]>
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 | ||
5db11c21 | 46 | list_add_tail(&op->list, &htable_ops_in_progress[index]); |
5db11c21 MM |
47 | } |
48 | ||
8bb8aefd | 49 | static struct orangefs_kernel_op_s *orangefs_devreq_remove_op(__u64 tag) |
5db11c21 | 50 | { |
8bb8aefd | 51 | struct orangefs_kernel_op_s *op, *next; |
5db11c21 MM |
52 | int index; |
53 | ||
54 | index = hash_func(tag, hash_table_size); | |
55 | ||
56 | spin_lock(&htable_ops_in_progress_lock); | |
57 | list_for_each_entry_safe(op, | |
58 | next, | |
59 | &htable_ops_in_progress[index], | |
60 | list) { | |
ed42fe05 AV |
61 | if (op->tag == tag && !op_state_purged(op)) { |
62 | list_del_init(&op->list); | |
63 | get_op(op); /* increase ref count. */ | |
5db11c21 MM |
64 | spin_unlock(&htable_ops_in_progress_lock); |
65 | return op; | |
66 | } | |
67 | } | |
68 | ||
69 | spin_unlock(&htable_ops_in_progress_lock); | |
70 | return NULL; | |
71 | } | |
72 | ||
8bb8aefd | 73 | static int orangefs_devreq_open(struct inode *inode, struct file *file) |
5db11c21 MM |
74 | { |
75 | int ret = -EINVAL; | |
76 | ||
77 | if (!(file->f_flags & O_NONBLOCK)) { | |
97f10027 MM |
78 | gossip_err("%s: device cannot be opened in blocking mode\n", |
79 | __func__); | |
5db11c21 MM |
80 | goto out; |
81 | } | |
82 | ret = -EACCES; | |
97f10027 | 83 | gossip_debug(GOSSIP_DEV_DEBUG, "client-core: opening device\n"); |
5db11c21 MM |
84 | mutex_lock(&devreq_mutex); |
85 | ||
86 | if (open_access_count == 0) { | |
fee25ce1 | 87 | open_access_count = 1; |
fb6d2526 | 88 | ret = 0; |
5db11c21 MM |
89 | } else { |
90 | DUMP_DEVICE_ERROR(); | |
91 | } | |
92 | mutex_unlock(&devreq_mutex); | |
93 | ||
94 | out: | |
95 | ||
96 | gossip_debug(GOSSIP_DEV_DEBUG, | |
97 | "pvfs2-client-core: open device complete (ret = %d)\n", | |
98 | ret); | |
99 | return ret; | |
100 | } | |
101 | ||
97f10027 | 102 | /* Function for read() callers into the device */ |
8bb8aefd | 103 | static ssize_t orangefs_devreq_read(struct file *file, |
5db11c21 MM |
104 | char __user *buf, |
105 | size_t count, loff_t *offset) | |
106 | { | |
8bb8aefd YL |
107 | struct orangefs_kernel_op_s *op, *temp; |
108 | __s32 proto_ver = ORANGEFS_KERNEL_PROTO_VERSION; | |
109 | static __s32 magic = ORANGEFS_DEVREQ_MAGIC; | |
110 | struct orangefs_kernel_op_s *cur_op = NULL; | |
24c8d080 | 111 | unsigned long ret; |
5db11c21 | 112 | |
24c8d080 | 113 | /* We do not support blocking IO. */ |
5db11c21 | 114 | if (!(file->f_flags & O_NONBLOCK)) { |
97f10027 MM |
115 | gossip_err("%s: blocking read from client-core.\n", |
116 | __func__); | |
5db11c21 | 117 | return -EINVAL; |
24c8d080 MB |
118 | } |
119 | ||
120 | /* | |
a762ae6d | 121 | * The client will do an ioctl to find MAX_DEV_REQ_UPSIZE, then |
24c8d080 MB |
122 | * always read with that size buffer. |
123 | */ | |
a762ae6d | 124 | if (count != MAX_DEV_REQ_UPSIZE) { |
24c8d080 MB |
125 | gossip_err("orangefs: client-core tried to read wrong size\n"); |
126 | return -EINVAL; | |
127 | } | |
128 | ||
ed42fe05 | 129 | restart: |
24c8d080 | 130 | /* Get next op (if any) from top of list. */ |
8bb8aefd YL |
131 | spin_lock(&orangefs_request_list_lock); |
132 | list_for_each_entry_safe(op, temp, &orangefs_request_list, list) { | |
24c8d080 MB |
133 | __s32 fsid; |
134 | /* This lock is held past the end of the loop when we break. */ | |
135 | spin_lock(&op->lock); | |
ed42fe05 AV |
136 | if (unlikely(op_state_purged(op))) { |
137 | spin_unlock(&op->lock); | |
138 | continue; | |
139 | } | |
24c8d080 MB |
140 | |
141 | fsid = fsid_of_op(op); | |
8bb8aefd | 142 | if (fsid != ORANGEFS_FS_ID_NULL) { |
24c8d080 MB |
143 | int ret; |
144 | /* Skip ops whose filesystem needs to be mounted. */ | |
145 | ret = fs_mount_pending(fsid); | |
146 | if (ret == 1) { | |
5db11c21 | 147 | gossip_debug(GOSSIP_DEV_DEBUG, |
24c8d080 MB |
148 | "orangefs: skipping op tag %llu %s\n", |
149 | llu(op->tag), get_opname_string(op)); | |
150 | spin_unlock(&op->lock); | |
151 | continue; | |
97f10027 MM |
152 | /* |
153 | * Skip ops whose filesystem we don't know about unless | |
154 | * it is being mounted. | |
155 | */ | |
24c8d080 MB |
156 | /* XXX: is there a better way to detect this? */ |
157 | } else if (ret == -1 && | |
97f10027 MM |
158 | !(op->upcall.type == |
159 | ORANGEFS_VFS_OP_FS_MOUNT || | |
160 | op->upcall.type == | |
161 | ORANGEFS_VFS_OP_GETATTR)) { | |
24c8d080 MB |
162 | gossip_debug(GOSSIP_DEV_DEBUG, |
163 | "orangefs: skipping op tag %llu %s\n", | |
164 | llu(op->tag), get_opname_string(op)); | |
165 | gossip_err( | |
166 | "orangefs: ERROR: fs_mount_pending %d\n", | |
167 | fsid); | |
168 | spin_unlock(&op->lock); | |
5db11c21 | 169 | continue; |
5db11c21 MM |
170 | } |
171 | } | |
24c8d080 MB |
172 | /* |
173 | * Either this op does not pertain to a filesystem, is mounting | |
174 | * a filesystem, or pertains to a mounted filesystem. Let it | |
175 | * through. | |
176 | */ | |
177 | cur_op = op; | |
178 | break; | |
179 | } | |
180 | ||
181 | /* | |
182 | * At this point we either have a valid op and can continue or have not | |
183 | * found an op and must ask the client to try again later. | |
184 | */ | |
185 | if (!cur_op) { | |
8bb8aefd | 186 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 187 | return -EAGAIN; |
5db11c21 MM |
188 | } |
189 | ||
24c8d080 MB |
190 | gossip_debug(GOSSIP_DEV_DEBUG, "orangefs: reading op tag %llu %s\n", |
191 | llu(cur_op->tag), get_opname_string(cur_op)); | |
5db11c21 | 192 | |
24c8d080 MB |
193 | /* |
194 | * Such an op should never be on the list in the first place. If so, we | |
195 | * will abort. | |
196 | */ | |
197 | if (op_state_in_progress(cur_op) || op_state_serviced(cur_op)) { | |
198 | gossip_err("orangefs: ERROR: Current op already queued.\n"); | |
199 | list_del(&cur_op->list); | |
5db11c21 | 200 | spin_unlock(&cur_op->lock); |
8bb8aefd | 201 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 202 | return -EAGAIN; |
5db11c21 | 203 | } |
ed42fe05 AV |
204 | list_del_init(&cur_op->list); |
205 | get_op(op); | |
8bb8aefd | 206 | spin_unlock(&orangefs_request_list_lock); |
ed42fe05 | 207 | |
24c8d080 MB |
208 | spin_unlock(&cur_op->lock); |
209 | ||
210 | /* Push the upcall out. */ | |
211 | ret = copy_to_user(buf, &proto_ver, sizeof(__s32)); | |
212 | if (ret != 0) | |
213 | goto error; | |
214 | ret = copy_to_user(buf+sizeof(__s32), &magic, sizeof(__s32)); | |
215 | if (ret != 0) | |
216 | goto error; | |
217 | ret = copy_to_user(buf+2 * sizeof(__s32), &cur_op->tag, sizeof(__u64)); | |
218 | if (ret != 0) | |
219 | goto error; | |
220 | ret = copy_to_user(buf+2*sizeof(__s32)+sizeof(__u64), &cur_op->upcall, | |
8bb8aefd | 221 | sizeof(struct orangefs_upcall_s)); |
24c8d080 MB |
222 | if (ret != 0) |
223 | goto error; | |
224 | ||
ed42fe05 AV |
225 | spin_lock(&htable_ops_in_progress_lock); |
226 | spin_lock(&cur_op->lock); | |
227 | if (unlikely(op_state_given_up(cur_op))) { | |
228 | spin_unlock(&cur_op->lock); | |
229 | spin_unlock(&htable_ops_in_progress_lock); | |
230 | op_release(cur_op); | |
231 | goto restart; | |
232 | } | |
233 | ||
234 | /* | |
235 | * Set the operation to be in progress and move it between lists since | |
236 | * it has been sent to the client. | |
237 | */ | |
238 | set_op_state_inprogress(cur_op); | |
239 | orangefs_devreq_add_op(cur_op); | |
240 | spin_unlock(&cur_op->lock); | |
241 | spin_unlock(&htable_ops_in_progress_lock); | |
242 | op_release(cur_op); | |
243 | ||
24c8d080 | 244 | /* The client only asks to read one size buffer. */ |
a762ae6d | 245 | return MAX_DEV_REQ_UPSIZE; |
24c8d080 MB |
246 | error: |
247 | /* | |
248 | * We were unable to copy the op data to the client. Put the op back in | |
249 | * list. If client has crashed, the op will be purged later when the | |
250 | * device is released. | |
251 | */ | |
252 | gossip_err("orangefs: Failed to copy data to user space\n"); | |
8bb8aefd | 253 | spin_lock(&orangefs_request_list_lock); |
24c8d080 | 254 | spin_lock(&cur_op->lock); |
ed42fe05 AV |
255 | if (likely(!op_state_given_up(cur_op))) { |
256 | set_op_state_waiting(cur_op); | |
257 | list_add(&cur_op->list, &orangefs_request_list); | |
258 | } | |
24c8d080 | 259 | spin_unlock(&cur_op->lock); |
8bb8aefd | 260 | spin_unlock(&orangefs_request_list_lock); |
ed42fe05 | 261 | op_release(cur_op); |
24c8d080 | 262 | return -EFAULT; |
5db11c21 MM |
263 | } |
264 | ||
97f10027 | 265 | /* |
b3ae4755 MM |
266 | * Function for writev() callers into the device. |
267 | * | |
268 | * Userspace should have written: | |
269 | * - __u32 version | |
270 | * - __u32 magic | |
271 | * - __u64 tag | |
272 | * - struct orangefs_downcall_s | |
273 | * - trailer buffer (in the case of READDIR operations) | |
97f10027 | 274 | */ |
b3ae4755 MM |
275 | static ssize_t orangefs_devreq_write_iter(struct kiocb *iocb, |
276 | struct iov_iter *iter) | |
5db11c21 | 277 | { |
b3ae4755 | 278 | ssize_t ret; |
8bb8aefd | 279 | struct orangefs_kernel_op_s *op = NULL; |
b3ae4755 MM |
280 | struct { |
281 | __u32 version; | |
282 | __u32 magic; | |
283 | __u64 tag; | |
284 | } head; | |
285 | int total = ret = iov_iter_count(iter); | |
286 | int n; | |
287 | int downcall_size = sizeof(struct orangefs_downcall_s); | |
288 | int head_size = sizeof(head); | |
289 | ||
290 | gossip_debug(GOSSIP_DEV_DEBUG, "%s: total:%d: ret:%zd:\n", | |
291 | __func__, | |
292 | total, | |
293 | ret); | |
5db11c21 | 294 | |
b3ae4755 | 295 | if (total < MAX_DEV_REQ_DOWNSIZE) { |
cf0c2771 | 296 | gossip_err("%s: total:%d: must be at least:%u:\n", |
b3ae4755 MM |
297 | __func__, |
298 | total, | |
cf0c2771 | 299 | (unsigned int) MAX_DEV_REQ_DOWNSIZE); |
ed42fe05 | 300 | return -EFAULT; |
5db11c21 | 301 | } |
b3ae4755 MM |
302 | |
303 | n = copy_from_iter(&head, head_size, iter); | |
304 | if (n < head_size) { | |
305 | gossip_err("%s: failed to copy head.\n", __func__); | |
ed42fe05 | 306 | return -EFAULT; |
97f10027 | 307 | } |
b3ae4755 MM |
308 | |
309 | if (head.version < ORANGEFS_MINIMUM_USERSPACE_VERSION) { | |
310 | gossip_err("%s: userspace claims version" | |
311 | "%d, minimum version required: %d.\n", | |
312 | __func__, | |
313 | head.version, | |
314 | ORANGEFS_MINIMUM_USERSPACE_VERSION); | |
ed42fe05 | 315 | return -EPROTO; |
5db11c21 | 316 | } |
5db11c21 | 317 | |
b3ae4755 MM |
318 | if (head.magic != ORANGEFS_DEVREQ_MAGIC) { |
319 | gossip_err("Error: Device magic number does not match.\n"); | |
ed42fe05 | 320 | return -EPROTO; |
b3ae4755 | 321 | } |
5db11c21 | 322 | |
b3ae4755 MM |
323 | op = orangefs_devreq_remove_op(head.tag); |
324 | if (!op) { | |
325 | gossip_err("WARNING: No one's waiting for tag %llu\n", | |
326 | llu(head.tag)); | |
ed42fe05 | 327 | return ret; |
b3ae4755 | 328 | } |
5db11c21 | 329 | |
b3ae4755 MM |
330 | n = copy_from_iter(&op->downcall, downcall_size, iter); |
331 | if (n != downcall_size) { | |
332 | gossip_err("%s: failed to copy downcall.\n", __func__); | |
b3ae4755 | 333 | ret = -EFAULT; |
ed42fe05 | 334 | goto Broken; |
5db11c21 MM |
335 | } |
336 | ||
b3ae4755 MM |
337 | if (op->downcall.status) |
338 | goto wakeup; | |
97f10027 | 339 | |
b3ae4755 MM |
340 | /* |
341 | * We've successfully peeled off the head and the downcall. | |
342 | * Something has gone awry if total doesn't equal the | |
343 | * sum of head_size, downcall_size and trailer_size. | |
344 | */ | |
345 | if ((head_size + downcall_size + op->downcall.trailer_size) != total) { | |
346 | gossip_err("%s: funky write, head_size:%d" | |
347 | ": downcall_size:%d: trailer_size:%lld" | |
348 | ": total size:%d:\n", | |
349 | __func__, | |
350 | head_size, | |
351 | downcall_size, | |
352 | op->downcall.trailer_size, | |
353 | total); | |
b3ae4755 | 354 | ret = -EFAULT; |
ed42fe05 | 355 | goto Broken; |
b3ae4755 | 356 | } |
97f10027 | 357 | |
b3ae4755 MM |
358 | /* Only READDIR operations should have trailers. */ |
359 | if ((op->downcall.type != ORANGEFS_VFS_OP_READDIR) && | |
360 | (op->downcall.trailer_size != 0)) { | |
361 | gossip_err("%s: %x operation with trailer.", | |
362 | __func__, | |
363 | op->downcall.type); | |
b3ae4755 | 364 | ret = -EFAULT; |
ed42fe05 | 365 | goto Broken; |
b3ae4755 | 366 | } |
97f10027 | 367 | |
b3ae4755 MM |
368 | /* READDIR operations should always have trailers. */ |
369 | if ((op->downcall.type == ORANGEFS_VFS_OP_READDIR) && | |
370 | (op->downcall.trailer_size == 0)) { | |
371 | gossip_err("%s: %x operation with no trailer.", | |
372 | __func__, | |
373 | op->downcall.type); | |
b3ae4755 | 374 | ret = -EFAULT; |
ed42fe05 | 375 | goto Broken; |
b3ae4755 | 376 | } |
97f10027 | 377 | |
b3ae4755 MM |
378 | if (op->downcall.type != ORANGEFS_VFS_OP_READDIR) |
379 | goto wakeup; | |
5db11c21 | 380 | |
b3ae4755 MM |
381 | op->downcall.trailer_buf = |
382 | vmalloc(op->downcall.trailer_size); | |
383 | if (op->downcall.trailer_buf == NULL) { | |
384 | gossip_err("%s: failed trailer vmalloc.\n", | |
385 | __func__); | |
b3ae4755 | 386 | ret = -ENOMEM; |
ed42fe05 | 387 | goto Broken; |
b3ae4755 MM |
388 | } |
389 | memset(op->downcall.trailer_buf, 0, op->downcall.trailer_size); | |
390 | n = copy_from_iter(op->downcall.trailer_buf, | |
391 | op->downcall.trailer_size, | |
392 | iter); | |
393 | if (n != op->downcall.trailer_size) { | |
394 | gossip_err("%s: failed to copy trailer.\n", __func__); | |
395 | vfree(op->downcall.trailer_buf); | |
b3ae4755 | 396 | ret = -EFAULT; |
ed42fe05 | 397 | goto Broken; |
b3ae4755 | 398 | } |
97f10027 | 399 | |
b3ae4755 | 400 | wakeup: |
97f10027 | 401 | |
b3ae4755 MM |
402 | /* |
403 | * If this operation is an I/O operation we need to wait | |
404 | * for all data to be copied before we can return to avoid | |
405 | * buffer corruption and races that can pull the buffers | |
406 | * out from under us. | |
407 | * | |
408 | * Essentially we're synchronizing with other parts of the | |
409 | * vfs implicitly by not allowing the user space | |
410 | * application reading/writing this device to return until | |
411 | * the buffers are done being used. | |
412 | */ | |
413 | if (op->downcall.type == ORANGEFS_VFS_OP_FILE_IO) { | |
b3ae4755 | 414 | DEFINE_WAIT(wait_entry); |
97f10027 MM |
415 | |
416 | /* | |
b3ae4755 MM |
417 | * tell the vfs op waiting on a waitqueue |
418 | * that this op is done | |
97f10027 | 419 | */ |
b3ae4755 | 420 | spin_lock(&op->lock); |
4f55e397 AV |
421 | if (unlikely(op_state_given_up(op))) { |
422 | spin_unlock(&op->lock); | |
423 | goto out; | |
424 | } | |
425 | set_op_state_serviced(op); | |
b3ae4755 | 426 | spin_unlock(&op->lock); |
5db11c21 | 427 | |
b3ae4755 | 428 | while (1) { |
5db11c21 | 429 | spin_lock(&op->lock); |
b3ae4755 MM |
430 | prepare_to_wait_exclusive( |
431 | &op->io_completion_waitq, | |
432 | &wait_entry, | |
433 | TASK_INTERRUPTIBLE); | |
434 | if (op->io_completed) { | |
5db11c21 | 435 | spin_unlock(&op->lock); |
5db11c21 MM |
436 | break; |
437 | } | |
ce6c414e | 438 | spin_unlock(&op->lock); |
4f55e397 AV |
439 | if (unlikely(signal_pending(current))) { |
440 | gossip_debug(GOSSIP_DEV_DEBUG, | |
441 | "%s: signal on I/O wait, aborting\n", | |
442 | __func__); | |
443 | break; | |
b3ae4755 | 444 | } |
5db11c21 | 445 | |
4f55e397 AV |
446 | if (!schedule_timeout(op_timeout_secs * HZ)) { |
447 | gossip_debug(GOSSIP_DEV_DEBUG, | |
448 | "%s: timed out.\n", | |
449 | __func__); | |
450 | break; | |
451 | } | |
5db11c21 | 452 | } |
b3ae4755 MM |
453 | |
454 | spin_lock(&op->lock); | |
455 | finish_wait(&op->io_completion_waitq, &wait_entry); | |
456 | spin_unlock(&op->lock); | |
5db11c21 | 457 | } else { |
b3ae4755 MM |
458 | /* |
459 | * tell the vfs op waiting on a waitqueue that | |
831d0949 | 460 | * this op is done - |
b3ae4755 MM |
461 | * for every other operation (i.e. non-I/O), we need to |
462 | * wake up the callers for downcall completion | |
463 | * notification | |
464 | */ | |
831d0949 | 465 | spin_lock(&op->lock); |
ed42fe05 AV |
466 | if (!op_state_given_up(op)) |
467 | set_op_state_serviced(op); | |
831d0949 | 468 | spin_unlock(&op->lock); |
5db11c21 | 469 | } |
b3ae4755 | 470 | out: |
ed42fe05 | 471 | op_release(op); |
b3ae4755 | 472 | return ret; |
ed42fe05 AV |
473 | |
474 | Broken: | |
475 | spin_lock(&op->lock); | |
476 | if (!op_state_given_up(op)) { | |
477 | op->downcall.status = ret; | |
478 | set_op_state_serviced(op); | |
479 | } | |
480 | spin_unlock(&op->lock); | |
481 | goto out; | |
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 | 541 | |
fee25ce1 | 542 | open_access_count = -1; |
5db11c21 MM |
543 | |
544 | unmounted = mark_all_pending_mounts(); | |
8bb8aefd | 545 | gossip_debug(GOSSIP_DEV_DEBUG, "ORANGEFS Device Close: Filesystem(s) %s\n", |
5db11c21 | 546 | (unmounted ? "UNMOUNTED" : "MOUNTED")); |
5db11c21 MM |
547 | |
548 | /* | |
549 | * Walk through the list of ops in the request list, mark them | |
550 | * as purged and wake them up. | |
551 | */ | |
552 | purge_waiting_ops(); | |
553 | /* | |
554 | * Walk through the hash table of in progress operations; mark | |
555 | * them as purged and wake them up | |
556 | */ | |
557 | purge_inprogress_ops(); | |
558 | gossip_debug(GOSSIP_DEV_DEBUG, | |
559 | "pvfs2-client-core: device close complete\n"); | |
fee25ce1 AV |
560 | open_access_count = 0; |
561 | mutex_unlock(&devreq_mutex); | |
5db11c21 MM |
562 | return 0; |
563 | } | |
564 | ||
565 | int is_daemon_in_service(void) | |
566 | { | |
567 | int in_service; | |
568 | ||
569 | /* | |
570 | * What this function does is checks if client-core is alive | |
571 | * based on the access count we maintain on the device. | |
572 | */ | |
573 | mutex_lock(&devreq_mutex); | |
574 | in_service = open_access_count == 1 ? 0 : -EIO; | |
575 | mutex_unlock(&devreq_mutex); | |
576 | return in_service; | |
577 | } | |
578 | ||
579 | static inline long check_ioctl_command(unsigned int command) | |
580 | { | |
581 | /* Check for valid ioctl codes */ | |
8bb8aefd | 582 | if (_IOC_TYPE(command) != ORANGEFS_DEV_MAGIC) { |
5db11c21 MM |
583 | gossip_err("device ioctl magic numbers don't match! Did you rebuild pvfs2-client-core/libpvfs2? [cmd %x, magic %x != %x]\n", |
584 | command, | |
585 | _IOC_TYPE(command), | |
8bb8aefd | 586 | ORANGEFS_DEV_MAGIC); |
5db11c21 MM |
587 | return -EINVAL; |
588 | } | |
589 | /* and valid ioctl commands */ | |
8bb8aefd | 590 | if (_IOC_NR(command) >= ORANGEFS_DEV_MAXNR || _IOC_NR(command) <= 0) { |
5db11c21 | 591 | gossip_err("Invalid ioctl command number [%d >= %d]\n", |
8bb8aefd | 592 | _IOC_NR(command), ORANGEFS_DEV_MAXNR); |
5db11c21 MM |
593 | return -ENOIOCTLCMD; |
594 | } | |
595 | return 0; | |
596 | } | |
597 | ||
598 | static long dispatch_ioctl_command(unsigned int command, unsigned long arg) | |
599 | { | |
8bb8aefd | 600 | static __s32 magic = ORANGEFS_DEVREQ_MAGIC; |
a762ae6d MB |
601 | static __s32 max_up_size = MAX_DEV_REQ_UPSIZE; |
602 | static __s32 max_down_size = MAX_DEV_REQ_DOWNSIZE; | |
8bb8aefd | 603 | struct ORANGEFS_dev_map_desc user_desc; |
5db11c21 MM |
604 | int ret = 0; |
605 | struct dev_mask_info_s mask_info = { 0 }; | |
606 | struct dev_mask2_info_s mask2_info = { 0, 0 }; | |
607 | int upstream_kmod = 1; | |
608 | struct list_head *tmp = NULL; | |
8bb8aefd | 609 | struct orangefs_sb_info_s *orangefs_sb = NULL; |
5db11c21 MM |
610 | |
611 | /* mtmoore: add locking here */ | |
612 | ||
613 | switch (command) { | |
8bb8aefd | 614 | case ORANGEFS_DEV_GET_MAGIC: |
5db11c21 MM |
615 | return ((put_user(magic, (__s32 __user *) arg) == -EFAULT) ? |
616 | -EIO : | |
617 | 0); | |
8bb8aefd | 618 | case ORANGEFS_DEV_GET_MAX_UPSIZE: |
5db11c21 MM |
619 | return ((put_user(max_up_size, |
620 | (__s32 __user *) arg) == -EFAULT) ? | |
621 | -EIO : | |
622 | 0); | |
8bb8aefd | 623 | case ORANGEFS_DEV_GET_MAX_DOWNSIZE: |
5db11c21 MM |
624 | return ((put_user(max_down_size, |
625 | (__s32 __user *) arg) == -EFAULT) ? | |
626 | -EIO : | |
627 | 0); | |
8bb8aefd | 628 | case ORANGEFS_DEV_MAP: |
5db11c21 | 629 | ret = copy_from_user(&user_desc, |
8bb8aefd | 630 | (struct ORANGEFS_dev_map_desc __user *) |
5db11c21 | 631 | arg, |
8bb8aefd | 632 | sizeof(struct ORANGEFS_dev_map_desc)); |
7d221485 | 633 | if (orangefs_get_bufmap_init()) { |
90d26aa8 MB |
634 | return -EINVAL; |
635 | } else { | |
636 | return ret ? | |
637 | -EIO : | |
638 | orangefs_bufmap_initialize(&user_desc); | |
639 | } | |
8bb8aefd | 640 | case ORANGEFS_DEV_REMOUNT_ALL: |
5db11c21 | 641 | gossip_debug(GOSSIP_DEV_DEBUG, |
97f10027 MM |
642 | "%s: got ORANGEFS_DEV_REMOUNT_ALL\n", |
643 | __func__); | |
5db11c21 MM |
644 | |
645 | /* | |
8bb8aefd | 646 | * remount all mounted orangefs volumes to regain the lost |
5db11c21 MM |
647 | * dynamic mount tables (if any) -- NOTE: this is done |
648 | * without keeping the superblock list locked due to the | |
649 | * upcall/downcall waiting. also, the request semaphore is | |
650 | * used to ensure that no operations will be serviced until | |
651 | * all of the remounts are serviced (to avoid ops between | |
652 | * mounts to fail) | |
653 | */ | |
654 | ret = mutex_lock_interruptible(&request_mutex); | |
655 | if (ret < 0) | |
656 | return ret; | |
657 | gossip_debug(GOSSIP_DEV_DEBUG, | |
97f10027 MM |
658 | "%s: priority remount in progress\n", |
659 | __func__); | |
8bb8aefd YL |
660 | list_for_each(tmp, &orangefs_superblocks) { |
661 | orangefs_sb = | |
97f10027 MM |
662 | list_entry(tmp, |
663 | struct orangefs_sb_info_s, | |
664 | list); | |
8bb8aefd | 665 | if (orangefs_sb && (orangefs_sb->sb)) { |
5db11c21 | 666 | gossip_debug(GOSSIP_DEV_DEBUG, |
97f10027 MM |
667 | "%s: Remounting SB %p\n", |
668 | __func__, | |
8bb8aefd | 669 | orangefs_sb); |
5db11c21 | 670 | |
8bb8aefd | 671 | ret = orangefs_remount(orangefs_sb->sb); |
5db11c21 MM |
672 | if (ret) { |
673 | gossip_debug(GOSSIP_DEV_DEBUG, | |
674 | "SB %p remount failed\n", | |
8bb8aefd | 675 | orangefs_sb); |
97f10027 | 676 | break; |
5db11c21 MM |
677 | } |
678 | } | |
679 | } | |
680 | gossip_debug(GOSSIP_DEV_DEBUG, | |
97f10027 MM |
681 | "%s: priority remount complete\n", |
682 | __func__); | |
5db11c21 MM |
683 | mutex_unlock(&request_mutex); |
684 | return ret; | |
685 | ||
8bb8aefd | 686 | case ORANGEFS_DEV_UPSTREAM: |
5db11c21 MM |
687 | ret = copy_to_user((void __user *)arg, |
688 | &upstream_kmod, | |
689 | sizeof(upstream_kmod)); | |
690 | ||
691 | if (ret != 0) | |
692 | return -EIO; | |
693 | else | |
694 | return ret; | |
695 | ||
8bb8aefd | 696 | case ORANGEFS_DEV_CLIENT_MASK: |
5db11c21 MM |
697 | ret = copy_from_user(&mask2_info, |
698 | (void __user *)arg, | |
699 | sizeof(struct dev_mask2_info_s)); | |
700 | ||
701 | if (ret != 0) | |
702 | return -EIO; | |
703 | ||
704 | client_debug_mask.mask1 = mask2_info.mask1_value; | |
705 | client_debug_mask.mask2 = mask2_info.mask2_value; | |
706 | ||
707 | pr_info("%s: client debug mask has been been received " | |
708 | ":%llx: :%llx:\n", | |
709 | __func__, | |
710 | (unsigned long long)client_debug_mask.mask1, | |
711 | (unsigned long long)client_debug_mask.mask2); | |
712 | ||
713 | return ret; | |
714 | ||
8bb8aefd | 715 | case ORANGEFS_DEV_CLIENT_STRING: |
5db11c21 MM |
716 | ret = copy_from_user(&client_debug_array_string, |
717 | (void __user *)arg, | |
8bb8aefd | 718 | ORANGEFS_MAX_DEBUG_STRING_LEN); |
5db11c21 | 719 | if (ret != 0) { |
97f10027 | 720 | pr_info("%s: CLIENT_STRING: copy_from_user failed\n", |
5db11c21 MM |
721 | __func__); |
722 | return -EIO; | |
723 | } | |
724 | ||
97f10027 | 725 | pr_info("%s: client debug array string has been received.\n", |
5db11c21 MM |
726 | __func__); |
727 | ||
728 | if (!help_string_initialized) { | |
729 | ||
730 | /* Free the "we don't know yet" default string... */ | |
731 | kfree(debug_help_string); | |
732 | ||
733 | /* build a proper debug help string */ | |
734 | if (orangefs_prepare_debugfs_help_string(0)) { | |
97f10027 | 735 | gossip_err("%s: no debug help string \n", |
5db11c21 MM |
736 | __func__); |
737 | return -EIO; | |
738 | } | |
739 | ||
740 | /* Replace the boilerplate boot-time debug-help file. */ | |
741 | debugfs_remove(help_file_dentry); | |
742 | ||
743 | help_file_dentry = | |
744 | debugfs_create_file( | |
745 | ORANGEFS_KMOD_DEBUG_HELP_FILE, | |
746 | 0444, | |
747 | debug_dir, | |
748 | debug_help_string, | |
749 | &debug_help_fops); | |
750 | ||
751 | if (!help_file_dentry) { | |
752 | gossip_err("%s: debugfs_create_file failed for" | |
753 | " :%s:!\n", | |
754 | __func__, | |
755 | ORANGEFS_KMOD_DEBUG_HELP_FILE); | |
756 | return -EIO; | |
757 | } | |
758 | } | |
759 | ||
760 | debug_mask_to_string(&client_debug_mask, 1); | |
761 | ||
762 | debugfs_remove(client_debug_dentry); | |
763 | ||
8bb8aefd | 764 | orangefs_client_debug_init(); |
5db11c21 MM |
765 | |
766 | help_string_initialized++; | |
767 | ||
768 | return ret; | |
769 | ||
8bb8aefd | 770 | case ORANGEFS_DEV_DEBUG: |
5db11c21 MM |
771 | ret = copy_from_user(&mask_info, |
772 | (void __user *)arg, | |
773 | sizeof(mask_info)); | |
774 | ||
775 | if (ret != 0) | |
776 | return -EIO; | |
777 | ||
778 | if (mask_info.mask_type == KERNEL_MASK) { | |
779 | if ((mask_info.mask_value == 0) | |
780 | && (kernel_mask_set_mod_init)) { | |
781 | /* | |
782 | * the kernel debug mask was set when the | |
783 | * kernel module was loaded; don't override | |
784 | * it if the client-core was started without | |
8bb8aefd | 785 | * a value for ORANGEFS_KMODMASK. |
5db11c21 MM |
786 | */ |
787 | return 0; | |
788 | } | |
789 | debug_mask_to_string(&mask_info.mask_value, | |
790 | mask_info.mask_type); | |
791 | gossip_debug_mask = mask_info.mask_value; | |
97f10027 | 792 | pr_info("%s: kernel debug mask has been modified to " |
5db11c21 | 793 | ":%s: :%llx:\n", |
97f10027 | 794 | __func__, |
5db11c21 MM |
795 | kernel_debug_string, |
796 | (unsigned long long)gossip_debug_mask); | |
797 | } else if (mask_info.mask_type == CLIENT_MASK) { | |
798 | debug_mask_to_string(&mask_info.mask_value, | |
799 | mask_info.mask_type); | |
97f10027 | 800 | pr_info("%s: client debug mask has been modified to" |
5db11c21 | 801 | ":%s: :%llx:\n", |
97f10027 | 802 | __func__, |
5db11c21 MM |
803 | client_debug_string, |
804 | llu(mask_info.mask_value)); | |
805 | } else { | |
806 | gossip_lerr("Invalid mask type....\n"); | |
807 | return -EINVAL; | |
808 | } | |
809 | ||
810 | return ret; | |
811 | ||
812 | default: | |
813 | return -ENOIOCTLCMD; | |
814 | } | |
815 | return -ENOIOCTLCMD; | |
816 | } | |
817 | ||
8bb8aefd | 818 | static long orangefs_devreq_ioctl(struct file *file, |
5db11c21 MM |
819 | unsigned int command, unsigned long arg) |
820 | { | |
821 | long ret; | |
822 | ||
823 | /* Check for properly constructed commands */ | |
824 | ret = check_ioctl_command(command); | |
825 | if (ret < 0) | |
826 | return (int)ret; | |
827 | ||
828 | return (int)dispatch_ioctl_command(command, arg); | |
829 | } | |
830 | ||
831 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
832 | ||
8bb8aefd YL |
833 | /* Compat structure for the ORANGEFS_DEV_MAP ioctl */ |
834 | struct ORANGEFS_dev_map_desc32 { | |
5db11c21 MM |
835 | compat_uptr_t ptr; |
836 | __s32 total_size; | |
837 | __s32 size; | |
838 | __s32 count; | |
839 | }; | |
840 | ||
841 | static unsigned long translate_dev_map26(unsigned long args, long *error) | |
842 | { | |
8bb8aefd | 843 | struct ORANGEFS_dev_map_desc32 __user *p32 = (void __user *)args; |
5db11c21 MM |
844 | /* |
845 | * Depending on the architecture, allocate some space on the | |
846 | * user-call-stack based on our expected layout. | |
847 | */ | |
8bb8aefd | 848 | struct ORANGEFS_dev_map_desc __user *p = |
5db11c21 | 849 | compat_alloc_user_space(sizeof(*p)); |
84d02150 | 850 | compat_uptr_t addr; |
5db11c21 MM |
851 | |
852 | *error = 0; | |
853 | /* get the ptr from the 32 bit user-space */ | |
854 | if (get_user(addr, &p32->ptr)) | |
855 | goto err; | |
856 | /* try to put that into a 64-bit layout */ | |
857 | if (put_user(compat_ptr(addr), &p->ptr)) | |
858 | goto err; | |
859 | /* copy the remaining fields */ | |
860 | if (copy_in_user(&p->total_size, &p32->total_size, sizeof(__s32))) | |
861 | goto err; | |
862 | if (copy_in_user(&p->size, &p32->size, sizeof(__s32))) | |
863 | goto err; | |
864 | if (copy_in_user(&p->count, &p32->count, sizeof(__s32))) | |
865 | goto err; | |
866 | return (unsigned long)p; | |
867 | err: | |
868 | *error = -EFAULT; | |
869 | return 0; | |
870 | } | |
871 | ||
872 | /* | |
873 | * 32 bit user-space apps' ioctl handlers when kernel modules | |
874 | * is compiled as a 64 bit one | |
875 | */ | |
8bb8aefd | 876 | static long orangefs_devreq_compat_ioctl(struct file *filp, unsigned int cmd, |
5db11c21 MM |
877 | unsigned long args) |
878 | { | |
879 | long ret; | |
880 | unsigned long arg = args; | |
881 | ||
882 | /* Check for properly constructed commands */ | |
883 | ret = check_ioctl_command(cmd); | |
884 | if (ret < 0) | |
885 | return ret; | |
8bb8aefd | 886 | if (cmd == ORANGEFS_DEV_MAP) { |
5db11c21 MM |
887 | /* |
888 | * convert the arguments to what we expect internally | |
889 | * in kernel space | |
890 | */ | |
891 | arg = translate_dev_map26(args, &ret); | |
892 | if (ret < 0) { | |
893 | gossip_err("Could not translate dev map\n"); | |
894 | return ret; | |
895 | } | |
896 | } | |
897 | /* no other ioctl requires translation */ | |
898 | return dispatch_ioctl_command(cmd, arg); | |
899 | } | |
900 | ||
2c590d5f MM |
901 | #endif /* CONFIG_COMPAT is in .config */ |
902 | ||
5db11c21 | 903 | /* the assigned character device major number */ |
8bb8aefd | 904 | static int orangefs_dev_major; |
5db11c21 MM |
905 | |
906 | /* | |
8bb8aefd | 907 | * Initialize orangefs device specific state: |
5db11c21 MM |
908 | * Must be called at module load time only |
909 | */ | |
8bb8aefd | 910 | int orangefs_dev_init(void) |
5db11c21 | 911 | { |
8bb8aefd YL |
912 | /* register orangefs-req device */ |
913 | orangefs_dev_major = register_chrdev(0, | |
914 | ORANGEFS_REQDEVICE_NAME, | |
915 | &orangefs_devreq_file_operations); | |
916 | if (orangefs_dev_major < 0) { | |
5db11c21 MM |
917 | gossip_debug(GOSSIP_DEV_DEBUG, |
918 | "Failed to register /dev/%s (error %d)\n", | |
8bb8aefd | 919 | ORANGEFS_REQDEVICE_NAME, orangefs_dev_major); |
8bb8aefd | 920 | return orangefs_dev_major; |
5db11c21 MM |
921 | } |
922 | ||
923 | gossip_debug(GOSSIP_DEV_DEBUG, | |
924 | "*** /dev/%s character device registered ***\n", | |
8bb8aefd | 925 | ORANGEFS_REQDEVICE_NAME); |
5db11c21 | 926 | gossip_debug(GOSSIP_DEV_DEBUG, "'mknod /dev/%s c %d 0'.\n", |
8bb8aefd | 927 | ORANGEFS_REQDEVICE_NAME, orangefs_dev_major); |
5db11c21 MM |
928 | return 0; |
929 | } | |
930 | ||
8bb8aefd | 931 | void orangefs_dev_cleanup(void) |
5db11c21 | 932 | { |
8bb8aefd | 933 | unregister_chrdev(orangefs_dev_major, ORANGEFS_REQDEVICE_NAME); |
5db11c21 MM |
934 | gossip_debug(GOSSIP_DEV_DEBUG, |
935 | "*** /dev/%s character device unregistered ***\n", | |
8bb8aefd | 936 | ORANGEFS_REQDEVICE_NAME); |
5db11c21 MM |
937 | } |
938 | ||
8bb8aefd | 939 | static unsigned int orangefs_devreq_poll(struct file *file, |
5db11c21 MM |
940 | struct poll_table_struct *poll_table) |
941 | { | |
942 | int poll_revent_mask = 0; | |
943 | ||
83595db0 | 944 | poll_wait(file, &orangefs_request_list_waitq, poll_table); |
5db11c21 | 945 | |
83595db0 AV |
946 | if (!list_empty(&orangefs_request_list)) |
947 | poll_revent_mask |= POLL_IN; | |
5db11c21 MM |
948 | return poll_revent_mask; |
949 | } | |
950 | ||
8bb8aefd | 951 | const struct file_operations orangefs_devreq_file_operations = { |
5db11c21 | 952 | .owner = THIS_MODULE, |
8bb8aefd YL |
953 | .read = orangefs_devreq_read, |
954 | .write_iter = orangefs_devreq_write_iter, | |
955 | .open = orangefs_devreq_open, | |
956 | .release = orangefs_devreq_release, | |
957 | .unlocked_ioctl = orangefs_devreq_ioctl, | |
5db11c21 MM |
958 | |
959 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
8bb8aefd | 960 | .compat_ioctl = orangefs_devreq_compat_ioctl, |
5db11c21 | 961 | #endif |
8bb8aefd | 962 | .poll = orangefs_devreq_poll |
5db11c21 | 963 | }; |