Loading...
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.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) + \
78sizeof(__u64) + sizeof(struct orangefs_upcall_s))
79#define MAX_DEV_REQ_DOWNSIZE (2 * sizeof(__s32) + \
80sizeof(__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 */
93enum 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 * An array of client_debug_mask will be built to hold debug keyword/mask
104 * values fetched from userspace.
105 */
106struct client_debug_mask {
107 char *keyword;
108 __u64 mask1;
109 __u64 mask2;
110};
111
112/*
113 * orangefs kernel memory related flags
114 */
115
116#if ((defined ORANGEFS_KERNEL_DEBUG) && (defined CONFIG_DEBUG_SLAB))
117#define ORANGEFS_CACHE_CREATE_FLAGS SLAB_RED_ZONE
118#else
119#define ORANGEFS_CACHE_CREATE_FLAGS 0
120#endif /* ((defined ORANGEFS_KERNEL_DEBUG) && (defined CONFIG_DEBUG_SLAB)) */
121
122/* orangefs xattr and acl related defines */
123#define ORANGEFS_XATTR_INDEX_POSIX_ACL_ACCESS 1
124#define ORANGEFS_XATTR_INDEX_POSIX_ACL_DEFAULT 2
125#define ORANGEFS_XATTR_INDEX_TRUSTED 3
126#define ORANGEFS_XATTR_INDEX_DEFAULT 4
127
128#define ORANGEFS_XATTR_NAME_ACL_ACCESS XATTR_NAME_POSIX_ACL_ACCESS
129#define ORANGEFS_XATTR_NAME_ACL_DEFAULT XATTR_NAME_POSIX_ACL_DEFAULT
130#define ORANGEFS_XATTR_NAME_TRUSTED_PREFIX "trusted."
131#define ORANGEFS_XATTR_NAME_DEFAULT_PREFIX ""
132
133/* these functions are defined in orangefs-utils.c */
134int orangefs_prepare_cdm_array(char *debug_array_string);
135int orangefs_prepare_debugfs_help_string(int);
136
137/* defined in orangefs-debugfs.c */
138int orangefs_client_debug_init(void);
139
140void debug_string_to_mask(char *, void *, int);
141void do_c_mask(int, char *, struct client_debug_mask **);
142void do_k_mask(int, char *, __u64 **);
143
144void debug_mask_to_string(void *, int);
145void do_k_string(void *, int);
146void do_c_string(void *, int);
147int check_amalgam_keyword(void *, int);
148int keyword_is_amalgam(char *);
149
150/*these variables are defined in orangefs-mod.c */
151extern char kernel_debug_string[ORANGEFS_MAX_DEBUG_STRING_LEN];
152extern char client_debug_string[ORANGEFS_MAX_DEBUG_STRING_LEN];
153extern char client_debug_array_string[ORANGEFS_MAX_DEBUG_STRING_LEN];
154extern unsigned int kernel_mask_set_mod_init;
155
156extern int orangefs_init_acl(struct inode *inode, struct inode *dir);
157extern const struct xattr_handler *orangefs_xattr_handlers[];
158
159extern struct posix_acl *orangefs_get_acl(struct inode *inode, int type);
160extern int orangefs_set_acl(struct inode *inode, struct posix_acl *acl, int type);
161
162/*
163 * Redefine xtvec structure so that we could move helper functions out of
164 * the define
165 */
166struct xtvec {
167 __kernel_off_t xtv_off; /* must be off_t */
168 __kernel_size_t xtv_len; /* must be size_t */
169};
170
171/*
172 * orangefs data structures
173 */
174struct orangefs_kernel_op_s {
175 enum orangefs_vfs_op_states op_state;
176 __u64 tag;
177
178 /*
179 * Set uses_shared_memory to non zero if this operation uses
180 * shared memory. If true, then a retry on the op must also
181 * get a new shared memory buffer and re-populate it.
182 * Cancels don't care - it only matters for service_operation()
183 * retry logics and cancels don't go through it anymore. It
184 * safely stays non-zero when we use it as slot_to_free.
185 */
186 union {
187 int uses_shared_memory;
188 int slot_to_free;
189 };
190
191 struct orangefs_upcall_s upcall;
192 struct orangefs_downcall_s downcall;
193
194 struct completion waitq;
195 spinlock_t lock;
196
197 int attempts;
198
199 struct list_head list;
200};
201
202#define set_op_state_waiting(op) ((op)->op_state = OP_VFS_STATE_WAITING)
203#define set_op_state_inprogress(op) ((op)->op_state = OP_VFS_STATE_INPROGR)
204#define set_op_state_given_up(op) ((op)->op_state = OP_VFS_STATE_GIVEN_UP)
205static inline void set_op_state_serviced(struct orangefs_kernel_op_s *op)
206{
207 op->op_state = OP_VFS_STATE_SERVICED;
208 complete(&op->waitq);
209}
210
211#define op_state_waiting(op) ((op)->op_state & OP_VFS_STATE_WAITING)
212#define op_state_in_progress(op) ((op)->op_state & OP_VFS_STATE_INPROGR)
213#define op_state_serviced(op) ((op)->op_state & OP_VFS_STATE_SERVICED)
214#define op_state_purged(op) ((op)->op_state & OP_VFS_STATE_PURGED)
215#define op_state_given_up(op) ((op)->op_state & OP_VFS_STATE_GIVEN_UP)
216#define op_is_cancel(op) ((op)->upcall.type == ORANGEFS_VFS_OP_CANCEL)
217
218void op_release(struct orangefs_kernel_op_s *op);
219
220extern void orangefs_bufmap_put(int);
221static inline void put_cancel(struct orangefs_kernel_op_s *op)
222{
223 orangefs_bufmap_put(op->slot_to_free);
224 op_release(op);
225}
226
227static inline void set_op_state_purged(struct orangefs_kernel_op_s *op)
228{
229 spin_lock(&op->lock);
230 if (unlikely(op_is_cancel(op))) {
231 list_del_init(&op->list);
232 spin_unlock(&op->lock);
233 put_cancel(op);
234 } else {
235 op->op_state |= OP_VFS_STATE_PURGED;
236 complete(&op->waitq);
237 spin_unlock(&op->lock);
238 }
239}
240
241/* per inode private orangefs info */
242struct orangefs_inode_s {
243 struct orangefs_object_kref refn;
244 char link_target[ORANGEFS_NAME_MAX];
245 __s64 blksize;
246 /*
247 * Reading/Writing Extended attributes need to acquire the appropriate
248 * reader/writer semaphore on the orangefs_inode_s structure.
249 */
250 struct rw_semaphore xattr_sem;
251
252 struct inode vfs_inode;
253 sector_t last_failed_block_index_read;
254
255 /*
256 * State of in-memory attributes not yet flushed to disk associated
257 * with this object
258 */
259 unsigned long pinode_flags;
260};
261
262#define P_ATIME_FLAG 0
263#define P_MTIME_FLAG 1
264#define P_CTIME_FLAG 2
265#define P_MODE_FLAG 3
266
267#define ClearAtimeFlag(pinode) clear_bit(P_ATIME_FLAG, &(pinode)->pinode_flags)
268#define SetAtimeFlag(pinode) set_bit(P_ATIME_FLAG, &(pinode)->pinode_flags)
269#define AtimeFlag(pinode) test_bit(P_ATIME_FLAG, &(pinode)->pinode_flags)
270
271#define ClearMtimeFlag(pinode) clear_bit(P_MTIME_FLAG, &(pinode)->pinode_flags)
272#define SetMtimeFlag(pinode) set_bit(P_MTIME_FLAG, &(pinode)->pinode_flags)
273#define MtimeFlag(pinode) test_bit(P_MTIME_FLAG, &(pinode)->pinode_flags)
274
275#define ClearCtimeFlag(pinode) clear_bit(P_CTIME_FLAG, &(pinode)->pinode_flags)
276#define SetCtimeFlag(pinode) set_bit(P_CTIME_FLAG, &(pinode)->pinode_flags)
277#define CtimeFlag(pinode) test_bit(P_CTIME_FLAG, &(pinode)->pinode_flags)
278
279#define ClearModeFlag(pinode) clear_bit(P_MODE_FLAG, &(pinode)->pinode_flags)
280#define SetModeFlag(pinode) set_bit(P_MODE_FLAG, &(pinode)->pinode_flags)
281#define ModeFlag(pinode) test_bit(P_MODE_FLAG, &(pinode)->pinode_flags)
282
283/* per superblock private orangefs info */
284struct orangefs_sb_info_s {
285 struct orangefs_khandle root_khandle;
286 __s32 fs_id;
287 int id;
288 int flags;
289#define ORANGEFS_OPT_INTR 0x01
290#define ORANGEFS_OPT_LOCAL_LOCK 0x02
291 char devname[ORANGEFS_MAX_SERVER_ADDR_LEN];
292 struct super_block *sb;
293 int mount_pending;
294 struct list_head list;
295};
296
297/*
298 * structure that holds the state of any async I/O operation issued
299 * through the VFS. Needed especially to handle cancellation requests
300 * or even completion notification so that the VFS client-side daemon
301 * can free up its vfs_request slots.
302 */
303struct orangefs_kiocb_s {
304 /* the pointer to the task that initiated the AIO */
305 struct task_struct *tsk;
306
307 /* pointer to the kiocb that kicked this operation */
308 struct kiocb *kiocb;
309
310 /* buffer index that was used for the I/O */
311 struct orangefs_bufmap *bufmap;
312 int buffer_index;
313
314 /* orangefs kernel operation type */
315 struct orangefs_kernel_op_s *op;
316
317 /* The user space buffers from/to which I/O is being staged */
318 struct iovec *iov;
319
320 /* number of elements in the iovector */
321 unsigned long nr_segs;
322
323 /* set to indicate the type of the operation */
324 int rw;
325
326 /* file offset */
327 loff_t offset;
328
329 /* and the count in bytes */
330 size_t bytes_to_be_copied;
331
332 ssize_t bytes_copied;
333 int needs_cleanup;
334};
335
336struct orangefs_stats {
337 unsigned long cache_hits;
338 unsigned long cache_misses;
339 unsigned long reads;
340 unsigned long writes;
341};
342
343extern struct orangefs_stats g_orangefs_stats;
344
345/*
346 * NOTE: See Documentation/filesystems/porting for information
347 * on implementing FOO_I and properly accessing fs private data
348 */
349static inline struct orangefs_inode_s *ORANGEFS_I(struct inode *inode)
350{
351 return container_of(inode, struct orangefs_inode_s, vfs_inode);
352}
353
354static inline struct orangefs_sb_info_s *ORANGEFS_SB(struct super_block *sb)
355{
356 return (struct orangefs_sb_info_s *) sb->s_fs_info;
357}
358
359/* ino_t descends from "unsigned long", 8 bytes, 64 bits. */
360static inline ino_t orangefs_khandle_to_ino(struct orangefs_khandle *khandle)
361{
362 union {
363 unsigned char u[8];
364 __u64 ino;
365 } ihandle;
366
367 ihandle.u[0] = khandle->u[0] ^ khandle->u[4];
368 ihandle.u[1] = khandle->u[1] ^ khandle->u[5];
369 ihandle.u[2] = khandle->u[2] ^ khandle->u[6];
370 ihandle.u[3] = khandle->u[3] ^ khandle->u[7];
371 ihandle.u[4] = khandle->u[12] ^ khandle->u[8];
372 ihandle.u[5] = khandle->u[13] ^ khandle->u[9];
373 ihandle.u[6] = khandle->u[14] ^ khandle->u[10];
374 ihandle.u[7] = khandle->u[15] ^ khandle->u[11];
375
376 return ihandle.ino;
377}
378
379static inline struct orangefs_khandle *get_khandle_from_ino(struct inode *inode)
380{
381 return &(ORANGEFS_I(inode)->refn.khandle);
382}
383
384static inline __s32 get_fsid_from_ino(struct inode *inode)
385{
386 return ORANGEFS_I(inode)->refn.fs_id;
387}
388
389static inline ino_t get_ino_from_khandle(struct inode *inode)
390{
391 struct orangefs_khandle *khandle;
392 ino_t ino;
393
394 khandle = get_khandle_from_ino(inode);
395 ino = orangefs_khandle_to_ino(khandle);
396 return ino;
397}
398
399static inline ino_t get_parent_ino_from_dentry(struct dentry *dentry)
400{
401 return get_ino_from_khandle(dentry->d_parent->d_inode);
402}
403
404static inline int is_root_handle(struct inode *inode)
405{
406 gossip_debug(GOSSIP_DCACHE_DEBUG,
407 "%s: root handle: %pU, this handle: %pU:\n",
408 __func__,
409 &ORANGEFS_SB(inode->i_sb)->root_khandle,
410 get_khandle_from_ino(inode));
411
412 if (ORANGEFS_khandle_cmp(&(ORANGEFS_SB(inode->i_sb)->root_khandle),
413 get_khandle_from_ino(inode)))
414 return 0;
415 else
416 return 1;
417}
418
419static inline int match_handle(struct orangefs_khandle resp_handle,
420 struct inode *inode)
421{
422 gossip_debug(GOSSIP_DCACHE_DEBUG,
423 "%s: one handle: %pU, another handle:%pU:\n",
424 __func__,
425 &resp_handle,
426 get_khandle_from_ino(inode));
427
428 if (ORANGEFS_khandle_cmp(&resp_handle, get_khandle_from_ino(inode)))
429 return 0;
430 else
431 return 1;
432}
433
434/*
435 * defined in orangefs-cache.c
436 */
437int op_cache_initialize(void);
438int op_cache_finalize(void);
439struct orangefs_kernel_op_s *op_alloc(__s32 type);
440void orangefs_new_tag(struct orangefs_kernel_op_s *op);
441char *get_opname_string(struct orangefs_kernel_op_s *new_op);
442
443int orangefs_inode_cache_initialize(void);
444int orangefs_inode_cache_finalize(void);
445
446/*
447 * defined in orangefs-mod.c
448 */
449void purge_inprogress_ops(void);
450
451/*
452 * defined in waitqueue.c
453 */
454void purge_waiting_ops(void);
455
456/*
457 * defined in super.c
458 */
459struct dentry *orangefs_mount(struct file_system_type *fst,
460 int flags,
461 const char *devname,
462 void *data);
463
464void orangefs_kill_sb(struct super_block *sb);
465int orangefs_remount(struct orangefs_sb_info_s *);
466
467int fsid_key_table_initialize(void);
468void fsid_key_table_finalize(void);
469
470/*
471 * defined in inode.c
472 */
473__u32 convert_to_orangefs_mask(unsigned long lite_mask);
474struct inode *orangefs_new_inode(struct super_block *sb,
475 struct inode *dir,
476 int mode,
477 dev_t dev,
478 struct orangefs_object_kref *ref);
479
480int orangefs_setattr(struct dentry *dentry, struct iattr *iattr);
481
482int orangefs_getattr(struct vfsmount *mnt,
483 struct dentry *dentry,
484 struct kstat *kstat);
485
486int orangefs_permission(struct inode *inode, int mask);
487
488/*
489 * defined in xattr.c
490 */
491int orangefs_setxattr(struct dentry *dentry,
492 const char *name,
493 const void *value,
494 size_t size,
495 int flags);
496
497ssize_t orangefs_getxattr(struct dentry *dentry,
498 const char *name,
499 void *buffer,
500 size_t size);
501
502ssize_t orangefs_listxattr(struct dentry *dentry, char *buffer, size_t size);
503
504/*
505 * defined in namei.c
506 */
507struct inode *orangefs_iget(struct super_block *sb,
508 struct orangefs_object_kref *ref);
509
510ssize_t orangefs_inode_read(struct inode *inode,
511 struct iov_iter *iter,
512 loff_t *offset,
513 loff_t readahead_size);
514
515/*
516 * defined in devorangefs-req.c
517 */
518int orangefs_dev_init(void);
519void orangefs_dev_cleanup(void);
520int is_daemon_in_service(void);
521bool __is_daemon_in_service(void);
522
523/*
524 * defined in orangefs-utils.c
525 */
526__s32 fsid_of_op(struct orangefs_kernel_op_s *op);
527
528int orangefs_flush_inode(struct inode *inode);
529
530ssize_t orangefs_inode_getxattr(struct inode *inode,
531 const char *prefix,
532 const char *name,
533 void *buffer,
534 size_t size);
535
536int orangefs_inode_setxattr(struct inode *inode,
537 const char *prefix,
538 const char *name,
539 const void *value,
540 size_t size,
541 int flags);
542
543int orangefs_inode_getattr(struct inode *inode, int new, int size);
544
545int orangefs_inode_check_changed(struct inode *inode);
546
547int orangefs_inode_setattr(struct inode *inode, struct iattr *iattr);
548
549void orangefs_make_bad_inode(struct inode *inode);
550
551int orangefs_unmount_sb(struct super_block *sb);
552
553bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op);
554
555int orangefs_normalize_to_errno(__s32 error_code);
556
557extern struct mutex devreq_mutex;
558extern struct mutex request_mutex;
559extern int debug;
560extern int op_timeout_secs;
561extern int slot_timeout_secs;
562extern struct list_head orangefs_superblocks;
563extern spinlock_t orangefs_superblocks_lock;
564extern struct list_head orangefs_request_list;
565extern spinlock_t orangefs_request_list_lock;
566extern wait_queue_head_t orangefs_request_list_waitq;
567extern struct list_head *htable_ops_in_progress;
568extern spinlock_t htable_ops_in_progress_lock;
569extern int hash_table_size;
570
571extern const struct address_space_operations orangefs_address_operations;
572extern struct backing_dev_info orangefs_backing_dev_info;
573extern struct inode_operations orangefs_file_inode_operations;
574extern const struct file_operations orangefs_file_operations;
575extern struct inode_operations orangefs_symlink_inode_operations;
576extern struct inode_operations orangefs_dir_inode_operations;
577extern const struct file_operations orangefs_dir_operations;
578extern const struct dentry_operations orangefs_dentry_operations;
579extern const struct file_operations orangefs_devreq_file_operations;
580
581extern wait_queue_head_t orangefs_bufmap_init_waitq;
582
583/*
584 * misc convenience macros
585 */
586
587#define ORANGEFS_OP_INTERRUPTIBLE 1 /* service_operation() is interruptible */
588#define ORANGEFS_OP_PRIORITY 2 /* service_operation() is high priority */
589#define ORANGEFS_OP_CANCELLATION 4 /* this is a cancellation */
590#define ORANGEFS_OP_NO_MUTEX 8 /* don't acquire request_mutex */
591#define ORANGEFS_OP_ASYNC 16 /* Queue it, but don't wait */
592
593int service_operation(struct orangefs_kernel_op_s *op,
594 const char *op_name,
595 int flags);
596
597#define get_interruptible_flag(inode) \
598 ((ORANGEFS_SB(inode->i_sb)->flags & ORANGEFS_OPT_INTR) ? \
599 ORANGEFS_OP_INTERRUPTIBLE : 0)
600
601#define fill_default_sys_attrs(sys_attr, type, mode) \
602do { \
603 sys_attr.owner = from_kuid(current_user_ns(), current_fsuid()); \
604 sys_attr.group = from_kgid(current_user_ns(), current_fsgid()); \
605 sys_attr.perms = ORANGEFS_util_translate_mode(mode); \
606 sys_attr.mtime = 0; \
607 sys_attr.atime = 0; \
608 sys_attr.ctime = 0; \
609 sys_attr.mask = ORANGEFS_ATTR_SYS_ALL_SETABLE; \
610} while (0)
611
612static inline void orangefs_i_size_write(struct inode *inode, loff_t i_size)
613{
614#if BITS_PER_LONG == 32 && defined(CONFIG_SMP)
615 mutex_lock(&inode->i_mutex);
616#endif
617 i_size_write(inode, i_size);
618#if BITS_PER_LONG == 32 && defined(CONFIG_SMP)
619 mutex_unlock(&inode->i_mutex);
620#endif
621}
622
623#endif /* __ORANGEFSKERNEL_H */
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.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) + \
78sizeof(__u64) + sizeof(struct orangefs_upcall_s))
79#define MAX_DEV_REQ_DOWNSIZE (2 * sizeof(__s32) + \
80sizeof(__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 */
93enum 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
112extern int orangefs_init_acl(struct inode *inode, struct inode *dir);
113extern const struct xattr_handler *orangefs_xattr_handlers[];
114
115extern struct posix_acl *orangefs_get_acl(struct inode *inode, int type);
116extern 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 */
122struct 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 */
130struct 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)
161static 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
174void op_release(struct orangefs_kernel_op_s *op);
175
176extern void orangefs_bufmap_put(int);
177static 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
183static 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 */
198struct 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};
219
220#define P_ATIME_FLAG 0
221#define P_MTIME_FLAG 1
222#define P_CTIME_FLAG 2
223#define P_MODE_FLAG 3
224
225#define ClearAtimeFlag(pinode) clear_bit(P_ATIME_FLAG, &(pinode)->pinode_flags)
226#define SetAtimeFlag(pinode) set_bit(P_ATIME_FLAG, &(pinode)->pinode_flags)
227#define AtimeFlag(pinode) test_bit(P_ATIME_FLAG, &(pinode)->pinode_flags)
228
229#define ClearMtimeFlag(pinode) clear_bit(P_MTIME_FLAG, &(pinode)->pinode_flags)
230#define SetMtimeFlag(pinode) set_bit(P_MTIME_FLAG, &(pinode)->pinode_flags)
231#define MtimeFlag(pinode) test_bit(P_MTIME_FLAG, &(pinode)->pinode_flags)
232
233#define ClearCtimeFlag(pinode) clear_bit(P_CTIME_FLAG, &(pinode)->pinode_flags)
234#define SetCtimeFlag(pinode) set_bit(P_CTIME_FLAG, &(pinode)->pinode_flags)
235#define CtimeFlag(pinode) test_bit(P_CTIME_FLAG, &(pinode)->pinode_flags)
236
237#define ClearModeFlag(pinode) clear_bit(P_MODE_FLAG, &(pinode)->pinode_flags)
238#define SetModeFlag(pinode) set_bit(P_MODE_FLAG, &(pinode)->pinode_flags)
239#define ModeFlag(pinode) test_bit(P_MODE_FLAG, &(pinode)->pinode_flags)
240
241/* per superblock private orangefs info */
242struct orangefs_sb_info_s {
243 struct orangefs_khandle root_khandle;
244 __s32 fs_id;
245 int id;
246 int flags;
247#define ORANGEFS_OPT_INTR 0x01
248#define ORANGEFS_OPT_LOCAL_LOCK 0x02
249 char devname[ORANGEFS_MAX_SERVER_ADDR_LEN];
250 struct super_block *sb;
251 int mount_pending;
252 struct list_head list;
253};
254
255/*
256 * structure that holds the state of any async I/O operation issued
257 * through the VFS. Needed especially to handle cancellation requests
258 * or even completion notification so that the VFS client-side daemon
259 * can free up its vfs_request slots.
260 */
261struct orangefs_kiocb_s {
262 /* the pointer to the task that initiated the AIO */
263 struct task_struct *tsk;
264
265 /* pointer to the kiocb that kicked this operation */
266 struct kiocb *kiocb;
267
268 /* buffer index that was used for the I/O */
269 struct orangefs_bufmap *bufmap;
270 int buffer_index;
271
272 /* orangefs kernel operation type */
273 struct orangefs_kernel_op_s *op;
274
275 /* The user space buffers from/to which I/O is being staged */
276 struct iovec *iov;
277
278 /* number of elements in the iovector */
279 unsigned long nr_segs;
280
281 /* set to indicate the type of the operation */
282 int rw;
283
284 /* file offset */
285 loff_t offset;
286
287 /* and the count in bytes */
288 size_t bytes_to_be_copied;
289
290 ssize_t bytes_copied;
291 int needs_cleanup;
292};
293
294struct orangefs_stats {
295 unsigned long cache_hits;
296 unsigned long cache_misses;
297 unsigned long reads;
298 unsigned long writes;
299};
300
301extern struct orangefs_stats orangefs_stats;
302
303/*
304 * NOTE: See Documentation/filesystems/porting for information
305 * on implementing FOO_I and properly accessing fs private data
306 */
307static inline struct orangefs_inode_s *ORANGEFS_I(struct inode *inode)
308{
309 return container_of(inode, struct orangefs_inode_s, vfs_inode);
310}
311
312static inline struct orangefs_sb_info_s *ORANGEFS_SB(struct super_block *sb)
313{
314 return (struct orangefs_sb_info_s *) sb->s_fs_info;
315}
316
317/* ino_t descends from "unsigned long", 8 bytes, 64 bits. */
318static inline ino_t orangefs_khandle_to_ino(struct orangefs_khandle *khandle)
319{
320 union {
321 unsigned char u[8];
322 __u64 ino;
323 } ihandle;
324
325 ihandle.u[0] = khandle->u[0] ^ khandle->u[4];
326 ihandle.u[1] = khandle->u[1] ^ khandle->u[5];
327 ihandle.u[2] = khandle->u[2] ^ khandle->u[6];
328 ihandle.u[3] = khandle->u[3] ^ khandle->u[7];
329 ihandle.u[4] = khandle->u[12] ^ khandle->u[8];
330 ihandle.u[5] = khandle->u[13] ^ khandle->u[9];
331 ihandle.u[6] = khandle->u[14] ^ khandle->u[10];
332 ihandle.u[7] = khandle->u[15] ^ khandle->u[11];
333
334 return ihandle.ino;
335}
336
337static inline struct orangefs_khandle *get_khandle_from_ino(struct inode *inode)
338{
339 return &(ORANGEFS_I(inode)->refn.khandle);
340}
341
342static inline __s32 get_fsid_from_ino(struct inode *inode)
343{
344 return ORANGEFS_I(inode)->refn.fs_id;
345}
346
347static inline ino_t get_ino_from_khandle(struct inode *inode)
348{
349 struct orangefs_khandle *khandle;
350 ino_t ino;
351
352 khandle = get_khandle_from_ino(inode);
353 ino = orangefs_khandle_to_ino(khandle);
354 return ino;
355}
356
357static inline ino_t get_parent_ino_from_dentry(struct dentry *dentry)
358{
359 return get_ino_from_khandle(dentry->d_parent->d_inode);
360}
361
362static inline int is_root_handle(struct inode *inode)
363{
364 gossip_debug(GOSSIP_DCACHE_DEBUG,
365 "%s: root handle: %pU, this handle: %pU:\n",
366 __func__,
367 &ORANGEFS_SB(inode->i_sb)->root_khandle,
368 get_khandle_from_ino(inode));
369
370 if (ORANGEFS_khandle_cmp(&(ORANGEFS_SB(inode->i_sb)->root_khandle),
371 get_khandle_from_ino(inode)))
372 return 0;
373 else
374 return 1;
375}
376
377static inline int match_handle(struct orangefs_khandle resp_handle,
378 struct inode *inode)
379{
380 gossip_debug(GOSSIP_DCACHE_DEBUG,
381 "%s: one handle: %pU, another handle:%pU:\n",
382 __func__,
383 &resp_handle,
384 get_khandle_from_ino(inode));
385
386 if (ORANGEFS_khandle_cmp(&resp_handle, get_khandle_from_ino(inode)))
387 return 0;
388 else
389 return 1;
390}
391
392/*
393 * defined in orangefs-cache.c
394 */
395int op_cache_initialize(void);
396int op_cache_finalize(void);
397struct orangefs_kernel_op_s *op_alloc(__s32 type);
398void orangefs_new_tag(struct orangefs_kernel_op_s *op);
399char *get_opname_string(struct orangefs_kernel_op_s *new_op);
400
401int orangefs_inode_cache_initialize(void);
402int orangefs_inode_cache_finalize(void);
403
404/*
405 * defined in orangefs-mod.c
406 */
407void purge_inprogress_ops(void);
408
409/*
410 * defined in waitqueue.c
411 */
412void purge_waiting_ops(void);
413
414/*
415 * defined in super.c
416 */
417extern uint64_t orangefs_features;
418
419struct dentry *orangefs_mount(struct file_system_type *fst,
420 int flags,
421 const char *devname,
422 void *data);
423
424void orangefs_kill_sb(struct super_block *sb);
425int orangefs_remount(struct orangefs_sb_info_s *);
426
427int fsid_key_table_initialize(void);
428void fsid_key_table_finalize(void);
429
430/*
431 * defined in inode.c
432 */
433__u32 convert_to_orangefs_mask(unsigned long lite_mask);
434struct inode *orangefs_new_inode(struct super_block *sb,
435 struct inode *dir,
436 int mode,
437 dev_t dev,
438 struct orangefs_object_kref *ref);
439
440int orangefs_setattr(struct dentry *dentry, struct iattr *iattr);
441
442int orangefs_getattr(struct vfsmount *mnt,
443 struct dentry *dentry,
444 struct kstat *kstat);
445
446int orangefs_permission(struct inode *inode, int mask);
447
448/*
449 * defined in xattr.c
450 */
451int orangefs_setxattr(struct dentry *dentry,
452 const char *name,
453 const void *value,
454 size_t size,
455 int flags);
456
457ssize_t orangefs_getxattr(struct dentry *dentry,
458 const char *name,
459 void *buffer,
460 size_t size);
461
462ssize_t orangefs_listxattr(struct dentry *dentry, char *buffer, size_t size);
463
464/*
465 * defined in namei.c
466 */
467struct inode *orangefs_iget(struct super_block *sb,
468 struct orangefs_object_kref *ref);
469
470ssize_t orangefs_inode_read(struct inode *inode,
471 struct iov_iter *iter,
472 loff_t *offset,
473 loff_t readahead_size);
474
475/*
476 * defined in devorangefs-req.c
477 */
478extern uint32_t orangefs_userspace_version;
479
480int orangefs_dev_init(void);
481void orangefs_dev_cleanup(void);
482int is_daemon_in_service(void);
483bool __is_daemon_in_service(void);
484
485/*
486 * defined in orangefs-utils.c
487 */
488__s32 fsid_of_op(struct orangefs_kernel_op_s *op);
489
490int orangefs_flush_inode(struct inode *inode);
491
492ssize_t orangefs_inode_getxattr(struct inode *inode,
493 const char *name,
494 void *buffer,
495 size_t size);
496
497int orangefs_inode_setxattr(struct inode *inode,
498 const char *name,
499 const void *value,
500 size_t size,
501 int flags);
502
503int orangefs_inode_getattr(struct inode *inode, int new, int bypass);
504
505int orangefs_inode_check_changed(struct inode *inode);
506
507int orangefs_inode_setattr(struct inode *inode, struct iattr *iattr);
508
509void orangefs_make_bad_inode(struct inode *inode);
510
511int orangefs_unmount_sb(struct super_block *sb);
512
513bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op);
514
515int orangefs_normalize_to_errno(__s32 error_code);
516
517extern struct mutex orangefs_request_mutex;
518extern int op_timeout_secs;
519extern int slot_timeout_secs;
520extern int orangefs_dcache_timeout_msecs;
521extern int orangefs_getattr_timeout_msecs;
522extern struct list_head orangefs_superblocks;
523extern spinlock_t orangefs_superblocks_lock;
524extern struct list_head orangefs_request_list;
525extern spinlock_t orangefs_request_list_lock;
526extern wait_queue_head_t orangefs_request_list_waitq;
527extern struct list_head *orangefs_htable_ops_in_progress;
528extern spinlock_t orangefs_htable_ops_in_progress_lock;
529extern int hash_table_size;
530
531extern const struct address_space_operations orangefs_address_operations;
532extern struct backing_dev_info orangefs_backing_dev_info;
533extern const struct inode_operations orangefs_file_inode_operations;
534extern const struct file_operations orangefs_file_operations;
535extern const struct inode_operations orangefs_symlink_inode_operations;
536extern const struct inode_operations orangefs_dir_inode_operations;
537extern const struct file_operations orangefs_dir_operations;
538extern const struct dentry_operations orangefs_dentry_operations;
539extern const struct file_operations orangefs_devreq_file_operations;
540
541extern wait_queue_head_t orangefs_bufmap_init_waitq;
542
543/*
544 * misc convenience macros
545 */
546
547#define ORANGEFS_OP_INTERRUPTIBLE 1 /* service_operation() is interruptible */
548#define ORANGEFS_OP_PRIORITY 2 /* service_operation() is high priority */
549#define ORANGEFS_OP_CANCELLATION 4 /* this is a cancellation */
550#define ORANGEFS_OP_NO_MUTEX 8 /* don't acquire request_mutex */
551#define ORANGEFS_OP_ASYNC 16 /* Queue it, but don't wait */
552
553int service_operation(struct orangefs_kernel_op_s *op,
554 const char *op_name,
555 int flags);
556
557#define get_interruptible_flag(inode) \
558 ((ORANGEFS_SB(inode->i_sb)->flags & ORANGEFS_OPT_INTR) ? \
559 ORANGEFS_OP_INTERRUPTIBLE : 0)
560
561#define fill_default_sys_attrs(sys_attr, type, mode) \
562do { \
563 sys_attr.owner = from_kuid(&init_user_ns, current_fsuid()); \
564 sys_attr.group = from_kgid(&init_user_ns, current_fsgid()); \
565 sys_attr.perms = ORANGEFS_util_translate_mode(mode); \
566 sys_attr.mtime = 0; \
567 sys_attr.atime = 0; \
568 sys_attr.ctime = 0; \
569 sys_attr.mask = ORANGEFS_ATTR_SYS_ALL_SETABLE; \
570} while (0)
571
572static inline void orangefs_i_size_write(struct inode *inode, loff_t i_size)
573{
574#if BITS_PER_LONG == 32 && defined(CONFIG_SMP)
575 inode_lock(inode);
576#endif
577 i_size_write(inode, i_size);
578#if BITS_PER_LONG == 32 && defined(CONFIG_SMP)
579 inode_unlock(inode);
580#endif
581}
582
583static inline void orangefs_set_timeout(struct dentry *dentry)
584{
585 unsigned long time = jiffies + orangefs_dcache_timeout_msecs*HZ/1000;
586
587 dentry->d_fsdata = (void *) time;
588}
589
590#endif /* __ORANGEFSKERNEL_H */