projects/linux·blob master · refs

815 lines · 21.9 KiB

1// SPDX-License-Identifier: GPL-2.0
2/*
3 * (C) 2001 Clemson University and The University of Chicago
4 *
5 * Changes by Acxiom Corporation to add protocol version to kernel
6 * communication, Copyright Acxiom Corporation, 2005.
7 *
8 * See COPYING in top-level directory.
9 */
10
11#include "protocol.h"
12#include "orangefs-kernel.h"
13#include "orangefs-dev-proto.h"
14#include "orangefs-bufmap.h"
15#include "orangefs-debugfs.h"
16
17#include <linux/debugfs.h>
18#include <linux/slab.h>
19
20/* this file implements the /dev/pvfs2-req device node */
21
22uint32_t orangefs_userspace_version;
23
24static int open_access_count;
25
26static DEFINE_MUTEX(devreq_mutex);
27
28#define DUMP_DEVICE_ERROR() \
29do { \
30 gossip_err("*****************************************************\n");\
31 gossip_err("ORANGEFS Device Error: You cannot open the device file "); \
32 gossip_err("\n/dev/%s more than once. Please make sure that\nthere " \
33 "are no ", ORANGEFS_REQDEVICE_NAME); \
34 gossip_err("instances of a program using this device\ncurrently " \
35 "running. (You must verify this!)\n"); \
36 gossip_err("For example, you can use the lsof program as follows:\n");\
37 gossip_err("'lsof | grep %s' (run this as root)\n", \
38 ORANGEFS_REQDEVICE_NAME); \
39 gossip_err(" open_access_count = %d\n", open_access_count); \
40 gossip_err("*****************************************************\n");\
41} while (0)
42
43static int hash_func(__u64 tag, int table_size)
44{
45 return do_div(tag, (unsigned int)table_size);
46}
47
48static void orangefs_devreq_add_op(struct orangefs_kernel_op_s *op)
49{
50 int index = hash_func(op->tag, hash_table_size);
51
52 list_add_tail(&op->list, &orangefs_htable_ops_in_progress[index]);
53}
54
55/*
56 * find the op with this tag and remove it from the in progress
57 * hash table.
58 */
59static struct orangefs_kernel_op_s *orangefs_devreq_remove_op(__u64 tag)
60{
61 struct orangefs_kernel_op_s *op, *next;
62 int index;
63
64 index = hash_func(tag, hash_table_size);
65
66 spin_lock(&orangefs_htable_ops_in_progress_lock);
67 list_for_each_entry_safe(op,
68 next,
69 &orangefs_htable_ops_in_progress[index],
70 list) {
71 if (op->tag == tag && !op_state_purged(op) &&
72 !op_state_given_up(op)) {
73 list_del_init(&op->list);
74 spin_unlock(&orangefs_htable_ops_in_progress_lock);
75 return op;
76 }
77 }
78
79 spin_unlock(&orangefs_htable_ops_in_progress_lock);
80 return NULL;
81}
82
83/* Returns whether any FS are still pending remounted */
84static int mark_all_pending_mounts(void)
85{
86 int unmounted = 1;
87 struct orangefs_sb_info_s *orangefs_sb = NULL;
88
89 spin_lock(&orangefs_superblocks_lock);
90 list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) {
91 /* All of these file system require a remount */
92 orangefs_sb->mount_pending = 1;
93 unmounted = 0;
94 }
95 spin_unlock(&orangefs_superblocks_lock);
96 return unmounted;
97}
98
99/*
100 * Determine if a given file system needs to be remounted or not
101 * Returns -1 on error
102 * 0 if already mounted
103 * 1 if needs remount
104 */
105static int fs_mount_pending(__s32 fsid)
106{
107 int mount_pending = -1;
108 struct orangefs_sb_info_s *orangefs_sb = NULL;
109
110 spin_lock(&orangefs_superblocks_lock);
111 list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) {
112 if (orangefs_sb->fs_id == fsid) {
113 mount_pending = orangefs_sb->mount_pending;
114 break;
115 }
116 }
117 spin_unlock(&orangefs_superblocks_lock);
118 return mount_pending;
119}
120
121static int orangefs_devreq_open(struct inode *inode, struct file *file)
122{
123 int ret = -EINVAL;
124
125 /* in order to ensure that the filesystem driver sees correct UIDs */
126 if (file->f_cred->user_ns != &init_user_ns) {
127 gossip_err("%s: device cannot be opened outside init_user_ns\n",
128 __func__);
129 goto out;
130 }
131
132 if (!(file->f_flags & O_NONBLOCK)) {
133 gossip_err("%s: device cannot be opened in blocking mode\n",
134 __func__);
135 goto out;
136 }
137 ret = -EACCES;
138 gossip_debug(GOSSIP_DEV_DEBUG, "client-core: opening device\n");
139 mutex_lock(&devreq_mutex);
140
141 if (open_access_count == 0) {
142 open_access_count = 1;
143 ret = 0;
144 } else {
145 DUMP_DEVICE_ERROR();
146 }
147 mutex_unlock(&devreq_mutex);
148
149out:
150
151 gossip_debug(GOSSIP_DEV_DEBUG,
152 "pvfs2-client-core: open device complete (ret = %d)\n",
153 ret);
154 return ret;
155}
156
157/* Function for read() callers into the device */
158static ssize_t orangefs_devreq_read(struct file *file,
159 char __user *buf,
160 size_t count, loff_t *offset)
161{
162 struct orangefs_kernel_op_s *op, *temp;
163 __s32 proto_ver = ORANGEFS_KERNEL_PROTO_VERSION;
164 static __s32 magic = ORANGEFS_DEVREQ_MAGIC;
165 struct orangefs_kernel_op_s *cur_op;
166 unsigned long ret;
167
168 /* We do not support blocking IO. */
169 if (!(file->f_flags & O_NONBLOCK)) {
170 gossip_err("%s: blocking read from client-core.\n",
171 __func__);
172 return -EINVAL;
173 }
174
175 /*
176 * The client will do an ioctl to find MAX_DEV_REQ_UPSIZE, then
177 * always read with that size buffer.
178 */
179 if (count != MAX_DEV_REQ_UPSIZE) {
180 gossip_err("orangefs: client-core tried to read wrong size\n");
181 return -EINVAL;
182 }
183
184 /* Check for an empty list before locking. */
185 if (list_empty(&orangefs_request_list))
186 return -EAGAIN;
187
188restart:
189 cur_op = NULL;
190 /* Get next op (if any) from top of list. */
191 spin_lock(&orangefs_request_list_lock);
192 list_for_each_entry_safe(op, temp, &orangefs_request_list, list) {
193 __s32 fsid;
194 /* This lock is held past the end of the loop when we break. */
195 spin_lock(&op->lock);
196 if (unlikely(op_state_purged(op) || op_state_given_up(op))) {
197 spin_unlock(&op->lock);
198 continue;
199 }
200
201 fsid = fsid_of_op(op);
202 if (fsid != ORANGEFS_FS_ID_NULL) {
203 int ret;
204 /* Skip ops whose filesystem needs to be mounted. */
205 ret = fs_mount_pending(fsid);
206 if (ret == 1) {
207 gossip_debug(GOSSIP_DEV_DEBUG,
208 "%s: mount pending, skipping op tag "
209 "%llu %s\n",
210 __func__,
211 llu(op->tag),
212 get_opname_string(op));
213 spin_unlock(&op->lock);
214 continue;
215 /*
216 * Skip ops whose filesystem we don't know about unless
217 * it is being mounted or unmounted. It is possible for
218 * a filesystem we don't know about to be unmounted if
219 * it fails to mount in the kernel after userspace has
220 * been sent the mount request.
221 */
222 /* XXX: is there a better way to detect this? */
223 } else if (ret == -1 &&
224 !(op->upcall.type ==
225 ORANGEFS_VFS_OP_FS_MOUNT ||
226 op->upcall.type ==
227 ORANGEFS_VFS_OP_GETATTR ||
228 op->upcall.type ==
229 ORANGEFS_VFS_OP_FS_UMOUNT)) {
230 gossip_debug(GOSSIP_DEV_DEBUG,
231 "orangefs: skipping op tag %llu %s\n",
232 llu(op->tag), get_opname_string(op));
233 gossip_err(
234 "orangefs: ERROR: fs_mount_pending %d\n",
235 fsid);
236 spin_unlock(&op->lock);
237 continue;
238 }
239 }
240 /*
241 * Either this op does not pertain to a filesystem, is mounting
242 * a filesystem, or pertains to a mounted filesystem. Let it
243 * through.
244 */
245 cur_op = op;
246 break;
247 }
248
249 /*
250 * At this point we either have a valid op and can continue or have not
251 * found an op and must ask the client to try again later.
252 */
253 if (!cur_op) {
254 spin_unlock(&orangefs_request_list_lock);
255 return -EAGAIN;
256 }
257
258 gossip_debug(GOSSIP_DEV_DEBUG, "%s: reading op tag %llu %s\n",
259 __func__,
260 llu(cur_op->tag),
261 get_opname_string(cur_op));
262
263 /*
264 * Such an op should never be on the list in the first place. If so, we
265 * will abort.
266 */
267 if (op_state_in_progress(cur_op) || op_state_serviced(cur_op)) {
268 gossip_err("orangefs: ERROR: Current op already queued.\n");
269 list_del_init(&cur_op->list);
270 spin_unlock(&cur_op->lock);
271 spin_unlock(&orangefs_request_list_lock);
272 return -EAGAIN;
273 }
274
275 list_del_init(&cur_op->list);
276 spin_unlock(&orangefs_request_list_lock);
277
278 spin_unlock(&cur_op->lock);
279
280 /* Push the upcall out. */
281 ret = copy_to_user(buf, &proto_ver, sizeof(__s32));
282 if (ret != 0)
283 goto error;
284 ret = copy_to_user(buf + sizeof(__s32), &magic, sizeof(__s32));
285 if (ret != 0)
286 goto error;
287 ret = copy_to_user(buf + 2 * sizeof(__s32),
288 &cur_op->tag,
289 sizeof(__u64));
290 if (ret != 0)
291 goto error;
292 ret = copy_to_user(buf + 2 * sizeof(__s32) + sizeof(__u64),
293 &cur_op->upcall,
294 sizeof(struct orangefs_upcall_s));
295 if (ret != 0)
296 goto error;
297
298 spin_lock(&orangefs_htable_ops_in_progress_lock);
299 spin_lock(&cur_op->lock);
300 if (unlikely(op_state_given_up(cur_op))) {
301 spin_unlock(&cur_op->lock);
302 spin_unlock(&orangefs_htable_ops_in_progress_lock);
303 complete(&cur_op->waitq);
304 goto restart;
305 }
306
307 /*
308 * Set the operation to be in progress and move it between lists since
309 * it has been sent to the client.
310 */
311 set_op_state_inprogress(cur_op);
312 gossip_debug(GOSSIP_DEV_DEBUG,
313 "%s: 1 op:%s: op_state:%d: process:%s:\n",
314 __func__,
315 get_opname_string(cur_op),
316 cur_op->op_state,
317 current->comm);
318 orangefs_devreq_add_op(cur_op);
319 spin_unlock(&cur_op->lock);
320 spin_unlock(&orangefs_htable_ops_in_progress_lock);
321
322 /* The client only asks to read one size buffer. */
323 return MAX_DEV_REQ_UPSIZE;
324error:
325 /*
326 * We were unable to copy the op data to the client. Put the op back in
327 * list. If client has crashed, the op will be purged later when the
328 * device is released.
329 */
330 gossip_err("orangefs: Failed to copy data to user space\n");
331 spin_lock(&orangefs_request_list_lock);
332 spin_lock(&cur_op->lock);
333 if (likely(!op_state_given_up(cur_op))) {
334 set_op_state_waiting(cur_op);
335 gossip_debug(GOSSIP_DEV_DEBUG,
336 "%s: 2 op:%s: op_state:%d: process:%s:\n",
337 __func__,
338 get_opname_string(cur_op),
339 cur_op->op_state,
340 current->comm);
341 list_add(&cur_op->list, &orangefs_request_list);
342 spin_unlock(&cur_op->lock);
343 } else {
344 spin_unlock(&cur_op->lock);
345 complete(&cur_op->waitq);
346 }
347 spin_unlock(&orangefs_request_list_lock);
348 return -EFAULT;
349}
350
351/*
352 * Function for writev() callers into the device.
353 *
354 * Userspace should have written:
355 * - __u32 version
356 * - __u32 magic
357 * - __u64 tag
358 * - struct orangefs_downcall_s
359 * - trailer buffer (in the case of READDIR operations)
360 */
361static ssize_t orangefs_devreq_write_iter(struct kiocb *iocb,
362 struct iov_iter *iter)
363{
364 ssize_t ret;
365 struct orangefs_kernel_op_s *op = NULL;
366 struct {
367 __u32 version;
368 __u32 magic;
369 __u64 tag;
370 } head;
371 int total = ret = iov_iter_count(iter);
372 int downcall_size = sizeof(struct orangefs_downcall_s);
373 int head_size = sizeof(head);
374
375 gossip_debug(GOSSIP_DEV_DEBUG, "%s: total:%d: ret:%zd:\n",
376 __func__,
377 total,
378 ret);
379
380 if (total < MAX_DEV_REQ_DOWNSIZE) {
381 gossip_err("%s: total:%d: must be at least:%u:\n",
382 __func__,
383 total,
384 (unsigned int) MAX_DEV_REQ_DOWNSIZE);
385 return -EFAULT;
386 }
387
388 if (!copy_from_iter_full(&head, head_size, iter)) {
389 gossip_err("%s: failed to copy head.\n", __func__);
390 return -EFAULT;
391 }
392
393 if (head.version < ORANGEFS_MINIMUM_USERSPACE_VERSION) {
394 gossip_err("%s: userspace claims version"
395 "%d, minimum version required: %d.\n",
396 __func__,
397 head.version,
398 ORANGEFS_MINIMUM_USERSPACE_VERSION);
399 return -EPROTO;
400 }
401
402 if (head.magic != ORANGEFS_DEVREQ_MAGIC) {
403 gossip_err("Error: Device magic number does not match.\n");
404 return -EPROTO;
405 }
406
407 if (!orangefs_userspace_version) {
408 orangefs_userspace_version = head.version;
409 } else if (orangefs_userspace_version != head.version) {
410 gossip_err("Error: userspace version changes\n");
411 return -EPROTO;
412 }
413
414 /* remove the op from the in progress hash table */
415 op = orangefs_devreq_remove_op(head.tag);
416 if (!op) {
417 gossip_debug(GOSSIP_DEV_DEBUG,
418 "%s: No one's waiting for tag %llu\n",
419 __func__, llu(head.tag));
420 return ret;
421 }
422
423 if (!copy_from_iter_full(&op->downcall, downcall_size, iter)) {
424 gossip_err("%s: failed to copy downcall.\n", __func__);
425 goto Efault;
426 }
427
428 if (op->downcall.status)
429 goto wakeup;
430
431 /*
432 * We've successfully peeled off the head and the downcall.
433 * Something has gone awry if total doesn't equal the
434 * sum of head_size, downcall_size and trailer_size.
435 */
436 if ((head_size + downcall_size + op->downcall.trailer_size) != total) {
437 gossip_err("%s: funky write, head_size:%d"
438 ": downcall_size:%d: trailer_size:%lld"
439 ": total size:%d:\n",
440 __func__,
441 head_size,
442 downcall_size,
443 op->downcall.trailer_size,
444 total);
445 goto Efault;
446 }
447
448 /* Only READDIR operations should have trailers. */
449 if ((op->downcall.type != ORANGEFS_VFS_OP_READDIR) &&
450 (op->downcall.trailer_size != 0)) {
451 gossip_err("%s: %x operation with trailer.",
452 __func__,
453 op->downcall.type);
454 goto Efault;
455 }
456
457 /* READDIR operations should always have trailers. */
458 if ((op->downcall.type == ORANGEFS_VFS_OP_READDIR) &&
459 (op->downcall.trailer_size == 0)) {
460 gossip_err("%s: %x operation with no trailer.",
461 __func__,
462 op->downcall.type);
463 goto Efault;
464 }
465
466 if (op->downcall.type != ORANGEFS_VFS_OP_READDIR)
467 goto wakeup;
468
469 op->downcall.trailer_buf = vzalloc(op->downcall.trailer_size);
470 if (!op->downcall.trailer_buf)
471 goto Enomem;
472
473 if (!copy_from_iter_full(op->downcall.trailer_buf,
474 op->downcall.trailer_size, iter)) {
475 gossip_err("%s: failed to copy trailer.\n", __func__);
476 vfree(op->downcall.trailer_buf);
477 op->downcall.trailer_buf = NULL;
478 goto Efault;
479 }
480
481wakeup:
482 /*
483 * Return to vfs waitqueue, and back to service_operation
484 * through wait_for_matching_downcall.
485 */
486 spin_lock(&op->lock);
487 if (unlikely(op_is_cancel(op))) {
488 spin_unlock(&op->lock);
489 put_cancel(op);
490 } else if (unlikely(op_state_given_up(op))) {
491 spin_unlock(&op->lock);
492 complete(&op->waitq);
493 } else {
494 set_op_state_serviced(op);
495 gossip_debug(GOSSIP_DEV_DEBUG,
496 "%s: op:%s: op_state:%d: process:%s:\n",
497 __func__,
498 get_opname_string(op),
499 op->op_state,
500 current->comm);
501 spin_unlock(&op->lock);
502 }
503 return ret;
504
505Efault:
506 op->downcall.status = -(ORANGEFS_ERROR_BIT | 9);
507 ret = -EFAULT;
508 goto wakeup;
509
510Enomem:
511 op->downcall.status = -(ORANGEFS_ERROR_BIT | 8);
512 ret = -ENOMEM;
513 goto wakeup;
514}
515
516/*
517 * NOTE: gets called when the last reference to this device is dropped.
518 * Using the open_access_count variable, we enforce a reference count
519 * on this file so that it can be opened by only one process at a time.
520 * the devreq_mutex is used to make sure all i/o has completed
521 * before we call orangefs_bufmap_finalize, and similar such tricky
522 * situations
523 */
524static int orangefs_devreq_release(struct inode *inode, struct file *file)
525{
526 int unmounted = 0;
527
528 gossip_debug(GOSSIP_DEV_DEBUG,
529 "%s:pvfs2-client-core: exiting, closing device\n",
530 __func__);
531
532 mutex_lock(&devreq_mutex);
533 orangefs_bufmap_finalize();
534
535 open_access_count = -1;
536
537 unmounted = mark_all_pending_mounts();
538 gossip_debug(GOSSIP_DEV_DEBUG, "ORANGEFS Device Close: Filesystem(s) %s\n",
539 (unmounted ? "UNMOUNTED" : "MOUNTED"));
540
541 purge_waiting_ops();
542 purge_inprogress_ops();
543
544 orangefs_bufmap_run_down();
545
546 gossip_debug(GOSSIP_DEV_DEBUG,
547 "pvfs2-client-core: device close complete\n");
548 open_access_count = 0;
549 orangefs_userspace_version = 0;
550 mutex_unlock(&devreq_mutex);
551 return 0;
552}
553
554int is_daemon_in_service(void)
555{
556 int in_service;
557
558 /*
559 * What this function does is checks if client-core is alive
560 * based on the access count we maintain on the device.
561 */
562 mutex_lock(&devreq_mutex);
563 in_service = open_access_count == 1 ? 0 : -EIO;
564 mutex_unlock(&devreq_mutex);
565 return in_service;
566}
567
568bool __is_daemon_in_service(void)
569{
570 return open_access_count == 1;
571}
572
573static inline long check_ioctl_command(unsigned int command)
574{
575 /* Check for valid ioctl codes */
576 if (_IOC_TYPE(command) != ORANGEFS_DEV_MAGIC) {
577 gossip_err("device ioctl magic numbers don't match! Did you rebuild pvfs2-client-core/libpvfs2? [cmd %x, magic %x != %x]\n",
578 command,
579 _IOC_TYPE(command),
580 ORANGEFS_DEV_MAGIC);
581 return -EINVAL;
582 }
583 /* and valid ioctl commands */
584 if (_IOC_NR(command) >= ORANGEFS_DEV_MAXNR || _IOC_NR(command) <= 0) {
585 gossip_err("Invalid ioctl command number [%d >= %d]\n",
586 _IOC_NR(command), ORANGEFS_DEV_MAXNR);
587 return -ENOIOCTLCMD;
588 }
589 return 0;
590}
591
592static long dispatch_ioctl_command(unsigned int command, unsigned long arg)
593{
594 static __s32 magic = ORANGEFS_DEVREQ_MAGIC;
595 static __s32 max_up_size = MAX_DEV_REQ_UPSIZE;
596 static __s32 max_down_size = MAX_DEV_REQ_DOWNSIZE;
597 struct ORANGEFS_dev_map_desc user_desc;
598 int ret = 0;
599 int upstream_kmod = 1;
600 struct orangefs_sb_info_s *orangefs_sb;
601
602 /* mtmoore: add locking here */
603
604 switch (command) {
605 case ORANGEFS_DEV_GET_MAGIC:
606 return ((put_user(magic, (__s32 __user *) arg) == -EFAULT) ?
607 -EIO :
608 0);
609 case ORANGEFS_DEV_GET_MAX_UPSIZE:
610 return ((put_user(max_up_size,
611 (__s32 __user *) arg) == -EFAULT) ?
612 -EIO :
613 0);
614 case ORANGEFS_DEV_GET_MAX_DOWNSIZE:
615 return ((put_user(max_down_size,
616 (__s32 __user *) arg) == -EFAULT) ?
617 -EIO :
618 0);
619 case ORANGEFS_DEV_MAP:
620 ret = copy_from_user(&user_desc,
621 (struct ORANGEFS_dev_map_desc __user *)
622 arg,
623 sizeof(struct ORANGEFS_dev_map_desc));
624 /* WTF -EIO and not -EFAULT? */
625 return ret ? -EIO : orangefs_bufmap_initialize(&user_desc);
626 case ORANGEFS_DEV_REMOUNT_ALL:
627 gossip_debug(GOSSIP_DEV_DEBUG,
628 "%s: got ORANGEFS_DEV_REMOUNT_ALL\n",
629 __func__);
630
631 /*
632 * remount all mounted orangefs volumes to regain the lost
633 * dynamic mount tables (if any) -- NOTE: this is done
634 * without keeping the superblock list locked due to the
635 * upcall/downcall waiting. also, the request mutex is
636 * used to ensure that no operations will be serviced until
637 * all of the remounts are serviced (to avoid ops between
638 * mounts to fail)
639 */
640 ret = mutex_lock_interruptible(&orangefs_request_mutex);
641 if (ret < 0)
642 return ret;
643 gossip_debug(GOSSIP_DEV_DEBUG,
644 "%s: priority remount in progress\n",
645 __func__);
646 spin_lock(&orangefs_superblocks_lock);
647 list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) {
648 /*
649 * We have to drop the spinlock, so entries can be
650 * removed. They can't be freed, though, so we just
651 * keep the forward pointers and zero the back ones -
652 * that way we can get to the rest of the list.
653 */
654 if (!orangefs_sb->list.prev)
655 continue;
656 gossip_debug(GOSSIP_DEV_DEBUG,
657 "%s: Remounting SB %p\n",
658 __func__,
659 orangefs_sb);
660
661 spin_unlock(&orangefs_superblocks_lock);
662 ret = orangefs_remount(orangefs_sb);
663 spin_lock(&orangefs_superblocks_lock);
664 if (ret) {
665 gossip_debug(GOSSIP_DEV_DEBUG,
666 "SB %p remount failed\n",
667 orangefs_sb);
668 break;
669 }
670 }
671 spin_unlock(&orangefs_superblocks_lock);
672 gossip_debug(GOSSIP_DEV_DEBUG,
673 "%s: priority remount complete\n",
674 __func__);
675 mutex_unlock(&orangefs_request_mutex);
676 return ret;
677
678 case ORANGEFS_DEV_UPSTREAM:
679 ret = copy_to_user((void __user *)arg,
680 &upstream_kmod,
681 sizeof(upstream_kmod));
682
683 if (ret != 0)
684 return -EIO;
685 else
686 return ret;
687
688 case ORANGEFS_DEV_CLIENT_MASK:
689 return orangefs_debugfs_new_client_mask((void __user *)arg);
690 case ORANGEFS_DEV_CLIENT_STRING:
691 return orangefs_debugfs_new_client_string((void __user *)arg);
692 case ORANGEFS_DEV_DEBUG:
693 return orangefs_debugfs_new_debug((void __user *)arg);
694 default:
695 return -ENOIOCTLCMD;
696 }
697 return -ENOIOCTLCMD;
698}
699
700static long orangefs_devreq_ioctl(struct file *file,
701 unsigned int command, unsigned long arg)
702{
703 long ret;
704
705 /* Check for properly constructed commands */
706 ret = check_ioctl_command(command);
707 if (ret < 0)
708 return (int)ret;
709
710 return (int)dispatch_ioctl_command(command, arg);
711}
712
713#ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */
714
715/* Compat structure for the ORANGEFS_DEV_MAP ioctl */
716struct ORANGEFS_dev_map_desc32 {
717 compat_uptr_t ptr;
718 __s32 total_size;
719 __s32 size;
720 __s32 count;
721};
722
723/*
724 * 32 bit user-space apps' ioctl handlers when kernel modules
725 * is compiled as a 64 bit one
726 */
727static long orangefs_devreq_compat_ioctl(struct file *filp, unsigned int cmd,
728 unsigned long args)
729{
730 long ret;
731
732 /* Check for properly constructed commands */
733 ret = check_ioctl_command(cmd);
734 if (ret < 0)
735 return ret;
736 if (cmd == ORANGEFS_DEV_MAP) {
737 struct ORANGEFS_dev_map_desc desc;
738 struct ORANGEFS_dev_map_desc32 d32;
739
740 if (copy_from_user(&d32, (void __user *)args, sizeof(d32)))
741 return -EFAULT;
742
743 desc.ptr = compat_ptr(d32.ptr);
744 desc.total_size = d32.total_size;
745 desc.size = d32.size;
746 desc.count = d32.count;
747 return orangefs_bufmap_initialize(&desc);
748 }
749 /* no other ioctl requires translation */
750 return dispatch_ioctl_command(cmd, args);
751}
752
753#endif /* CONFIG_COMPAT is in .config */
754
755static __poll_t orangefs_devreq_poll(struct file *file,
756 struct poll_table_struct *poll_table)
757{
758 __poll_t poll_revent_mask = 0;
759
760 poll_wait(file, &orangefs_request_list_waitq, poll_table);
761
762 if (!list_empty(&orangefs_request_list))
763 poll_revent_mask |= EPOLLIN;
764 return poll_revent_mask;
765}
766
767/* the assigned character device major number */
768static int orangefs_dev_major;
769
770static const struct file_operations orangefs_devreq_file_operations = {
771 .owner = THIS_MODULE,
772 .read = orangefs_devreq_read,
773 .write_iter = orangefs_devreq_write_iter,
774 .open = orangefs_devreq_open,
775 .release = orangefs_devreq_release,
776 .unlocked_ioctl = orangefs_devreq_ioctl,
777
778#ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */
779 .compat_ioctl = orangefs_devreq_compat_ioctl,
780#endif
781 .poll = orangefs_devreq_poll
782};
783
784/*
785 * Initialize orangefs device specific state:
786 * Must be called at module load time only
787 */
788int orangefs_dev_init(void)
789{
790 /* register orangefs-req device */
791 orangefs_dev_major = register_chrdev(0,
792 ORANGEFS_REQDEVICE_NAME,
793 &orangefs_devreq_file_operations);
794 if (orangefs_dev_major < 0) {
795 gossip_debug(GOSSIP_DEV_DEBUG,
796 "Failed to register /dev/%s (error %d)\n",
797 ORANGEFS_REQDEVICE_NAME, orangefs_dev_major);
798 return orangefs_dev_major;
799 }
800
801 gossip_debug(GOSSIP_DEV_DEBUG,
802 "*** /dev/%s character device registered ***\n",
803 ORANGEFS_REQDEVICE_NAME);
804 gossip_debug(GOSSIP_DEV_DEBUG, "'mknod /dev/%s c %d 0'.\n",
805 ORANGEFS_REQDEVICE_NAME, orangefs_dev_major);
806 return 0;
807}
808
809void orangefs_dev_cleanup(void)
810{
811 unregister_chrdev(orangefs_dev_major, ORANGEFS_REQDEVICE_NAME);
812 gossip_debug(GOSSIP_DEV_DEBUG,
813 "*** /dev/%s character device unregistered ***\n",
814 ORANGEFS_REQDEVICE_NAME);
815}