry->d_fsdata = (void *) time;
552 }
553 
554 #endif /* __ORANGEFSKERNEL_H */
555