]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - fs/orangefs/devorangefs-req.c
Reset TreeId to zero on SMB2 TREE_CONNECT
[mirror_ubuntu-zesty-kernel.git] / fs / orangefs / devorangefs-req.c
CommitLineData
5db11c21
MM
1/*
2 * (C) 2001 Clemson University and The University of Chicago
3 *
4 * Changes by Acxiom Corporation to add protocol version to kernel
5 * communication, Copyright Acxiom Corporation, 2005.
6 *
7 * See COPYING in top-level directory.
8 */
9
10#include "protocol.h"
575e9461
MM
11#include "orangefs-kernel.h"
12#include "orangefs-dev-proto.h"
13#include "orangefs-bufmap.h"
44f46410 14#include "orangefs-debugfs.h"
5db11c21
MM
15
16#include <linux/debugfs.h>
17#include <linux/slab.h>
18
19/* this file implements the /dev/pvfs2-req device node */
20
482664dd 21uint32_t orangefs_userspace_version;
f2ee3b75 22
5db11c21
MM
23static int open_access_count;
24
a0fe0515
MB
25static DEFINE_MUTEX(devreq_mutex);
26
5db11c21
MM
27#define DUMP_DEVICE_ERROR() \
28do { \
29 gossip_err("*****************************************************\n");\
8bb8aefd 30 gossip_err("ORANGEFS Device Error: You cannot open the device file "); \
5db11c21 31 gossip_err("\n/dev/%s more than once. Please make sure that\nthere " \
8bb8aefd 32 "are no ", ORANGEFS_REQDEVICE_NAME); \
5db11c21
MM
33 gossip_err("instances of a program using this device\ncurrently " \
34 "running. (You must verify this!)\n"); \
35 gossip_err("For example, you can use the lsof program as follows:\n");\
36 gossip_err("'lsof | grep %s' (run this as root)\n", \
8bb8aefd 37 ORANGEFS_REQDEVICE_NAME); \
5db11c21
MM
38 gossip_err(" open_access_count = %d\n", open_access_count); \
39 gossip_err("*****************************************************\n");\
40} while (0)
41
42static int hash_func(__u64 tag, int table_size)
43{
2c590d5f 44 return do_div(tag, (unsigned int)table_size);
5db11c21
MM
45}
46
8bb8aefd 47static void orangefs_devreq_add_op(struct orangefs_kernel_op_s *op)
5db11c21
MM
48{
49 int index = hash_func(op->tag, hash_table_size);
50
1d503617 51 list_add_tail(&op->list, &orangefs_htable_ops_in_progress[index]);
5db11c21
MM
52}
53
ca9f518e
MM
54/*
55 * find the op with this tag and remove it from the in progress
56 * hash table.
57 */
8bb8aefd 58static struct orangefs_kernel_op_s *orangefs_devreq_remove_op(__u64 tag)
5db11c21 59{
8bb8aefd 60 struct orangefs_kernel_op_s *op, *next;
5db11c21
MM
61 int index;
62
63 index = hash_func(tag, hash_table_size);
64
1d503617 65 spin_lock(&orangefs_htable_ops_in_progress_lock);
5db11c21
MM
66 list_for_each_entry_safe(op,
67 next,
1d503617 68 &orangefs_htable_ops_in_progress[index],
5db11c21 69 list) {
05a50a5b
AV
70 if (op->tag == tag && !op_state_purged(op) &&
71 !op_state_given_up(op)) {
ed42fe05 72 list_del_init(&op->list);
1d503617 73 spin_unlock(&orangefs_htable_ops_in_progress_lock);
5db11c21
MM
74 return op;
75 }
76 }
77
1d503617 78 spin_unlock(&orangefs_htable_ops_in_progress_lock);
5db11c21
MM
79 return NULL;
80}
81
acfcbaf1
MB
82/* Returns whether any FS are still pending remounted */
83static int mark_all_pending_mounts(void)
84{
85 int unmounted = 1;
86 struct orangefs_sb_info_s *orangefs_sb = NULL;
87
88 spin_lock(&orangefs_superblocks_lock);
89 list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) {
90 /* All of these file system require a remount */
91 orangefs_sb->mount_pending = 1;
92 unmounted = 0;
93 }
94 spin_unlock(&orangefs_superblocks_lock);
95 return unmounted;
96}
97
98/*
99 * Determine if a given file system needs to be remounted or not
100 * Returns -1 on error
101 * 0 if already mounted
102 * 1 if needs remount
103 */
104static int fs_mount_pending(__s32 fsid)
105{
106 int mount_pending = -1;
107 struct orangefs_sb_info_s *orangefs_sb = NULL;
108
109 spin_lock(&orangefs_superblocks_lock);
110 list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) {
111 if (orangefs_sb->fs_id == fsid) {
112 mount_pending = orangefs_sb->mount_pending;
113 break;
114 }
115 }
116 spin_unlock(&orangefs_superblocks_lock);
117 return mount_pending;
118}
119
8bb8aefd 120static int orangefs_devreq_open(struct inode *inode, struct file *file)
5db11c21
MM
121{
122 int ret = -EINVAL;
123
78fee0b6
JH
124 /* in order to ensure that the filesystem driver sees correct UIDs */
125 if (file->f_cred->user_ns != &init_user_ns) {
126 gossip_err("%s: device cannot be opened outside init_user_ns\n",
127 __func__);
128 goto out;
129 }
130
5db11c21 131 if (!(file->f_flags & O_NONBLOCK)) {
97f10027
MM
132 gossip_err("%s: device cannot be opened in blocking mode\n",
133 __func__);
5db11c21
MM
134 goto out;
135 }
136 ret = -EACCES;
97f10027 137 gossip_debug(GOSSIP_DEV_DEBUG, "client-core: opening device\n");
5db11c21
MM
138 mutex_lock(&devreq_mutex);
139
140 if (open_access_count == 0) {
fee25ce1 141 open_access_count = 1;
fb6d2526 142 ret = 0;
5db11c21
MM
143 } else {
144 DUMP_DEVICE_ERROR();
145 }
146 mutex_unlock(&devreq_mutex);
147
148out:
149
150 gossip_debug(GOSSIP_DEV_DEBUG,
151 "pvfs2-client-core: open device complete (ret = %d)\n",
152 ret);
153 return ret;
154}
155
97f10027 156/* Function for read() callers into the device */
8bb8aefd 157static ssize_t orangefs_devreq_read(struct file *file,
5db11c21
MM
158 char __user *buf,
159 size_t count, loff_t *offset)
160{
8bb8aefd
YL
161 struct orangefs_kernel_op_s *op, *temp;
162 __s32 proto_ver = ORANGEFS_KERNEL_PROTO_VERSION;
163 static __s32 magic = ORANGEFS_DEVREQ_MAGIC;
164 struct orangefs_kernel_op_s *cur_op = NULL;
24c8d080 165 unsigned long ret;
5db11c21 166
24c8d080 167 /* We do not support blocking IO. */
5db11c21 168 if (!(file->f_flags & O_NONBLOCK)) {
97f10027
MM
169 gossip_err("%s: blocking read from client-core.\n",
170 __func__);
5db11c21 171 return -EINVAL;
24c8d080
MB
172 }
173
174 /*
a762ae6d 175 * The client will do an ioctl to find MAX_DEV_REQ_UPSIZE, then
24c8d080
MB
176 * always read with that size buffer.
177 */
a762ae6d 178 if (count != MAX_DEV_REQ_UPSIZE) {
24c8d080
MB
179 gossip_err("orangefs: client-core tried to read wrong size\n");
180 return -EINVAL;
181 }
182
ed42fe05 183restart:
24c8d080 184 /* Get next op (if any) from top of list. */
8bb8aefd
YL
185 spin_lock(&orangefs_request_list_lock);
186 list_for_each_entry_safe(op, temp, &orangefs_request_list, list) {
24c8d080
MB
187 __s32 fsid;
188 /* This lock is held past the end of the loop when we break. */
189 spin_lock(&op->lock);
05a50a5b 190 if (unlikely(op_state_purged(op) || op_state_given_up(op))) {
ed42fe05
AV
191 spin_unlock(&op->lock);
192 continue;
193 }
24c8d080
MB
194
195 fsid = fsid_of_op(op);
8bb8aefd 196 if (fsid != ORANGEFS_FS_ID_NULL) {
24c8d080
MB
197 int ret;
198 /* Skip ops whose filesystem needs to be mounted. */
199 ret = fs_mount_pending(fsid);
200 if (ret == 1) {
5db11c21 201 gossip_debug(GOSSIP_DEV_DEBUG,
5090c967
MM
202 "%s: mount pending, skipping op tag "
203 "%llu %s\n",
204 __func__,
205 llu(op->tag),
206 get_opname_string(op));
24c8d080
MB
207 spin_unlock(&op->lock);
208 continue;
97f10027
MM
209 /*
210 * Skip ops whose filesystem we don't know about unless
211 * it is being mounted.
212 */
24c8d080
MB
213 /* XXX: is there a better way to detect this? */
214 } else if (ret == -1 &&
97f10027
MM
215 !(op->upcall.type ==
216 ORANGEFS_VFS_OP_FS_MOUNT ||
217 op->upcall.type ==
218 ORANGEFS_VFS_OP_GETATTR)) {
24c8d080
MB
219 gossip_debug(GOSSIP_DEV_DEBUG,
220 "orangefs: skipping op tag %llu %s\n",
221 llu(op->tag), get_opname_string(op));
222 gossip_err(
223 "orangefs: ERROR: fs_mount_pending %d\n",
224 fsid);
225 spin_unlock(&op->lock);
5db11c21 226 continue;
5db11c21
MM
227 }
228 }
24c8d080
MB
229 /*
230 * Either this op does not pertain to a filesystem, is mounting
231 * a filesystem, or pertains to a mounted filesystem. Let it
232 * through.
233 */
234 cur_op = op;
235 break;
236 }
237
238 /*
239 * At this point we either have a valid op and can continue or have not
240 * found an op and must ask the client to try again later.
241 */
242 if (!cur_op) {
8bb8aefd 243 spin_unlock(&orangefs_request_list_lock);
24c8d080 244 return -EAGAIN;
5db11c21
MM
245 }
246
ca9f518e
MM
247 gossip_debug(GOSSIP_DEV_DEBUG, "%s: reading op tag %llu %s\n",
248 __func__,
249 llu(cur_op->tag),
250 get_opname_string(cur_op));
5db11c21 251
24c8d080
MB
252 /*
253 * Such an op should never be on the list in the first place. If so, we
254 * will abort.
255 */
256 if (op_state_in_progress(cur_op) || op_state_serviced(cur_op)) {
257 gossip_err("orangefs: ERROR: Current op already queued.\n");
05a50a5b 258 list_del_init(&cur_op->list);
5db11c21 259 spin_unlock(&cur_op->lock);
8bb8aefd 260 spin_unlock(&orangefs_request_list_lock);
24c8d080 261 return -EAGAIN;
5db11c21 262 }
ca9f518e 263
ed42fe05 264 list_del_init(&cur_op->list);
8bb8aefd 265 spin_unlock(&orangefs_request_list_lock);
ed42fe05 266
24c8d080
MB
267 spin_unlock(&cur_op->lock);
268
269 /* Push the upcall out. */
270 ret = copy_to_user(buf, &proto_ver, sizeof(__s32));
271 if (ret != 0)
272 goto error;
273 ret = copy_to_user(buf+sizeof(__s32), &magic, sizeof(__s32));
274 if (ret != 0)
275 goto error;
276 ret = copy_to_user(buf+2 * sizeof(__s32), &cur_op->tag, sizeof(__u64));
277 if (ret != 0)
278 goto error;
279 ret = copy_to_user(buf+2*sizeof(__s32)+sizeof(__u64), &cur_op->upcall,
8bb8aefd 280 sizeof(struct orangefs_upcall_s));
24c8d080
MB
281 if (ret != 0)
282 goto error;
283
1d503617 284 spin_lock(&orangefs_htable_ops_in_progress_lock);
ed42fe05
AV
285 spin_lock(&cur_op->lock);
286 if (unlikely(op_state_given_up(cur_op))) {
287 spin_unlock(&cur_op->lock);
1d503617 288 spin_unlock(&orangefs_htable_ops_in_progress_lock);
05a50a5b 289 complete(&cur_op->waitq);
ed42fe05
AV
290 goto restart;
291 }
292
293 /*
294 * Set the operation to be in progress and move it between lists since
295 * it has been sent to the client.
296 */
297 set_op_state_inprogress(cur_op);
9d9e7ba9
MM
298 gossip_debug(GOSSIP_DEV_DEBUG,
299 "%s: 1 op:%s: op_state:%d: process:%s:\n",
300 __func__,
301 get_opname_string(cur_op),
302 cur_op->op_state,
303 current->comm);
ed42fe05
AV
304 orangefs_devreq_add_op(cur_op);
305 spin_unlock(&cur_op->lock);
1d503617 306 spin_unlock(&orangefs_htable_ops_in_progress_lock);
ed42fe05 307
24c8d080 308 /* The client only asks to read one size buffer. */
a762ae6d 309 return MAX_DEV_REQ_UPSIZE;
24c8d080
MB
310error:
311 /*
312 * We were unable to copy the op data to the client. Put the op back in
313 * list. If client has crashed, the op will be purged later when the
314 * device is released.
315 */
316 gossip_err("orangefs: Failed to copy data to user space\n");
8bb8aefd 317 spin_lock(&orangefs_request_list_lock);
24c8d080 318 spin_lock(&cur_op->lock);
ed42fe05
AV
319 if (likely(!op_state_given_up(cur_op))) {
320 set_op_state_waiting(cur_op);
9d9e7ba9
MM
321 gossip_debug(GOSSIP_DEV_DEBUG,
322 "%s: 2 op:%s: op_state:%d: process:%s:\n",
323 __func__,
324 get_opname_string(cur_op),
325 cur_op->op_state,
326 current->comm);
ed42fe05 327 list_add(&cur_op->list, &orangefs_request_list);
05a50a5b
AV
328 spin_unlock(&cur_op->lock);
329 } else {
330 spin_unlock(&cur_op->lock);
331 complete(&cur_op->waitq);
ed42fe05 332 }
8bb8aefd 333 spin_unlock(&orangefs_request_list_lock);
24c8d080 334 return -EFAULT;
5db11c21
MM
335}
336
97f10027 337/*
b3ae4755
MM
338 * Function for writev() callers into the device.
339 *
340 * Userspace should have written:
341 * - __u32 version
342 * - __u32 magic
343 * - __u64 tag
344 * - struct orangefs_downcall_s
345 * - trailer buffer (in the case of READDIR operations)
97f10027 346 */
b3ae4755
MM
347static ssize_t orangefs_devreq_write_iter(struct kiocb *iocb,
348 struct iov_iter *iter)
5db11c21 349{
b3ae4755 350 ssize_t ret;
8bb8aefd 351 struct orangefs_kernel_op_s *op = NULL;
b3ae4755
MM
352 struct {
353 __u32 version;
354 __u32 magic;
355 __u64 tag;
356 } head;
357 int total = ret = iov_iter_count(iter);
b3ae4755
MM
358 int downcall_size = sizeof(struct orangefs_downcall_s);
359 int head_size = sizeof(head);
360
361 gossip_debug(GOSSIP_DEV_DEBUG, "%s: total:%d: ret:%zd:\n",
362 __func__,
363 total,
364 ret);
5db11c21 365
b3ae4755 366 if (total < MAX_DEV_REQ_DOWNSIZE) {
cf0c2771 367 gossip_err("%s: total:%d: must be at least:%u:\n",
b3ae4755
MM
368 __func__,
369 total,
cf0c2771 370 (unsigned int) MAX_DEV_REQ_DOWNSIZE);
ed42fe05 371 return -EFAULT;
5db11c21 372 }
b3ae4755 373
cbbd26b8 374 if (!copy_from_iter_full(&head, head_size, iter)) {
b3ae4755 375 gossip_err("%s: failed to copy head.\n", __func__);
ed42fe05 376 return -EFAULT;
97f10027 377 }
b3ae4755
MM
378
379 if (head.version < ORANGEFS_MINIMUM_USERSPACE_VERSION) {
380 gossip_err("%s: userspace claims version"
381 "%d, minimum version required: %d.\n",
382 __func__,
383 head.version,
384 ORANGEFS_MINIMUM_USERSPACE_VERSION);
ed42fe05 385 return -EPROTO;
5db11c21 386 }
5db11c21 387
b3ae4755
MM
388 if (head.magic != ORANGEFS_DEVREQ_MAGIC) {
389 gossip_err("Error: Device magic number does not match.\n");
ed42fe05 390 return -EPROTO;
b3ae4755 391 }
5db11c21 392
482664dd
MB
393 if (!orangefs_userspace_version) {
394 orangefs_userspace_version = head.version;
395 } else if (orangefs_userspace_version != head.version) {
f2ee3b75
MB
396 gossip_err("Error: userspace version changes\n");
397 return -EPROTO;
398 }
399
ca9f518e 400 /* remove the op from the in progress hash table */
b3ae4755
MM
401 op = orangefs_devreq_remove_op(head.tag);
402 if (!op) {
403 gossip_err("WARNING: No one's waiting for tag %llu\n",
404 llu(head.tag));
ed42fe05 405 return ret;
b3ae4755 406 }
5db11c21 407
cbbd26b8 408 if (!copy_from_iter_full(&op->downcall, downcall_size, iter)) {
b3ae4755 409 gossip_err("%s: failed to copy downcall.\n", __func__);
5964c1b8 410 goto Efault;
5db11c21
MM
411 }
412
b3ae4755
MM
413 if (op->downcall.status)
414 goto wakeup;
97f10027 415
b3ae4755
MM
416 /*
417 * We've successfully peeled off the head and the downcall.
418 * Something has gone awry if total doesn't equal the
419 * sum of head_size, downcall_size and trailer_size.
420 */
421 if ((head_size + downcall_size + op->downcall.trailer_size) != total) {
422 gossip_err("%s: funky write, head_size:%d"
423 ": downcall_size:%d: trailer_size:%lld"
424 ": total size:%d:\n",
425 __func__,
426 head_size,
427 downcall_size,
428 op->downcall.trailer_size,
429 total);
5964c1b8 430 goto Efault;
b3ae4755 431 }
97f10027 432
b3ae4755
MM
433 /* Only READDIR operations should have trailers. */
434 if ((op->downcall.type != ORANGEFS_VFS_OP_READDIR) &&
435 (op->downcall.trailer_size != 0)) {
436 gossip_err("%s: %x operation with trailer.",
437 __func__,
438 op->downcall.type);
5964c1b8 439 goto Efault;
b3ae4755 440 }
97f10027 441
b3ae4755
MM
442 /* READDIR operations should always have trailers. */
443 if ((op->downcall.type == ORANGEFS_VFS_OP_READDIR) &&
444 (op->downcall.trailer_size == 0)) {
445 gossip_err("%s: %x operation with no trailer.",
446 __func__,
447 op->downcall.type);
5964c1b8 448 goto Efault;
b3ae4755 449 }
97f10027 450
b3ae4755
MM
451 if (op->downcall.type != ORANGEFS_VFS_OP_READDIR)
452 goto wakeup;
5db11c21 453
b3ae4755
MM
454 op->downcall.trailer_buf =
455 vmalloc(op->downcall.trailer_size);
456 if (op->downcall.trailer_buf == NULL) {
457 gossip_err("%s: failed trailer vmalloc.\n",
458 __func__);
5964c1b8 459 goto Enomem;
b3ae4755
MM
460 }
461 memset(op->downcall.trailer_buf, 0, op->downcall.trailer_size);
cbbd26b8
AV
462 if (!copy_from_iter_full(op->downcall.trailer_buf,
463 op->downcall.trailer_size, iter)) {
b3ae4755
MM
464 gossip_err("%s: failed to copy trailer.\n", __func__);
465 vfree(op->downcall.trailer_buf);
5964c1b8 466 goto Efault;
b3ae4755 467 }
97f10027 468
b3ae4755 469wakeup:
2a9e5c22 470 /*
9f08cfe9
MM
471 * Return to vfs waitqueue, and back to service_operation
472 * through wait_for_matching_downcall.
2a9e5c22
AV
473 */
474 spin_lock(&op->lock);
5964c1b8 475 if (unlikely(op_is_cancel(op))) {
2a9e5c22 476 spin_unlock(&op->lock);
78699e29 477 put_cancel(op);
5964c1b8
AV
478 } else if (unlikely(op_state_given_up(op))) {
479 spin_unlock(&op->lock);
05a50a5b 480 complete(&op->waitq);
5964c1b8
AV
481 } else {
482 set_op_state_serviced(op);
9d9e7ba9
MM
483 gossip_debug(GOSSIP_DEV_DEBUG,
484 "%s: op:%s: op_state:%d: process:%s:\n",
485 __func__,
486 get_opname_string(op),
487 op->op_state,
488 current->comm);
5964c1b8
AV
489 spin_unlock(&op->lock);
490 }
b3ae4755 491 return ret;
ed42fe05 492
5964c1b8
AV
493Efault:
494 op->downcall.status = -(ORANGEFS_ERROR_BIT | 9);
495 ret = -EFAULT;
496 goto wakeup;
497
498Enomem:
499 op->downcall.status = -(ORANGEFS_ERROR_BIT | 8);
500 ret = -ENOMEM;
501 goto wakeup;
5db11c21
MM
502}
503
5db11c21
MM
504/*
505 * NOTE: gets called when the last reference to this device is dropped.
506 * Using the open_access_count variable, we enforce a reference count
507 * on this file so that it can be opened by only one process at a time.
508 * the devreq_mutex is used to make sure all i/o has completed
8bb8aefd 509 * before we call orangefs_bufmap_finalize, and similar such tricky
5db11c21
MM
510 * situations
511 */
8bb8aefd 512static int orangefs_devreq_release(struct inode *inode, struct file *file)
5db11c21
MM
513{
514 int unmounted = 0;
515
516 gossip_debug(GOSSIP_DEV_DEBUG,
517 "%s:pvfs2-client-core: exiting, closing device\n",
518 __func__);
519
520 mutex_lock(&devreq_mutex);
ea2c9c9f 521 orangefs_bufmap_finalize();
5db11c21 522
fee25ce1 523 open_access_count = -1;
5db11c21
MM
524
525 unmounted = mark_all_pending_mounts();
8bb8aefd 526 gossip_debug(GOSSIP_DEV_DEBUG, "ORANGEFS Device Close: Filesystem(s) %s\n",
5db11c21 527 (unmounted ? "UNMOUNTED" : "MOUNTED"));
5db11c21 528
5db11c21 529 purge_waiting_ops();
5db11c21 530 purge_inprogress_ops();
ea2c9c9f
AV
531
532 orangefs_bufmap_run_down();
533
5db11c21
MM
534 gossip_debug(GOSSIP_DEV_DEBUG,
535 "pvfs2-client-core: device close complete\n");
fee25ce1 536 open_access_count = 0;
482664dd 537 orangefs_userspace_version = 0;
fee25ce1 538 mutex_unlock(&devreq_mutex);
5db11c21
MM
539 return 0;
540}
541
542int is_daemon_in_service(void)
543{
544 int in_service;
545
546 /*
547 * What this function does is checks if client-core is alive
548 * based on the access count we maintain on the device.
549 */
550 mutex_lock(&devreq_mutex);
551 in_service = open_access_count == 1 ? 0 : -EIO;
552 mutex_unlock(&devreq_mutex);
553 return in_service;
554}
555
78699e29
AV
556bool __is_daemon_in_service(void)
557{
558 return open_access_count == 1;
559}
560
5db11c21
MM
561static inline long check_ioctl_command(unsigned int command)
562{
563 /* Check for valid ioctl codes */
8bb8aefd 564 if (_IOC_TYPE(command) != ORANGEFS_DEV_MAGIC) {
5db11c21
MM
565 gossip_err("device ioctl magic numbers don't match! Did you rebuild pvfs2-client-core/libpvfs2? [cmd %x, magic %x != %x]\n",
566 command,
567 _IOC_TYPE(command),
8bb8aefd 568 ORANGEFS_DEV_MAGIC);
5db11c21
MM
569 return -EINVAL;
570 }
571 /* and valid ioctl commands */
8bb8aefd 572 if (_IOC_NR(command) >= ORANGEFS_DEV_MAXNR || _IOC_NR(command) <= 0) {
5db11c21 573 gossip_err("Invalid ioctl command number [%d >= %d]\n",
8bb8aefd 574 _IOC_NR(command), ORANGEFS_DEV_MAXNR);
5db11c21
MM
575 return -ENOIOCTLCMD;
576 }
577 return 0;
578}
579
580static long dispatch_ioctl_command(unsigned int command, unsigned long arg)
581{
8bb8aefd 582 static __s32 magic = ORANGEFS_DEVREQ_MAGIC;
a762ae6d
MB
583 static __s32 max_up_size = MAX_DEV_REQ_UPSIZE;
584 static __s32 max_down_size = MAX_DEV_REQ_DOWNSIZE;
8bb8aefd 585 struct ORANGEFS_dev_map_desc user_desc;
5db11c21 586 int ret = 0;
5db11c21 587 int upstream_kmod = 1;
45996492 588 struct orangefs_sb_info_s *orangefs_sb;
5db11c21
MM
589
590 /* mtmoore: add locking here */
591
592 switch (command) {
8bb8aefd 593 case ORANGEFS_DEV_GET_MAGIC:
5db11c21
MM
594 return ((put_user(magic, (__s32 __user *) arg) == -EFAULT) ?
595 -EIO :
596 0);
8bb8aefd 597 case ORANGEFS_DEV_GET_MAX_UPSIZE:
5db11c21
MM
598 return ((put_user(max_up_size,
599 (__s32 __user *) arg) == -EFAULT) ?
600 -EIO :
601 0);
8bb8aefd 602 case ORANGEFS_DEV_GET_MAX_DOWNSIZE:
5db11c21
MM
603 return ((put_user(max_down_size,
604 (__s32 __user *) arg) == -EFAULT) ?
605 -EIO :
606 0);
8bb8aefd 607 case ORANGEFS_DEV_MAP:
5db11c21 608 ret = copy_from_user(&user_desc,
8bb8aefd 609 (struct ORANGEFS_dev_map_desc __user *)
5db11c21 610 arg,
8bb8aefd 611 sizeof(struct ORANGEFS_dev_map_desc));
ea2c9c9f
AV
612 /* WTF -EIO and not -EFAULT? */
613 return ret ? -EIO : orangefs_bufmap_initialize(&user_desc);
8bb8aefd 614 case ORANGEFS_DEV_REMOUNT_ALL:
5db11c21 615 gossip_debug(GOSSIP_DEV_DEBUG,
97f10027
MM
616 "%s: got ORANGEFS_DEV_REMOUNT_ALL\n",
617 __func__);
5db11c21
MM
618
619 /*
8bb8aefd 620 * remount all mounted orangefs volumes to regain the lost
5db11c21
MM
621 * dynamic mount tables (if any) -- NOTE: this is done
622 * without keeping the superblock list locked due to the
adcf34a2 623 * upcall/downcall waiting. also, the request mutex is
5db11c21
MM
624 * used to ensure that no operations will be serviced until
625 * all of the remounts are serviced (to avoid ops between
626 * mounts to fail)
627 */
1d503617 628 ret = mutex_lock_interruptible(&orangefs_request_mutex);
5db11c21
MM
629 if (ret < 0)
630 return ret;
631 gossip_debug(GOSSIP_DEV_DEBUG,
97f10027
MM
632 "%s: priority remount in progress\n",
633 __func__);
45996492
AV
634 spin_lock(&orangefs_superblocks_lock);
635 list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) {
636 /*
637 * We have to drop the spinlock, so entries can be
638 * removed. They can't be freed, though, so we just
639 * keep the forward pointers and zero the back ones -
640 * that way we can get to the rest of the list.
641 */
642 if (!orangefs_sb->list.prev)
643 continue;
644 gossip_debug(GOSSIP_DEV_DEBUG,
645 "%s: Remounting SB %p\n",
646 __func__,
647 orangefs_sb);
648
649 spin_unlock(&orangefs_superblocks_lock);
650 ret = orangefs_remount(orangefs_sb);
651 spin_lock(&orangefs_superblocks_lock);
652 if (ret) {
5db11c21 653 gossip_debug(GOSSIP_DEV_DEBUG,
45996492 654 "SB %p remount failed\n",
8bb8aefd 655 orangefs_sb);
45996492 656 break;
5db11c21
MM
657 }
658 }
45996492 659 spin_unlock(&orangefs_superblocks_lock);
5db11c21 660 gossip_debug(GOSSIP_DEV_DEBUG,
97f10027
MM
661 "%s: priority remount complete\n",
662 __func__);
1d503617 663 mutex_unlock(&orangefs_request_mutex);
5db11c21
MM
664 return ret;
665
8bb8aefd 666 case ORANGEFS_DEV_UPSTREAM:
5db11c21
MM
667 ret = copy_to_user((void __user *)arg,
668 &upstream_kmod,
669 sizeof(upstream_kmod));
670
671 if (ret != 0)
672 return -EIO;
673 else
674 return ret;
675
8bb8aefd 676 case ORANGEFS_DEV_CLIENT_MASK:
44f46410 677 return orangefs_debugfs_new_client_mask((void __user *)arg);
8bb8aefd 678 case ORANGEFS_DEV_CLIENT_STRING:
44f46410 679 return orangefs_debugfs_new_client_string((void __user *)arg);
8bb8aefd 680 case ORANGEFS_DEV_DEBUG:
44f46410 681 return orangefs_debugfs_new_debug((void __user *)arg);
5db11c21
MM
682 default:
683 return -ENOIOCTLCMD;
684 }
685 return -ENOIOCTLCMD;
686}
687
8bb8aefd 688static long orangefs_devreq_ioctl(struct file *file,
5db11c21
MM
689 unsigned int command, unsigned long arg)
690{
691 long ret;
692
693 /* Check for properly constructed commands */
694 ret = check_ioctl_command(command);
695 if (ret < 0)
696 return (int)ret;
697
698 return (int)dispatch_ioctl_command(command, arg);
699}
700
701#ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */
702
8bb8aefd
YL
703/* Compat structure for the ORANGEFS_DEV_MAP ioctl */
704struct ORANGEFS_dev_map_desc32 {
5db11c21
MM
705 compat_uptr_t ptr;
706 __s32 total_size;
707 __s32 size;
708 __s32 count;
709};
710
711static unsigned long translate_dev_map26(unsigned long args, long *error)
712{
8bb8aefd 713 struct ORANGEFS_dev_map_desc32 __user *p32 = (void __user *)args;
5db11c21
MM
714 /*
715 * Depending on the architecture, allocate some space on the
716 * user-call-stack based on our expected layout.
717 */
8bb8aefd 718 struct ORANGEFS_dev_map_desc __user *p =
5db11c21 719 compat_alloc_user_space(sizeof(*p));
84d02150 720 compat_uptr_t addr;
5db11c21
MM
721
722 *error = 0;
723 /* get the ptr from the 32 bit user-space */
724 if (get_user(addr, &p32->ptr))
725 goto err;
726 /* try to put that into a 64-bit layout */
727 if (put_user(compat_ptr(addr), &p->ptr))
728 goto err;
729 /* copy the remaining fields */
730 if (copy_in_user(&p->total_size, &p32->total_size, sizeof(__s32)))
731 goto err;
732 if (copy_in_user(&p->size, &p32->size, sizeof(__s32)))
733 goto err;
734 if (copy_in_user(&p->count, &p32->count, sizeof(__s32)))
735 goto err;
736 return (unsigned long)p;
737err:
738 *error = -EFAULT;
739 return 0;
740}
741
742/*
743 * 32 bit user-space apps' ioctl handlers when kernel modules
744 * is compiled as a 64 bit one
745 */
8bb8aefd 746static long orangefs_devreq_compat_ioctl(struct file *filp, unsigned int cmd,
5db11c21
MM
747 unsigned long args)
748{
749 long ret;
750 unsigned long arg = args;
751
752 /* Check for properly constructed commands */
753 ret = check_ioctl_command(cmd);
754 if (ret < 0)
755 return ret;
8bb8aefd 756 if (cmd == ORANGEFS_DEV_MAP) {
5db11c21
MM
757 /*
758 * convert the arguments to what we expect internally
759 * in kernel space
760 */
761 arg = translate_dev_map26(args, &ret);
762 if (ret < 0) {
763 gossip_err("Could not translate dev map\n");
764 return ret;
765 }
766 }
767 /* no other ioctl requires translation */
768 return dispatch_ioctl_command(cmd, arg);
769}
770
2c590d5f
MM
771#endif /* CONFIG_COMPAT is in .config */
772
5db11c21 773/* the assigned character device major number */
8bb8aefd 774static int orangefs_dev_major;
5db11c21
MM
775
776/*
8bb8aefd 777 * Initialize orangefs device specific state:
5db11c21
MM
778 * Must be called at module load time only
779 */
8bb8aefd 780int orangefs_dev_init(void)
5db11c21 781{
8bb8aefd
YL
782 /* register orangefs-req device */
783 orangefs_dev_major = register_chrdev(0,
784 ORANGEFS_REQDEVICE_NAME,
785 &orangefs_devreq_file_operations);
786 if (orangefs_dev_major < 0) {
5db11c21
MM
787 gossip_debug(GOSSIP_DEV_DEBUG,
788 "Failed to register /dev/%s (error %d)\n",
8bb8aefd 789 ORANGEFS_REQDEVICE_NAME, orangefs_dev_major);
8bb8aefd 790 return orangefs_dev_major;
5db11c21
MM
791 }
792
793 gossip_debug(GOSSIP_DEV_DEBUG,
794 "*** /dev/%s character device registered ***\n",
8bb8aefd 795 ORANGEFS_REQDEVICE_NAME);
5db11c21 796 gossip_debug(GOSSIP_DEV_DEBUG, "'mknod /dev/%s c %d 0'.\n",
8bb8aefd 797 ORANGEFS_REQDEVICE_NAME, orangefs_dev_major);
5db11c21
MM
798 return 0;
799}
800
8bb8aefd 801void orangefs_dev_cleanup(void)
5db11c21 802{
8bb8aefd 803 unregister_chrdev(orangefs_dev_major, ORANGEFS_REQDEVICE_NAME);
5db11c21
MM
804 gossip_debug(GOSSIP_DEV_DEBUG,
805 "*** /dev/%s character device unregistered ***\n",
8bb8aefd 806 ORANGEFS_REQDEVICE_NAME);
5db11c21
MM
807}
808
8bb8aefd 809static unsigned int orangefs_devreq_poll(struct file *file,
5db11c21
MM
810 struct poll_table_struct *poll_table)
811{
812 int poll_revent_mask = 0;
813
83595db0 814 poll_wait(file, &orangefs_request_list_waitq, poll_table);
5db11c21 815
83595db0
AV
816 if (!list_empty(&orangefs_request_list))
817 poll_revent_mask |= POLL_IN;
5db11c21
MM
818 return poll_revent_mask;
819}
820
8bb8aefd 821const struct file_operations orangefs_devreq_file_operations = {
5db11c21 822 .owner = THIS_MODULE,
8bb8aefd
YL
823 .read = orangefs_devreq_read,
824 .write_iter = orangefs_devreq_write_iter,
825 .open = orangefs_devreq_open,
826 .release = orangefs_devreq_release,
827 .unlocked_ioctl = orangefs_devreq_ioctl,
5db11c21
MM
828
829#ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */
8bb8aefd 830 .compat_ioctl = orangefs_devreq_compat_ioctl,
5db11c21 831#endif
8bb8aefd 832 .poll = orangefs_devreq_poll
5db11c21 833};