Commit | Line | Data |
---|---|---|
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" | |
5db11c21 MM |
14 | |
15 | #include <linux/debugfs.h> | |
16 | #include <linux/slab.h> | |
17 | ||
18 | /* this file implements the /dev/pvfs2-req device node */ | |
19 | ||
20 | static int open_access_count; | |
21 | ||
22 | #define DUMP_DEVICE_ERROR() \ | |
23 | do { \ | |
24 | gossip_err("*****************************************************\n");\ | |
8bb8aefd | 25 | gossip_err("ORANGEFS Device Error: You cannot open the device file "); \ |
5db11c21 | 26 | gossip_err("\n/dev/%s more than once. Please make sure that\nthere " \ |
8bb8aefd | 27 | "are no ", ORANGEFS_REQDEVICE_NAME); \ |
5db11c21 MM |
28 | gossip_err("instances of a program using this device\ncurrently " \ |
29 | "running. (You must verify this!)\n"); \ | |
30 | gossip_err("For example, you can use the lsof program as follows:\n");\ | |
31 | gossip_err("'lsof | grep %s' (run this as root)\n", \ | |
8bb8aefd | 32 | ORANGEFS_REQDEVICE_NAME); \ |
5db11c21 MM |
33 | gossip_err(" open_access_count = %d\n", open_access_count); \ |
34 | gossip_err("*****************************************************\n");\ | |
35 | } while (0) | |
36 | ||
37 | static int hash_func(__u64 tag, int table_size) | |
38 | { | |
2c590d5f | 39 | return do_div(tag, (unsigned int)table_size); |
5db11c21 MM |
40 | } |
41 | ||
8bb8aefd | 42 | static void orangefs_devreq_add_op(struct orangefs_kernel_op_s *op) |
5db11c21 MM |
43 | { |
44 | int index = hash_func(op->tag, hash_table_size); | |
45 | ||
46 | spin_lock(&htable_ops_in_progress_lock); | |
47 | list_add_tail(&op->list, &htable_ops_in_progress[index]); | |
48 | spin_unlock(&htable_ops_in_progress_lock); | |
49 | } | |
50 | ||
8bb8aefd | 51 | static struct orangefs_kernel_op_s *orangefs_devreq_remove_op(__u64 tag) |
5db11c21 | 52 | { |
8bb8aefd | 53 | struct orangefs_kernel_op_s *op, *next; |
5db11c21 MM |
54 | int index; |
55 | ||
56 | index = hash_func(tag, hash_table_size); | |
57 | ||
58 | spin_lock(&htable_ops_in_progress_lock); | |
59 | list_for_each_entry_safe(op, | |
60 | next, | |
61 | &htable_ops_in_progress[index], | |
62 | list) { | |
63 | if (op->tag == tag) { | |
64 | list_del(&op->list); | |
65 | spin_unlock(&htable_ops_in_progress_lock); | |
66 | return op; | |
67 | } | |
68 | } | |
69 | ||
70 | spin_unlock(&htable_ops_in_progress_lock); | |
71 | return NULL; | |
72 | } | |
73 | ||
8bb8aefd | 74 | static int orangefs_devreq_open(struct inode *inode, struct file *file) |
5db11c21 MM |
75 | { |
76 | int ret = -EINVAL; | |
77 | ||
78 | if (!(file->f_flags & O_NONBLOCK)) { | |
97f10027 MM |
79 | gossip_err("%s: device cannot be opened in blocking mode\n", |
80 | __func__); | |
5db11c21 MM |
81 | goto out; |
82 | } | |
83 | ret = -EACCES; | |
97f10027 | 84 | gossip_debug(GOSSIP_DEV_DEBUG, "client-core: opening device\n"); |
5db11c21 MM |
85 | mutex_lock(&devreq_mutex); |
86 | ||
87 | if (open_access_count == 0) { | |
88 | ret = generic_file_open(inode, file); | |
89 | if (ret == 0) | |
90 | open_access_count++; | |
91 | } else { | |
92 | DUMP_DEVICE_ERROR(); | |
93 | } | |
94 | mutex_unlock(&devreq_mutex); | |
95 | ||
96 | out: | |
97 | ||
98 | gossip_debug(GOSSIP_DEV_DEBUG, | |
99 | "pvfs2-client-core: open device complete (ret = %d)\n", | |
100 | ret); | |
101 | return ret; | |
102 | } | |
103 | ||
97f10027 | 104 | /* Function for read() callers into the device */ |
8bb8aefd | 105 | static ssize_t orangefs_devreq_read(struct file *file, |
5db11c21 MM |
106 | char __user *buf, |
107 | size_t count, loff_t *offset) | |
108 | { | |
8bb8aefd YL |
109 | struct orangefs_kernel_op_s *op, *temp; |
110 | __s32 proto_ver = ORANGEFS_KERNEL_PROTO_VERSION; | |
111 | static __s32 magic = ORANGEFS_DEVREQ_MAGIC; | |
112 | struct orangefs_kernel_op_s *cur_op = NULL; | |
24c8d080 | 113 | unsigned long ret; |
5db11c21 | 114 | |
24c8d080 | 115 | /* We do not support blocking IO. */ |
5db11c21 | 116 | if (!(file->f_flags & O_NONBLOCK)) { |
97f10027 MM |
117 | gossip_err("%s: blocking read from client-core.\n", |
118 | __func__); | |
5db11c21 | 119 | return -EINVAL; |
24c8d080 MB |
120 | } |
121 | ||
122 | /* | |
a762ae6d | 123 | * The client will do an ioctl to find MAX_DEV_REQ_UPSIZE, then |
24c8d080 MB |
124 | * always read with that size buffer. |
125 | */ | |
a762ae6d | 126 | if (count != MAX_DEV_REQ_UPSIZE) { |
24c8d080 MB |
127 | gossip_err("orangefs: client-core tried to read wrong size\n"); |
128 | return -EINVAL; | |
129 | } | |
130 | ||
131 | /* Get next op (if any) from top of list. */ | |
8bb8aefd YL |
132 | spin_lock(&orangefs_request_list_lock); |
133 | list_for_each_entry_safe(op, temp, &orangefs_request_list, list) { | |
24c8d080 MB |
134 | __s32 fsid; |
135 | /* This lock is held past the end of the loop when we break. */ | |
136 | spin_lock(&op->lock); | |
137 | ||
138 | fsid = fsid_of_op(op); | |
8bb8aefd | 139 | if (fsid != ORANGEFS_FS_ID_NULL) { |
24c8d080 MB |
140 | int ret; |
141 | /* Skip ops whose filesystem needs to be mounted. */ | |
142 | ret = fs_mount_pending(fsid); | |
143 | if (ret == 1) { | |
5db11c21 | 144 | gossip_debug(GOSSIP_DEV_DEBUG, |
24c8d080 MB |
145 | "orangefs: skipping op tag %llu %s\n", |
146 | llu(op->tag), get_opname_string(op)); | |
147 | spin_unlock(&op->lock); | |
148 | continue; | |
97f10027 MM |
149 | /* |
150 | * Skip ops whose filesystem we don't know about unless | |
151 | * it is being mounted. | |
152 | */ | |
24c8d080 MB |
153 | /* XXX: is there a better way to detect this? */ |
154 | } else if (ret == -1 && | |
97f10027 MM |
155 | !(op->upcall.type == |
156 | ORANGEFS_VFS_OP_FS_MOUNT || | |
157 | op->upcall.type == | |
158 | ORANGEFS_VFS_OP_GETATTR)) { | |
24c8d080 MB |
159 | gossip_debug(GOSSIP_DEV_DEBUG, |
160 | "orangefs: skipping op tag %llu %s\n", | |
161 | llu(op->tag), get_opname_string(op)); | |
162 | gossip_err( | |
163 | "orangefs: ERROR: fs_mount_pending %d\n", | |
164 | fsid); | |
165 | spin_unlock(&op->lock); | |
5db11c21 | 166 | continue; |
5db11c21 MM |
167 | } |
168 | } | |
24c8d080 MB |
169 | /* |
170 | * Either this op does not pertain to a filesystem, is mounting | |
171 | * a filesystem, or pertains to a mounted filesystem. Let it | |
172 | * through. | |
173 | */ | |
174 | cur_op = op; | |
175 | break; | |
176 | } | |
177 | ||
178 | /* | |
179 | * At this point we either have a valid op and can continue or have not | |
180 | * found an op and must ask the client to try again later. | |
181 | */ | |
182 | if (!cur_op) { | |
8bb8aefd | 183 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 184 | return -EAGAIN; |
5db11c21 MM |
185 | } |
186 | ||
24c8d080 MB |
187 | gossip_debug(GOSSIP_DEV_DEBUG, "orangefs: reading op tag %llu %s\n", |
188 | llu(cur_op->tag), get_opname_string(cur_op)); | |
5db11c21 | 189 | |
24c8d080 MB |
190 | /* |
191 | * Such an op should never be on the list in the first place. If so, we | |
192 | * will abort. | |
193 | */ | |
194 | if (op_state_in_progress(cur_op) || op_state_serviced(cur_op)) { | |
195 | gossip_err("orangefs: ERROR: Current op already queued.\n"); | |
196 | list_del(&cur_op->list); | |
5db11c21 | 197 | spin_unlock(&cur_op->lock); |
8bb8aefd | 198 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 199 | return -EAGAIN; |
5db11c21 | 200 | } |
24c8d080 MB |
201 | |
202 | /* | |
203 | * Set the operation to be in progress and move it between lists since | |
204 | * it has been sent to the client. | |
205 | */ | |
206 | set_op_state_inprogress(cur_op); | |
207 | ||
208 | list_del(&cur_op->list); | |
8bb8aefd YL |
209 | spin_unlock(&orangefs_request_list_lock); |
210 | orangefs_devreq_add_op(cur_op); | |
24c8d080 MB |
211 | spin_unlock(&cur_op->lock); |
212 | ||
213 | /* Push the upcall out. */ | |
214 | ret = copy_to_user(buf, &proto_ver, sizeof(__s32)); | |
215 | if (ret != 0) | |
216 | goto error; | |
217 | ret = copy_to_user(buf+sizeof(__s32), &magic, sizeof(__s32)); | |
218 | if (ret != 0) | |
219 | goto error; | |
220 | ret = copy_to_user(buf+2 * sizeof(__s32), &cur_op->tag, sizeof(__u64)); | |
221 | if (ret != 0) | |
222 | goto error; | |
223 | ret = copy_to_user(buf+2*sizeof(__s32)+sizeof(__u64), &cur_op->upcall, | |
8bb8aefd | 224 | sizeof(struct orangefs_upcall_s)); |
24c8d080 MB |
225 | if (ret != 0) |
226 | goto error; | |
227 | ||
228 | /* The client only asks to read one size buffer. */ | |
a762ae6d | 229 | return MAX_DEV_REQ_UPSIZE; |
24c8d080 MB |
230 | error: |
231 | /* | |
232 | * We were unable to copy the op data to the client. Put the op back in | |
233 | * list. If client has crashed, the op will be purged later when the | |
234 | * device is released. | |
235 | */ | |
236 | gossip_err("orangefs: Failed to copy data to user space\n"); | |
8bb8aefd | 237 | spin_lock(&orangefs_request_list_lock); |
24c8d080 MB |
238 | spin_lock(&cur_op->lock); |
239 | set_op_state_waiting(cur_op); | |
8bb8aefd YL |
240 | orangefs_devreq_remove_op(cur_op->tag); |
241 | list_add(&cur_op->list, &orangefs_request_list); | |
24c8d080 | 242 | spin_unlock(&cur_op->lock); |
8bb8aefd | 243 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 244 | return -EFAULT; |
5db11c21 MM |
245 | } |
246 | ||
97f10027 | 247 | /* |
b3ae4755 MM |
248 | * Function for writev() callers into the device. |
249 | * | |
250 | * Userspace should have written: | |
251 | * - __u32 version | |
252 | * - __u32 magic | |
253 | * - __u64 tag | |
254 | * - struct orangefs_downcall_s | |
255 | * - trailer buffer (in the case of READDIR operations) | |
97f10027 | 256 | */ |
b3ae4755 MM |
257 | static ssize_t orangefs_devreq_write_iter(struct kiocb *iocb, |
258 | struct iov_iter *iter) | |
5db11c21 | 259 | { |
b3ae4755 | 260 | ssize_t ret; |
8bb8aefd | 261 | struct orangefs_kernel_op_s *op = NULL; |
b3ae4755 MM |
262 | struct { |
263 | __u32 version; | |
264 | __u32 magic; | |
265 | __u64 tag; | |
266 | } head; | |
267 | int total = ret = iov_iter_count(iter); | |
268 | int n; | |
269 | int downcall_size = sizeof(struct orangefs_downcall_s); | |
270 | int head_size = sizeof(head); | |
271 | ||
272 | gossip_debug(GOSSIP_DEV_DEBUG, "%s: total:%d: ret:%zd:\n", | |
273 | __func__, | |
274 | total, | |
275 | ret); | |
5db11c21 | 276 | |
b3ae4755 MM |
277 | if (total < MAX_DEV_REQ_DOWNSIZE) { |
278 | gossip_err("%s: total:%d: must be at least:%lu:\n", | |
279 | __func__, | |
280 | total, | |
281 | MAX_DEV_REQ_DOWNSIZE); | |
282 | ret = -EFAULT; | |
283 | goto out; | |
5db11c21 | 284 | } |
b3ae4755 MM |
285 | |
286 | n = copy_from_iter(&head, head_size, iter); | |
287 | if (n < head_size) { | |
288 | gossip_err("%s: failed to copy head.\n", __func__); | |
289 | ret = -EFAULT; | |
290 | goto out; | |
97f10027 | 291 | } |
b3ae4755 MM |
292 | |
293 | if (head.version < ORANGEFS_MINIMUM_USERSPACE_VERSION) { | |
294 | gossip_err("%s: userspace claims version" | |
295 | "%d, minimum version required: %d.\n", | |
296 | __func__, | |
297 | head.version, | |
298 | ORANGEFS_MINIMUM_USERSPACE_VERSION); | |
299 | ret = -EPROTO; | |
300 | goto out; | |
5db11c21 | 301 | } |
5db11c21 | 302 | |
b3ae4755 MM |
303 | if (head.magic != ORANGEFS_DEVREQ_MAGIC) { |
304 | gossip_err("Error: Device magic number does not match.\n"); | |
305 | ret = -EPROTO; | |
306 | goto out; | |
307 | } | |
5db11c21 | 308 | |
b3ae4755 MM |
309 | op = orangefs_devreq_remove_op(head.tag); |
310 | if (!op) { | |
311 | gossip_err("WARNING: No one's waiting for tag %llu\n", | |
312 | llu(head.tag)); | |
313 | goto out; | |
314 | } | |
5db11c21 | 315 | |
b3ae4755 | 316 | get_op(op); /* increase ref count. */ |
5db11c21 | 317 | |
b3ae4755 MM |
318 | n = copy_from_iter(&op->downcall, downcall_size, iter); |
319 | if (n != downcall_size) { | |
320 | gossip_err("%s: failed to copy downcall.\n", __func__); | |
321 | put_op(op); | |
322 | ret = -EFAULT; | |
323 | goto out; | |
5db11c21 MM |
324 | } |
325 | ||
b3ae4755 MM |
326 | if (op->downcall.status) |
327 | goto wakeup; | |
97f10027 | 328 | |
b3ae4755 MM |
329 | /* |
330 | * We've successfully peeled off the head and the downcall. | |
331 | * Something has gone awry if total doesn't equal the | |
332 | * sum of head_size, downcall_size and trailer_size. | |
333 | */ | |
334 | if ((head_size + downcall_size + op->downcall.trailer_size) != total) { | |
335 | gossip_err("%s: funky write, head_size:%d" | |
336 | ": downcall_size:%d: trailer_size:%lld" | |
337 | ": total size:%d:\n", | |
338 | __func__, | |
339 | head_size, | |
340 | downcall_size, | |
341 | op->downcall.trailer_size, | |
342 | total); | |
343 | put_op(op); | |
344 | ret = -EFAULT; | |
345 | goto out; | |
346 | } | |
97f10027 | 347 | |
b3ae4755 MM |
348 | /* Only READDIR operations should have trailers. */ |
349 | if ((op->downcall.type != ORANGEFS_VFS_OP_READDIR) && | |
350 | (op->downcall.trailer_size != 0)) { | |
351 | gossip_err("%s: %x operation with trailer.", | |
352 | __func__, | |
353 | op->downcall.type); | |
354 | put_op(op); | |
355 | ret = -EFAULT; | |
356 | goto out; | |
357 | } | |
97f10027 | 358 | |
b3ae4755 MM |
359 | /* READDIR operations should always have trailers. */ |
360 | if ((op->downcall.type == ORANGEFS_VFS_OP_READDIR) && | |
361 | (op->downcall.trailer_size == 0)) { | |
362 | gossip_err("%s: %x operation with no trailer.", | |
363 | __func__, | |
364 | op->downcall.type); | |
365 | put_op(op); | |
366 | ret = -EFAULT; | |
367 | goto out; | |
368 | } | |
97f10027 | 369 | |
b3ae4755 MM |
370 | if (op->downcall.type != ORANGEFS_VFS_OP_READDIR) |
371 | goto wakeup; | |
5db11c21 | 372 | |
b3ae4755 MM |
373 | op->downcall.trailer_buf = |
374 | vmalloc(op->downcall.trailer_size); | |
375 | if (op->downcall.trailer_buf == NULL) { | |
376 | gossip_err("%s: failed trailer vmalloc.\n", | |
377 | __func__); | |
378 | put_op(op); | |
379 | ret = -ENOMEM; | |
380 | goto out; | |
381 | } | |
382 | memset(op->downcall.trailer_buf, 0, op->downcall.trailer_size); | |
383 | n = copy_from_iter(op->downcall.trailer_buf, | |
384 | op->downcall.trailer_size, | |
385 | iter); | |
386 | if (n != op->downcall.trailer_size) { | |
387 | gossip_err("%s: failed to copy trailer.\n", __func__); | |
388 | vfree(op->downcall.trailer_buf); | |
389 | put_op(op); | |
390 | ret = -EFAULT; | |
391 | goto out; | |
392 | } | |
97f10027 | 393 | |
b3ae4755 | 394 | wakeup: |
97f10027 | 395 | |
b3ae4755 MM |
396 | /* |
397 | * If this operation is an I/O operation we need to wait | |
398 | * for all data to be copied before we can return to avoid | |
399 | * buffer corruption and races that can pull the buffers | |
400 | * out from under us. | |
401 | * | |
402 | * Essentially we're synchronizing with other parts of the | |
403 | * vfs implicitly by not allowing the user space | |
404 | * application reading/writing this device to return until | |
405 | * the buffers are done being used. | |
406 | */ | |
407 | if (op->downcall.type == ORANGEFS_VFS_OP_FILE_IO) { | |
408 | int timed_out = 0; | |
409 | DEFINE_WAIT(wait_entry); | |
97f10027 MM |
410 | |
411 | /* | |
b3ae4755 MM |
412 | * tell the vfs op waiting on a waitqueue |
413 | * that this op is done | |
97f10027 | 414 | */ |
b3ae4755 MM |
415 | spin_lock(&op->lock); |
416 | set_op_state_serviced(op); | |
417 | spin_unlock(&op->lock); | |
5db11c21 | 418 | |
b3ae4755 | 419 | wake_up_interruptible(&op->waitq); |
5db11c21 | 420 | |
b3ae4755 | 421 | while (1) { |
5db11c21 | 422 | spin_lock(&op->lock); |
b3ae4755 MM |
423 | prepare_to_wait_exclusive( |
424 | &op->io_completion_waitq, | |
425 | &wait_entry, | |
426 | TASK_INTERRUPTIBLE); | |
427 | if (op->io_completed) { | |
5db11c21 | 428 | spin_unlock(&op->lock); |
5db11c21 MM |
429 | break; |
430 | } | |
ce6c414e | 431 | spin_unlock(&op->lock); |
5db11c21 | 432 | |
b3ae4755 MM |
433 | if (!signal_pending(current)) { |
434 | int timeout = | |
435 | MSECS_TO_JIFFIES(1000 * | |
436 | op_timeout_secs); | |
437 | if (!schedule_timeout(timeout)) { | |
438 | gossip_debug(GOSSIP_DEV_DEBUG, | |
439 | "%s: timed out.\n", | |
440 | __func__); | |
441 | timed_out = 1; | |
442 | break; | |
443 | } | |
444 | continue; | |
445 | } | |
5db11c21 | 446 | |
b3ae4755 MM |
447 | gossip_debug(GOSSIP_DEV_DEBUG, |
448 | "%s: signal on I/O wait, aborting\n", | |
449 | __func__); | |
450 | break; | |
5db11c21 | 451 | } |
b3ae4755 MM |
452 | |
453 | spin_lock(&op->lock); | |
454 | finish_wait(&op->io_completion_waitq, &wait_entry); | |
455 | spin_unlock(&op->lock); | |
456 | ||
457 | /* NOTE: for I/O operations we handle releasing the op | |
458 | * object except in the case of timeout. the reason we | |
459 | * can't free the op in timeout cases is that the op | |
460 | * service logic in the vfs retries operations using | |
461 | * the same op ptr, thus it can't be freed. | |
462 | */ | |
463 | if (!timed_out) | |
464 | op_release(op); | |
5db11c21 | 465 | } else { |
b3ae4755 MM |
466 | /* |
467 | * tell the vfs op waiting on a waitqueue that | |
468 | * this op is done | |
469 | */ | |
470 | spin_lock(&op->lock); | |
471 | set_op_state_serviced(op); | |
472 | spin_unlock(&op->lock); | |
473 | /* | |
474 | * for every other operation (i.e. non-I/O), we need to | |
475 | * wake up the callers for downcall completion | |
476 | * notification | |
477 | */ | |
478 | wake_up_interruptible(&op->waitq); | |
5db11c21 | 479 | } |
b3ae4755 MM |
480 | out: |
481 | return ret; | |
5db11c21 MM |
482 | } |
483 | ||
484 | /* Returns whether any FS are still pending remounted */ | |
485 | static int mark_all_pending_mounts(void) | |
486 | { | |
487 | int unmounted = 1; | |
8bb8aefd | 488 | struct orangefs_sb_info_s *orangefs_sb = NULL; |
5db11c21 | 489 | |
8bb8aefd YL |
490 | spin_lock(&orangefs_superblocks_lock); |
491 | list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) { | |
5db11c21 | 492 | /* All of these file system require a remount */ |
8bb8aefd | 493 | orangefs_sb->mount_pending = 1; |
5db11c21 MM |
494 | unmounted = 0; |
495 | } | |
8bb8aefd | 496 | spin_unlock(&orangefs_superblocks_lock); |
5db11c21 MM |
497 | return unmounted; |
498 | } | |
499 | ||
500 | /* | |
501 | * Determine if a given file system needs to be remounted or not | |
502 | * Returns -1 on error | |
503 | * 0 if already mounted | |
504 | * 1 if needs remount | |
505 | */ | |
506 | int fs_mount_pending(__s32 fsid) | |
507 | { | |
508 | int mount_pending = -1; | |
8bb8aefd | 509 | struct orangefs_sb_info_s *orangefs_sb = NULL; |
5db11c21 | 510 | |
8bb8aefd YL |
511 | spin_lock(&orangefs_superblocks_lock); |
512 | list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) { | |
513 | if (orangefs_sb->fs_id == fsid) { | |
514 | mount_pending = orangefs_sb->mount_pending; | |
5db11c21 MM |
515 | break; |
516 | } | |
517 | } | |
8bb8aefd | 518 | spin_unlock(&orangefs_superblocks_lock); |
5db11c21 MM |
519 | return mount_pending; |
520 | } | |
521 | ||
522 | /* | |
523 | * NOTE: gets called when the last reference to this device is dropped. | |
524 | * Using the open_access_count variable, we enforce a reference count | |
525 | * on this file so that it can be opened by only one process at a time. | |
526 | * the devreq_mutex is used to make sure all i/o has completed | |
8bb8aefd | 527 | * before we call orangefs_bufmap_finalize, and similar such tricky |
5db11c21 MM |
528 | * situations |
529 | */ | |
8bb8aefd | 530 | static int orangefs_devreq_release(struct inode *inode, struct file *file) |
5db11c21 MM |
531 | { |
532 | int unmounted = 0; | |
533 | ||
534 | gossip_debug(GOSSIP_DEV_DEBUG, | |
535 | "%s:pvfs2-client-core: exiting, closing device\n", | |
536 | __func__); | |
537 | ||
538 | mutex_lock(&devreq_mutex); | |
7d221485 | 539 | if (orangefs_get_bufmap_init()) |
90d26aa8 | 540 | orangefs_bufmap_finalize(); |
5db11c21 MM |
541 | |
542 | open_access_count--; | |
543 | ||
544 | unmounted = mark_all_pending_mounts(); | |
8bb8aefd | 545 | gossip_debug(GOSSIP_DEV_DEBUG, "ORANGEFS Device Close: Filesystem(s) %s\n", |
5db11c21 MM |
546 | (unmounted ? "UNMOUNTED" : "MOUNTED")); |
547 | mutex_unlock(&devreq_mutex); | |
548 | ||
549 | /* | |
550 | * Walk through the list of ops in the request list, mark them | |
551 | * as purged and wake them up. | |
552 | */ | |
553 | purge_waiting_ops(); | |
554 | /* | |
555 | * Walk through the hash table of in progress operations; mark | |
556 | * them as purged and wake them up | |
557 | */ | |
558 | purge_inprogress_ops(); | |
559 | gossip_debug(GOSSIP_DEV_DEBUG, | |
560 | "pvfs2-client-core: device close complete\n"); | |
561 | return 0; | |
562 | } | |
563 | ||
564 | int is_daemon_in_service(void) | |
565 | { | |
566 | int in_service; | |
567 | ||
568 | /* | |
569 | * What this function does is checks if client-core is alive | |
570 | * based on the access count we maintain on the device. | |
571 | */ | |
572 | mutex_lock(&devreq_mutex); | |
573 | in_service = open_access_count == 1 ? 0 : -EIO; | |
574 | mutex_unlock(&devreq_mutex); | |
575 | return in_service; | |
576 | } | |
577 | ||
578 | static inline long check_ioctl_command(unsigned int command) | |
579 | { | |
580 | /* Check for valid ioctl codes */ | |
8bb8aefd | 581 | if (_IOC_TYPE(command) != ORANGEFS_DEV_MAGIC) { |
5db11c21 MM |
582 | gossip_err("device ioctl magic numbers don't match! Did you rebuild pvfs2-client-core/libpvfs2? [cmd %x, magic %x != %x]\n", |
583 | command, | |
584 | _IOC_TYPE(command), | |
8bb8aefd | 585 | ORANGEFS_DEV_MAGIC); |
5db11c21 MM |
586 | return -EINVAL; |
587 | } | |
588 | /* and valid ioctl commands */ | |
8bb8aefd | 589 | if (_IOC_NR(command) >= ORANGEFS_DEV_MAXNR || _IOC_NR(command) <= 0) { |
5db11c21 | 590 | gossip_err("Invalid ioctl command number [%d >= %d]\n", |
8bb8aefd | 591 | _IOC_NR(command), ORANGEFS_DEV_MAXNR); |
5db11c21 MM |
592 | return -ENOIOCTLCMD; |
593 | } | |
594 | return 0; | |
595 | } | |
596 | ||
597 | static long dispatch_ioctl_command(unsigned int command, unsigned long arg) | |
598 | { | |
8bb8aefd | 599 | static __s32 magic = ORANGEFS_DEVREQ_MAGIC; |
a762ae6d MB |
600 | static __s32 max_up_size = MAX_DEV_REQ_UPSIZE; |
601 | static __s32 max_down_size = MAX_DEV_REQ_DOWNSIZE; | |
8bb8aefd | 602 | struct ORANGEFS_dev_map_desc user_desc; |
5db11c21 MM |
603 | int ret = 0; |
604 | struct dev_mask_info_s mask_info = { 0 }; | |
605 | struct dev_mask2_info_s mask2_info = { 0, 0 }; | |
606 | int upstream_kmod = 1; | |
607 | struct list_head *tmp = NULL; | |
8bb8aefd | 608 | struct orangefs_sb_info_s *orangefs_sb = NULL; |
5db11c21 MM |
609 | |
610 | /* mtmoore: add locking here */ | |
611 | ||
612 | switch (command) { | |
8bb8aefd | 613 | case ORANGEFS_DEV_GET_MAGIC: |
5db11c21 MM |
614 | return ((put_user(magic, (__s32 __user *) arg) == -EFAULT) ? |
615 | -EIO : | |
616 | 0); | |
8bb8aefd | 617 | case ORANGEFS_DEV_GET_MAX_UPSIZE: |
5db11c21 MM |
618 | return ((put_user(max_up_size, |
619 | (__s32 __user *) arg) == -EFAULT) ? | |
620 | -EIO : | |
621 | 0); | |
8bb8aefd | 622 | case ORANGEFS_DEV_GET_MAX_DOWNSIZE: |
5db11c21 MM |
623 | return ((put_user(max_down_size, |
624 | (__s32 __user *) arg) == -EFAULT) ? | |
625 | -EIO : | |
626 | 0); | |
8bb8aefd | 627 | case ORANGEFS_DEV_MAP: |
5db11c21 | 628 | ret = copy_from_user(&user_desc, |
8bb8aefd | 629 | (struct ORANGEFS_dev_map_desc __user *) |
5db11c21 | 630 | arg, |
8bb8aefd | 631 | sizeof(struct ORANGEFS_dev_map_desc)); |
7d221485 | 632 | if (orangefs_get_bufmap_init()) { |
90d26aa8 MB |
633 | return -EINVAL; |
634 | } else { | |
635 | return ret ? | |
636 | -EIO : | |
637 | orangefs_bufmap_initialize(&user_desc); | |
638 | } | |
8bb8aefd | 639 | case ORANGEFS_DEV_REMOUNT_ALL: |
5db11c21 | 640 | gossip_debug(GOSSIP_DEV_DEBUG, |
97f10027 MM |
641 | "%s: got ORANGEFS_DEV_REMOUNT_ALL\n", |
642 | __func__); | |
5db11c21 MM |
643 | |
644 | /* | |
8bb8aefd | 645 | * remount all mounted orangefs volumes to regain the lost |
5db11c21 MM |
646 | * dynamic mount tables (if any) -- NOTE: this is done |
647 | * without keeping the superblock list locked due to the | |
648 | * upcall/downcall waiting. also, the request semaphore is | |
649 | * used to ensure that no operations will be serviced until | |
650 | * all of the remounts are serviced (to avoid ops between | |
651 | * mounts to fail) | |
652 | */ | |
653 | ret = mutex_lock_interruptible(&request_mutex); | |
654 | if (ret < 0) | |
655 | return ret; | |
656 | gossip_debug(GOSSIP_DEV_DEBUG, | |
97f10027 MM |
657 | "%s: priority remount in progress\n", |
658 | __func__); | |
8bb8aefd YL |
659 | list_for_each(tmp, &orangefs_superblocks) { |
660 | orangefs_sb = | |
97f10027 MM |
661 | list_entry(tmp, |
662 | struct orangefs_sb_info_s, | |
663 | list); | |
8bb8aefd | 664 | if (orangefs_sb && (orangefs_sb->sb)) { |
5db11c21 | 665 | gossip_debug(GOSSIP_DEV_DEBUG, |
97f10027 MM |
666 | "%s: Remounting SB %p\n", |
667 | __func__, | |
8bb8aefd | 668 | orangefs_sb); |
5db11c21 | 669 | |
8bb8aefd | 670 | ret = orangefs_remount(orangefs_sb->sb); |
5db11c21 MM |
671 | if (ret) { |
672 | gossip_debug(GOSSIP_DEV_DEBUG, | |
673 | "SB %p remount failed\n", | |
8bb8aefd | 674 | orangefs_sb); |
97f10027 | 675 | break; |
5db11c21 MM |
676 | } |
677 | } | |
678 | } | |
679 | gossip_debug(GOSSIP_DEV_DEBUG, | |
97f10027 MM |
680 | "%s: priority remount complete\n", |
681 | __func__); | |
5db11c21 MM |
682 | mutex_unlock(&request_mutex); |
683 | return ret; | |
684 | ||
8bb8aefd | 685 | case ORANGEFS_DEV_UPSTREAM: |
5db11c21 MM |
686 | ret = copy_to_user((void __user *)arg, |
687 | &upstream_kmod, | |
688 | sizeof(upstream_kmod)); | |
689 | ||
690 | if (ret != 0) | |
691 | return -EIO; | |
692 | else | |
693 | return ret; | |
694 | ||
8bb8aefd | 695 | case ORANGEFS_DEV_CLIENT_MASK: |
5db11c21 MM |
696 | ret = copy_from_user(&mask2_info, |
697 | (void __user *)arg, | |
698 | sizeof(struct dev_mask2_info_s)); | |
699 | ||
700 | if (ret != 0) | |
701 | return -EIO; | |
702 | ||
703 | client_debug_mask.mask1 = mask2_info.mask1_value; | |
704 | client_debug_mask.mask2 = mask2_info.mask2_value; | |
705 | ||
706 | pr_info("%s: client debug mask has been been received " | |
707 | ":%llx: :%llx:\n", | |
708 | __func__, | |
709 | (unsigned long long)client_debug_mask.mask1, | |
710 | (unsigned long long)client_debug_mask.mask2); | |
711 | ||
712 | return ret; | |
713 | ||
8bb8aefd | 714 | case ORANGEFS_DEV_CLIENT_STRING: |
5db11c21 MM |
715 | ret = copy_from_user(&client_debug_array_string, |
716 | (void __user *)arg, | |
8bb8aefd | 717 | ORANGEFS_MAX_DEBUG_STRING_LEN); |
5db11c21 | 718 | if (ret != 0) { |
97f10027 | 719 | pr_info("%s: CLIENT_STRING: copy_from_user failed\n", |
5db11c21 MM |
720 | __func__); |
721 | return -EIO; | |
722 | } | |
723 | ||
97f10027 | 724 | pr_info("%s: client debug array string has been received.\n", |
5db11c21 MM |
725 | __func__); |
726 | ||
727 | if (!help_string_initialized) { | |
728 | ||
729 | /* Free the "we don't know yet" default string... */ | |
730 | kfree(debug_help_string); | |
731 | ||
732 | /* build a proper debug help string */ | |
733 | if (orangefs_prepare_debugfs_help_string(0)) { | |
97f10027 | 734 | gossip_err("%s: no debug help string \n", |
5db11c21 MM |
735 | __func__); |
736 | return -EIO; | |
737 | } | |
738 | ||
739 | /* Replace the boilerplate boot-time debug-help file. */ | |
740 | debugfs_remove(help_file_dentry); | |
741 | ||
742 | help_file_dentry = | |
743 | debugfs_create_file( | |
744 | ORANGEFS_KMOD_DEBUG_HELP_FILE, | |
745 | 0444, | |
746 | debug_dir, | |
747 | debug_help_string, | |
748 | &debug_help_fops); | |
749 | ||
750 | if (!help_file_dentry) { | |
751 | gossip_err("%s: debugfs_create_file failed for" | |
752 | " :%s:!\n", | |
753 | __func__, | |
754 | ORANGEFS_KMOD_DEBUG_HELP_FILE); | |
755 | return -EIO; | |
756 | } | |
757 | } | |
758 | ||
759 | debug_mask_to_string(&client_debug_mask, 1); | |
760 | ||
761 | debugfs_remove(client_debug_dentry); | |
762 | ||
8bb8aefd | 763 | orangefs_client_debug_init(); |
5db11c21 MM |
764 | |
765 | help_string_initialized++; | |
766 | ||
767 | return ret; | |
768 | ||
8bb8aefd | 769 | case ORANGEFS_DEV_DEBUG: |
5db11c21 MM |
770 | ret = copy_from_user(&mask_info, |
771 | (void __user *)arg, | |
772 | sizeof(mask_info)); | |
773 | ||
774 | if (ret != 0) | |
775 | return -EIO; | |
776 | ||
777 | if (mask_info.mask_type == KERNEL_MASK) { | |
778 | if ((mask_info.mask_value == 0) | |
779 | && (kernel_mask_set_mod_init)) { | |
780 | /* | |
781 | * the kernel debug mask was set when the | |
782 | * kernel module was loaded; don't override | |
783 | * it if the client-core was started without | |
8bb8aefd | 784 | * a value for ORANGEFS_KMODMASK. |
5db11c21 MM |
785 | */ |
786 | return 0; | |
787 | } | |
788 | debug_mask_to_string(&mask_info.mask_value, | |
789 | mask_info.mask_type); | |
790 | gossip_debug_mask = mask_info.mask_value; | |
97f10027 | 791 | pr_info("%s: kernel debug mask has been modified to " |
5db11c21 | 792 | ":%s: :%llx:\n", |
97f10027 | 793 | __func__, |
5db11c21 MM |
794 | kernel_debug_string, |
795 | (unsigned long long)gossip_debug_mask); | |
796 | } else if (mask_info.mask_type == CLIENT_MASK) { | |
797 | debug_mask_to_string(&mask_info.mask_value, | |
798 | mask_info.mask_type); | |
97f10027 | 799 | pr_info("%s: client debug mask has been modified to" |
5db11c21 | 800 | ":%s: :%llx:\n", |
97f10027 | 801 | __func__, |
5db11c21 MM |
802 | client_debug_string, |
803 | llu(mask_info.mask_value)); | |
804 | } else { | |
805 | gossip_lerr("Invalid mask type....\n"); | |
806 | return -EINVAL; | |
807 | } | |
808 | ||
809 | return ret; | |
810 | ||
811 | default: | |
812 | return -ENOIOCTLCMD; | |
813 | } | |
814 | return -ENOIOCTLCMD; | |
815 | } | |
816 | ||
8bb8aefd | 817 | static long orangefs_devreq_ioctl(struct file *file, |
5db11c21 MM |
818 | unsigned int command, unsigned long arg) |
819 | { | |
820 | long ret; | |
821 | ||
822 | /* Check for properly constructed commands */ | |
823 | ret = check_ioctl_command(command); | |
824 | if (ret < 0) | |
825 | return (int)ret; | |
826 | ||
827 | return (int)dispatch_ioctl_command(command, arg); | |
828 | } | |
829 | ||
830 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
831 | ||
8bb8aefd YL |
832 | /* Compat structure for the ORANGEFS_DEV_MAP ioctl */ |
833 | struct ORANGEFS_dev_map_desc32 { | |
5db11c21 MM |
834 | compat_uptr_t ptr; |
835 | __s32 total_size; | |
836 | __s32 size; | |
837 | __s32 count; | |
838 | }; | |
839 | ||
840 | static unsigned long translate_dev_map26(unsigned long args, long *error) | |
841 | { | |
8bb8aefd | 842 | struct ORANGEFS_dev_map_desc32 __user *p32 = (void __user *)args; |
5db11c21 MM |
843 | /* |
844 | * Depending on the architecture, allocate some space on the | |
845 | * user-call-stack based on our expected layout. | |
846 | */ | |
8bb8aefd | 847 | struct ORANGEFS_dev_map_desc __user *p = |
5db11c21 | 848 | compat_alloc_user_space(sizeof(*p)); |
84d02150 | 849 | compat_uptr_t addr; |
5db11c21 MM |
850 | |
851 | *error = 0; | |
852 | /* get the ptr from the 32 bit user-space */ | |
853 | if (get_user(addr, &p32->ptr)) | |
854 | goto err; | |
855 | /* try to put that into a 64-bit layout */ | |
856 | if (put_user(compat_ptr(addr), &p->ptr)) | |
857 | goto err; | |
858 | /* copy the remaining fields */ | |
859 | if (copy_in_user(&p->total_size, &p32->total_size, sizeof(__s32))) | |
860 | goto err; | |
861 | if (copy_in_user(&p->size, &p32->size, sizeof(__s32))) | |
862 | goto err; | |
863 | if (copy_in_user(&p->count, &p32->count, sizeof(__s32))) | |
864 | goto err; | |
865 | return (unsigned long)p; | |
866 | err: | |
867 | *error = -EFAULT; | |
868 | return 0; | |
869 | } | |
870 | ||
871 | /* | |
872 | * 32 bit user-space apps' ioctl handlers when kernel modules | |
873 | * is compiled as a 64 bit one | |
874 | */ | |
8bb8aefd | 875 | static long orangefs_devreq_compat_ioctl(struct file *filp, unsigned int cmd, |
5db11c21 MM |
876 | unsigned long args) |
877 | { | |
878 | long ret; | |
879 | unsigned long arg = args; | |
880 | ||
881 | /* Check for properly constructed commands */ | |
882 | ret = check_ioctl_command(cmd); | |
883 | if (ret < 0) | |
884 | return ret; | |
8bb8aefd | 885 | if (cmd == ORANGEFS_DEV_MAP) { |
5db11c21 MM |
886 | /* |
887 | * convert the arguments to what we expect internally | |
888 | * in kernel space | |
889 | */ | |
890 | arg = translate_dev_map26(args, &ret); | |
891 | if (ret < 0) { | |
892 | gossip_err("Could not translate dev map\n"); | |
893 | return ret; | |
894 | } | |
895 | } | |
896 | /* no other ioctl requires translation */ | |
897 | return dispatch_ioctl_command(cmd, arg); | |
898 | } | |
899 | ||
2c590d5f MM |
900 | #endif /* CONFIG_COMPAT is in .config */ |
901 | ||
902 | /* | |
903 | * The following two ioctl32 functions had been refactored into the above | |
904 | * CONFIG_COMPAT ifdef, but that was an over simplification that was | |
905 | * not noticed until we tried to compile on power pc... | |
906 | */ | |
907 | #if (defined(CONFIG_COMPAT) && !defined(HAVE_REGISTER_IOCTL32_CONVERSION)) || !defined(CONFIG_COMPAT) | |
8bb8aefd | 908 | static int orangefs_ioctl32_init(void) |
5db11c21 MM |
909 | { |
910 | return 0; | |
911 | } | |
912 | ||
8bb8aefd | 913 | static void orangefs_ioctl32_cleanup(void) |
5db11c21 MM |
914 | { |
915 | return; | |
916 | } | |
2c590d5f | 917 | #endif |
5db11c21 MM |
918 | |
919 | /* the assigned character device major number */ | |
8bb8aefd | 920 | static int orangefs_dev_major; |
5db11c21 MM |
921 | |
922 | /* | |
8bb8aefd | 923 | * Initialize orangefs device specific state: |
5db11c21 MM |
924 | * Must be called at module load time only |
925 | */ | |
8bb8aefd | 926 | int orangefs_dev_init(void) |
5db11c21 MM |
927 | { |
928 | int ret; | |
929 | ||
930 | /* register the ioctl32 sub-system */ | |
8bb8aefd | 931 | ret = orangefs_ioctl32_init(); |
5db11c21 MM |
932 | if (ret < 0) |
933 | return ret; | |
934 | ||
8bb8aefd YL |
935 | /* register orangefs-req device */ |
936 | orangefs_dev_major = register_chrdev(0, | |
937 | ORANGEFS_REQDEVICE_NAME, | |
938 | &orangefs_devreq_file_operations); | |
939 | if (orangefs_dev_major < 0) { | |
5db11c21 MM |
940 | gossip_debug(GOSSIP_DEV_DEBUG, |
941 | "Failed to register /dev/%s (error %d)\n", | |
8bb8aefd YL |
942 | ORANGEFS_REQDEVICE_NAME, orangefs_dev_major); |
943 | orangefs_ioctl32_cleanup(); | |
944 | return orangefs_dev_major; | |
5db11c21 MM |
945 | } |
946 | ||
947 | gossip_debug(GOSSIP_DEV_DEBUG, | |
948 | "*** /dev/%s character device registered ***\n", | |
8bb8aefd | 949 | ORANGEFS_REQDEVICE_NAME); |
5db11c21 | 950 | gossip_debug(GOSSIP_DEV_DEBUG, "'mknod /dev/%s c %d 0'.\n", |
8bb8aefd | 951 | ORANGEFS_REQDEVICE_NAME, orangefs_dev_major); |
5db11c21 MM |
952 | return 0; |
953 | } | |
954 | ||
8bb8aefd | 955 | void orangefs_dev_cleanup(void) |
5db11c21 | 956 | { |
8bb8aefd | 957 | unregister_chrdev(orangefs_dev_major, ORANGEFS_REQDEVICE_NAME); |
5db11c21 MM |
958 | gossip_debug(GOSSIP_DEV_DEBUG, |
959 | "*** /dev/%s character device unregistered ***\n", | |
8bb8aefd | 960 | ORANGEFS_REQDEVICE_NAME); |
5db11c21 | 961 | /* unregister the ioctl32 sub-system */ |
8bb8aefd | 962 | orangefs_ioctl32_cleanup(); |
5db11c21 MM |
963 | } |
964 | ||
8bb8aefd | 965 | static unsigned int orangefs_devreq_poll(struct file *file, |
5db11c21 MM |
966 | struct poll_table_struct *poll_table) |
967 | { | |
968 | int poll_revent_mask = 0; | |
969 | ||
970 | if (open_access_count == 1) { | |
8bb8aefd | 971 | poll_wait(file, &orangefs_request_list_waitq, poll_table); |
5db11c21 | 972 | |
8bb8aefd YL |
973 | spin_lock(&orangefs_request_list_lock); |
974 | if (!list_empty(&orangefs_request_list)) | |
5db11c21 | 975 | poll_revent_mask |= POLL_IN; |
8bb8aefd | 976 | spin_unlock(&orangefs_request_list_lock); |
5db11c21 MM |
977 | } |
978 | return poll_revent_mask; | |
979 | } | |
980 | ||
8bb8aefd | 981 | const struct file_operations orangefs_devreq_file_operations = { |
5db11c21 | 982 | .owner = THIS_MODULE, |
8bb8aefd YL |
983 | .read = orangefs_devreq_read, |
984 | .write_iter = orangefs_devreq_write_iter, | |
985 | .open = orangefs_devreq_open, | |
986 | .release = orangefs_devreq_release, | |
987 | .unlocked_ioctl = orangefs_devreq_ioctl, | |
5db11c21 MM |
988 | |
989 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
8bb8aefd | 990 | .compat_ioctl = orangefs_devreq_compat_ioctl, |
5db11c21 | 991 | #endif |
8bb8aefd | 992 | .poll = orangefs_devreq_poll |
5db11c21 | 993 | }; |