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) { | |
fee25ce1 | 88 | open_access_count = 1; |
fb6d2526 | 89 | ret = 0; |
5db11c21 MM |
90 | } else { |
91 | DUMP_DEVICE_ERROR(); | |
92 | } | |
93 | mutex_unlock(&devreq_mutex); | |
94 | ||
95 | out: | |
96 | ||
97 | gossip_debug(GOSSIP_DEV_DEBUG, | |
98 | "pvfs2-client-core: open device complete (ret = %d)\n", | |
99 | ret); | |
100 | return ret; | |
101 | } | |
102 | ||
97f10027 | 103 | /* Function for read() callers into the device */ |
8bb8aefd | 104 | static ssize_t orangefs_devreq_read(struct file *file, |
5db11c21 MM |
105 | char __user *buf, |
106 | size_t count, loff_t *offset) | |
107 | { | |
8bb8aefd YL |
108 | struct orangefs_kernel_op_s *op, *temp; |
109 | __s32 proto_ver = ORANGEFS_KERNEL_PROTO_VERSION; | |
110 | static __s32 magic = ORANGEFS_DEVREQ_MAGIC; | |
111 | struct orangefs_kernel_op_s *cur_op = NULL; | |
24c8d080 | 112 | unsigned long ret; |
5db11c21 | 113 | |
24c8d080 | 114 | /* We do not support blocking IO. */ |
5db11c21 | 115 | if (!(file->f_flags & O_NONBLOCK)) { |
97f10027 MM |
116 | gossip_err("%s: blocking read from client-core.\n", |
117 | __func__); | |
5db11c21 | 118 | return -EINVAL; |
24c8d080 MB |
119 | } |
120 | ||
121 | /* | |
a762ae6d | 122 | * The client will do an ioctl to find MAX_DEV_REQ_UPSIZE, then |
24c8d080 MB |
123 | * always read with that size buffer. |
124 | */ | |
a762ae6d | 125 | if (count != MAX_DEV_REQ_UPSIZE) { |
24c8d080 MB |
126 | gossip_err("orangefs: client-core tried to read wrong size\n"); |
127 | return -EINVAL; | |
128 | } | |
129 | ||
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); | |
136 | ||
137 | fsid = fsid_of_op(op); | |
8bb8aefd | 138 | if (fsid != ORANGEFS_FS_ID_NULL) { |
24c8d080 MB |
139 | int ret; |
140 | /* Skip ops whose filesystem needs to be mounted. */ | |
141 | ret = fs_mount_pending(fsid); | |
142 | if (ret == 1) { | |
5db11c21 | 143 | gossip_debug(GOSSIP_DEV_DEBUG, |
24c8d080 MB |
144 | "orangefs: skipping op tag %llu %s\n", |
145 | llu(op->tag), get_opname_string(op)); | |
146 | spin_unlock(&op->lock); | |
147 | continue; | |
97f10027 MM |
148 | /* |
149 | * Skip ops whose filesystem we don't know about unless | |
150 | * it is being mounted. | |
151 | */ | |
24c8d080 MB |
152 | /* XXX: is there a better way to detect this? */ |
153 | } else if (ret == -1 && | |
97f10027 MM |
154 | !(op->upcall.type == |
155 | ORANGEFS_VFS_OP_FS_MOUNT || | |
156 | op->upcall.type == | |
157 | ORANGEFS_VFS_OP_GETATTR)) { | |
24c8d080 MB |
158 | gossip_debug(GOSSIP_DEV_DEBUG, |
159 | "orangefs: skipping op tag %llu %s\n", | |
160 | llu(op->tag), get_opname_string(op)); | |
161 | gossip_err( | |
162 | "orangefs: ERROR: fs_mount_pending %d\n", | |
163 | fsid); | |
164 | spin_unlock(&op->lock); | |
5db11c21 | 165 | continue; |
5db11c21 MM |
166 | } |
167 | } | |
24c8d080 MB |
168 | /* |
169 | * Either this op does not pertain to a filesystem, is mounting | |
170 | * a filesystem, or pertains to a mounted filesystem. Let it | |
171 | * through. | |
172 | */ | |
173 | cur_op = op; | |
174 | break; | |
175 | } | |
176 | ||
177 | /* | |
178 | * At this point we either have a valid op and can continue or have not | |
179 | * found an op and must ask the client to try again later. | |
180 | */ | |
181 | if (!cur_op) { | |
8bb8aefd | 182 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 183 | return -EAGAIN; |
5db11c21 MM |
184 | } |
185 | ||
24c8d080 MB |
186 | gossip_debug(GOSSIP_DEV_DEBUG, "orangefs: reading op tag %llu %s\n", |
187 | llu(cur_op->tag), get_opname_string(cur_op)); | |
5db11c21 | 188 | |
24c8d080 MB |
189 | /* |
190 | * Such an op should never be on the list in the first place. If so, we | |
191 | * will abort. | |
192 | */ | |
193 | if (op_state_in_progress(cur_op) || op_state_serviced(cur_op)) { | |
194 | gossip_err("orangefs: ERROR: Current op already queued.\n"); | |
195 | list_del(&cur_op->list); | |
5db11c21 | 196 | spin_unlock(&cur_op->lock); |
8bb8aefd | 197 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 198 | return -EAGAIN; |
5db11c21 | 199 | } |
24c8d080 MB |
200 | |
201 | /* | |
202 | * Set the operation to be in progress and move it between lists since | |
203 | * it has been sent to the client. | |
204 | */ | |
205 | set_op_state_inprogress(cur_op); | |
206 | ||
207 | list_del(&cur_op->list); | |
8bb8aefd YL |
208 | spin_unlock(&orangefs_request_list_lock); |
209 | orangefs_devreq_add_op(cur_op); | |
24c8d080 MB |
210 | spin_unlock(&cur_op->lock); |
211 | ||
212 | /* Push the upcall out. */ | |
213 | ret = copy_to_user(buf, &proto_ver, sizeof(__s32)); | |
214 | if (ret != 0) | |
215 | goto error; | |
216 | ret = copy_to_user(buf+sizeof(__s32), &magic, sizeof(__s32)); | |
217 | if (ret != 0) | |
218 | goto error; | |
219 | ret = copy_to_user(buf+2 * sizeof(__s32), &cur_op->tag, sizeof(__u64)); | |
220 | if (ret != 0) | |
221 | goto error; | |
222 | ret = copy_to_user(buf+2*sizeof(__s32)+sizeof(__u64), &cur_op->upcall, | |
8bb8aefd | 223 | sizeof(struct orangefs_upcall_s)); |
24c8d080 MB |
224 | if (ret != 0) |
225 | goto error; | |
226 | ||
227 | /* The client only asks to read one size buffer. */ | |
a762ae6d | 228 | return MAX_DEV_REQ_UPSIZE; |
24c8d080 MB |
229 | error: |
230 | /* | |
231 | * We were unable to copy the op data to the client. Put the op back in | |
232 | * list. If client has crashed, the op will be purged later when the | |
233 | * device is released. | |
234 | */ | |
235 | gossip_err("orangefs: Failed to copy data to user space\n"); | |
8bb8aefd | 236 | spin_lock(&orangefs_request_list_lock); |
24c8d080 MB |
237 | spin_lock(&cur_op->lock); |
238 | set_op_state_waiting(cur_op); | |
8bb8aefd YL |
239 | orangefs_devreq_remove_op(cur_op->tag); |
240 | list_add(&cur_op->list, &orangefs_request_list); | |
24c8d080 | 241 | spin_unlock(&cur_op->lock); |
8bb8aefd | 242 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 243 | return -EFAULT; |
5db11c21 MM |
244 | } |
245 | ||
97f10027 | 246 | /* |
b3ae4755 MM |
247 | * Function for writev() callers into the device. |
248 | * | |
249 | * Userspace should have written: | |
250 | * - __u32 version | |
251 | * - __u32 magic | |
252 | * - __u64 tag | |
253 | * - struct orangefs_downcall_s | |
254 | * - trailer buffer (in the case of READDIR operations) | |
97f10027 | 255 | */ |
b3ae4755 MM |
256 | static ssize_t orangefs_devreq_write_iter(struct kiocb *iocb, |
257 | struct iov_iter *iter) | |
5db11c21 | 258 | { |
b3ae4755 | 259 | ssize_t ret; |
8bb8aefd | 260 | struct orangefs_kernel_op_s *op = NULL; |
b3ae4755 MM |
261 | struct { |
262 | __u32 version; | |
263 | __u32 magic; | |
264 | __u64 tag; | |
265 | } head; | |
266 | int total = ret = iov_iter_count(iter); | |
267 | int n; | |
268 | int downcall_size = sizeof(struct orangefs_downcall_s); | |
269 | int head_size = sizeof(head); | |
270 | ||
271 | gossip_debug(GOSSIP_DEV_DEBUG, "%s: total:%d: ret:%zd:\n", | |
272 | __func__, | |
273 | total, | |
274 | ret); | |
5db11c21 | 275 | |
b3ae4755 | 276 | if (total < MAX_DEV_REQ_DOWNSIZE) { |
cf0c2771 | 277 | gossip_err("%s: total:%d: must be at least:%u:\n", |
b3ae4755 MM |
278 | __func__, |
279 | total, | |
cf0c2771 | 280 | (unsigned int) MAX_DEV_REQ_DOWNSIZE); |
b3ae4755 MM |
281 | ret = -EFAULT; |
282 | goto out; | |
5db11c21 | 283 | } |
b3ae4755 MM |
284 | |
285 | n = copy_from_iter(&head, head_size, iter); | |
286 | if (n < head_size) { | |
287 | gossip_err("%s: failed to copy head.\n", __func__); | |
288 | ret = -EFAULT; | |
289 | goto out; | |
97f10027 | 290 | } |
b3ae4755 MM |
291 | |
292 | if (head.version < ORANGEFS_MINIMUM_USERSPACE_VERSION) { | |
293 | gossip_err("%s: userspace claims version" | |
294 | "%d, minimum version required: %d.\n", | |
295 | __func__, | |
296 | head.version, | |
297 | ORANGEFS_MINIMUM_USERSPACE_VERSION); | |
298 | ret = -EPROTO; | |
299 | goto out; | |
5db11c21 | 300 | } |
5db11c21 | 301 | |
b3ae4755 MM |
302 | if (head.magic != ORANGEFS_DEVREQ_MAGIC) { |
303 | gossip_err("Error: Device magic number does not match.\n"); | |
304 | ret = -EPROTO; | |
305 | goto out; | |
306 | } | |
5db11c21 | 307 | |
b3ae4755 MM |
308 | op = orangefs_devreq_remove_op(head.tag); |
309 | if (!op) { | |
310 | gossip_err("WARNING: No one's waiting for tag %llu\n", | |
311 | llu(head.tag)); | |
312 | goto out; | |
313 | } | |
5db11c21 | 314 | |
b3ae4755 | 315 | get_op(op); /* increase ref count. */ |
5db11c21 | 316 | |
b3ae4755 MM |
317 | n = copy_from_iter(&op->downcall, downcall_size, iter); |
318 | if (n != downcall_size) { | |
319 | gossip_err("%s: failed to copy downcall.\n", __func__); | |
320 | put_op(op); | |
321 | ret = -EFAULT; | |
322 | goto out; | |
5db11c21 MM |
323 | } |
324 | ||
b3ae4755 MM |
325 | if (op->downcall.status) |
326 | goto wakeup; | |
97f10027 | 327 | |
b3ae4755 MM |
328 | /* |
329 | * We've successfully peeled off the head and the downcall. | |
330 | * Something has gone awry if total doesn't equal the | |
331 | * sum of head_size, downcall_size and trailer_size. | |
332 | */ | |
333 | if ((head_size + downcall_size + op->downcall.trailer_size) != total) { | |
334 | gossip_err("%s: funky write, head_size:%d" | |
335 | ": downcall_size:%d: trailer_size:%lld" | |
336 | ": total size:%d:\n", | |
337 | __func__, | |
338 | head_size, | |
339 | downcall_size, | |
340 | op->downcall.trailer_size, | |
341 | total); | |
342 | put_op(op); | |
343 | ret = -EFAULT; | |
344 | goto out; | |
345 | } | |
97f10027 | 346 | |
b3ae4755 MM |
347 | /* Only READDIR operations should have trailers. */ |
348 | if ((op->downcall.type != ORANGEFS_VFS_OP_READDIR) && | |
349 | (op->downcall.trailer_size != 0)) { | |
350 | gossip_err("%s: %x operation with trailer.", | |
351 | __func__, | |
352 | op->downcall.type); | |
353 | put_op(op); | |
354 | ret = -EFAULT; | |
355 | goto out; | |
356 | } | |
97f10027 | 357 | |
b3ae4755 MM |
358 | /* READDIR operations should always have trailers. */ |
359 | if ((op->downcall.type == ORANGEFS_VFS_OP_READDIR) && | |
360 | (op->downcall.trailer_size == 0)) { | |
361 | gossip_err("%s: %x operation with no trailer.", | |
362 | __func__, | |
363 | op->downcall.type); | |
364 | put_op(op); | |
365 | ret = -EFAULT; | |
366 | goto out; | |
367 | } | |
97f10027 | 368 | |
b3ae4755 MM |
369 | if (op->downcall.type != ORANGEFS_VFS_OP_READDIR) |
370 | goto wakeup; | |
5db11c21 | 371 | |
b3ae4755 MM |
372 | op->downcall.trailer_buf = |
373 | vmalloc(op->downcall.trailer_size); | |
374 | if (op->downcall.trailer_buf == NULL) { | |
375 | gossip_err("%s: failed trailer vmalloc.\n", | |
376 | __func__); | |
377 | put_op(op); | |
378 | ret = -ENOMEM; | |
379 | goto out; | |
380 | } | |
381 | memset(op->downcall.trailer_buf, 0, op->downcall.trailer_size); | |
382 | n = copy_from_iter(op->downcall.trailer_buf, | |
383 | op->downcall.trailer_size, | |
384 | iter); | |
385 | if (n != op->downcall.trailer_size) { | |
386 | gossip_err("%s: failed to copy trailer.\n", __func__); | |
387 | vfree(op->downcall.trailer_buf); | |
388 | put_op(op); | |
389 | ret = -EFAULT; | |
390 | goto out; | |
391 | } | |
97f10027 | 392 | |
b3ae4755 | 393 | wakeup: |
97f10027 | 394 | |
b3ae4755 MM |
395 | /* |
396 | * If this operation is an I/O operation we need to wait | |
397 | * for all data to be copied before we can return to avoid | |
398 | * buffer corruption and races that can pull the buffers | |
399 | * out from under us. | |
400 | * | |
401 | * Essentially we're synchronizing with other parts of the | |
402 | * vfs implicitly by not allowing the user space | |
403 | * application reading/writing this device to return until | |
404 | * the buffers are done being used. | |
405 | */ | |
406 | if (op->downcall.type == ORANGEFS_VFS_OP_FILE_IO) { | |
407 | int timed_out = 0; | |
408 | DEFINE_WAIT(wait_entry); | |
97f10027 MM |
409 | |
410 | /* | |
b3ae4755 MM |
411 | * tell the vfs op waiting on a waitqueue |
412 | * that this op is done | |
97f10027 | 413 | */ |
b3ae4755 MM |
414 | spin_lock(&op->lock); |
415 | set_op_state_serviced(op); | |
416 | spin_unlock(&op->lock); | |
5db11c21 | 417 | |
b3ae4755 | 418 | while (1) { |
5db11c21 | 419 | spin_lock(&op->lock); |
b3ae4755 MM |
420 | prepare_to_wait_exclusive( |
421 | &op->io_completion_waitq, | |
422 | &wait_entry, | |
423 | TASK_INTERRUPTIBLE); | |
424 | if (op->io_completed) { | |
5db11c21 | 425 | spin_unlock(&op->lock); |
5db11c21 MM |
426 | break; |
427 | } | |
ce6c414e | 428 | spin_unlock(&op->lock); |
5db11c21 | 429 | |
b3ae4755 MM |
430 | if (!signal_pending(current)) { |
431 | int timeout = | |
432 | MSECS_TO_JIFFIES(1000 * | |
433 | op_timeout_secs); | |
434 | if (!schedule_timeout(timeout)) { | |
435 | gossip_debug(GOSSIP_DEV_DEBUG, | |
436 | "%s: timed out.\n", | |
437 | __func__); | |
438 | timed_out = 1; | |
439 | break; | |
440 | } | |
441 | continue; | |
442 | } | |
5db11c21 | 443 | |
b3ae4755 MM |
444 | gossip_debug(GOSSIP_DEV_DEBUG, |
445 | "%s: signal on I/O wait, aborting\n", | |
446 | __func__); | |
447 | break; | |
5db11c21 | 448 | } |
b3ae4755 MM |
449 | |
450 | spin_lock(&op->lock); | |
451 | finish_wait(&op->io_completion_waitq, &wait_entry); | |
452 | spin_unlock(&op->lock); | |
453 | ||
454 | /* NOTE: for I/O operations we handle releasing the op | |
455 | * object except in the case of timeout. the reason we | |
456 | * can't free the op in timeout cases is that the op | |
457 | * service logic in the vfs retries operations using | |
458 | * the same op ptr, thus it can't be freed. | |
459 | */ | |
460 | if (!timed_out) | |
461 | op_release(op); | |
5db11c21 | 462 | } else { |
b3ae4755 MM |
463 | /* |
464 | * tell the vfs op waiting on a waitqueue that | |
831d0949 | 465 | * this op is done - |
b3ae4755 MM |
466 | * for every other operation (i.e. non-I/O), we need to |
467 | * wake up the callers for downcall completion | |
468 | * notification | |
469 | */ | |
831d0949 AV |
470 | spin_lock(&op->lock); |
471 | set_op_state_serviced(op); | |
472 | spin_unlock(&op->lock); | |
5db11c21 | 473 | } |
b3ae4755 MM |
474 | out: |
475 | return ret; | |
5db11c21 MM |
476 | } |
477 | ||
478 | /* Returns whether any FS are still pending remounted */ | |
479 | static int mark_all_pending_mounts(void) | |
480 | { | |
481 | int unmounted = 1; | |
8bb8aefd | 482 | struct orangefs_sb_info_s *orangefs_sb = NULL; |
5db11c21 | 483 | |
8bb8aefd YL |
484 | spin_lock(&orangefs_superblocks_lock); |
485 | list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) { | |
5db11c21 | 486 | /* All of these file system require a remount */ |
8bb8aefd | 487 | orangefs_sb->mount_pending = 1; |
5db11c21 MM |
488 | unmounted = 0; |
489 | } | |
8bb8aefd | 490 | spin_unlock(&orangefs_superblocks_lock); |
5db11c21 MM |
491 | return unmounted; |
492 | } | |
493 | ||
494 | /* | |
495 | * Determine if a given file system needs to be remounted or not | |
496 | * Returns -1 on error | |
497 | * 0 if already mounted | |
498 | * 1 if needs remount | |
499 | */ | |
500 | int fs_mount_pending(__s32 fsid) | |
501 | { | |
502 | int mount_pending = -1; | |
8bb8aefd | 503 | struct orangefs_sb_info_s *orangefs_sb = NULL; |
5db11c21 | 504 | |
8bb8aefd YL |
505 | spin_lock(&orangefs_superblocks_lock); |
506 | list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) { | |
507 | if (orangefs_sb->fs_id == fsid) { | |
508 | mount_pending = orangefs_sb->mount_pending; | |
5db11c21 MM |
509 | break; |
510 | } | |
511 | } | |
8bb8aefd | 512 | spin_unlock(&orangefs_superblocks_lock); |
5db11c21 MM |
513 | return mount_pending; |
514 | } | |
515 | ||
516 | /* | |
517 | * NOTE: gets called when the last reference to this device is dropped. | |
518 | * Using the open_access_count variable, we enforce a reference count | |
519 | * on this file so that it can be opened by only one process at a time. | |
520 | * the devreq_mutex is used to make sure all i/o has completed | |
8bb8aefd | 521 | * before we call orangefs_bufmap_finalize, and similar such tricky |
5db11c21 MM |
522 | * situations |
523 | */ | |
8bb8aefd | 524 | static int orangefs_devreq_release(struct inode *inode, struct file *file) |
5db11c21 MM |
525 | { |
526 | int unmounted = 0; | |
527 | ||
528 | gossip_debug(GOSSIP_DEV_DEBUG, | |
529 | "%s:pvfs2-client-core: exiting, closing device\n", | |
530 | __func__); | |
531 | ||
532 | mutex_lock(&devreq_mutex); | |
7d221485 | 533 | if (orangefs_get_bufmap_init()) |
90d26aa8 | 534 | orangefs_bufmap_finalize(); |
5db11c21 | 535 | |
fee25ce1 | 536 | open_access_count = -1; |
5db11c21 MM |
537 | |
538 | unmounted = mark_all_pending_mounts(); | |
8bb8aefd | 539 | gossip_debug(GOSSIP_DEV_DEBUG, "ORANGEFS Device Close: Filesystem(s) %s\n", |
5db11c21 | 540 | (unmounted ? "UNMOUNTED" : "MOUNTED")); |
5db11c21 MM |
541 | |
542 | /* | |
543 | * Walk through the list of ops in the request list, mark them | |
544 | * as purged and wake them up. | |
545 | */ | |
546 | purge_waiting_ops(); | |
547 | /* | |
548 | * Walk through the hash table of in progress operations; mark | |
549 | * them as purged and wake them up | |
550 | */ | |
551 | purge_inprogress_ops(); | |
552 | gossip_debug(GOSSIP_DEV_DEBUG, | |
553 | "pvfs2-client-core: device close complete\n"); | |
fee25ce1 AV |
554 | open_access_count = 0; |
555 | mutex_unlock(&devreq_mutex); | |
5db11c21 MM |
556 | return 0; |
557 | } | |
558 | ||
559 | int is_daemon_in_service(void) | |
560 | { | |
561 | int in_service; | |
562 | ||
563 | /* | |
564 | * What this function does is checks if client-core is alive | |
565 | * based on the access count we maintain on the device. | |
566 | */ | |
567 | mutex_lock(&devreq_mutex); | |
568 | in_service = open_access_count == 1 ? 0 : -EIO; | |
569 | mutex_unlock(&devreq_mutex); | |
570 | return in_service; | |
571 | } | |
572 | ||
573 | static inline long check_ioctl_command(unsigned int command) | |
574 | { | |
575 | /* Check for valid ioctl codes */ | |
8bb8aefd | 576 | if (_IOC_TYPE(command) != ORANGEFS_DEV_MAGIC) { |
5db11c21 MM |
577 | gossip_err("device ioctl magic numbers don't match! Did you rebuild pvfs2-client-core/libpvfs2? [cmd %x, magic %x != %x]\n", |
578 | command, | |
579 | _IOC_TYPE(command), | |
8bb8aefd | 580 | ORANGEFS_DEV_MAGIC); |
5db11c21 MM |
581 | return -EINVAL; |
582 | } | |
583 | /* and valid ioctl commands */ | |
8bb8aefd | 584 | if (_IOC_NR(command) >= ORANGEFS_DEV_MAXNR || _IOC_NR(command) <= 0) { |
5db11c21 | 585 | gossip_err("Invalid ioctl command number [%d >= %d]\n", |
8bb8aefd | 586 | _IOC_NR(command), ORANGEFS_DEV_MAXNR); |
5db11c21 MM |
587 | return -ENOIOCTLCMD; |
588 | } | |
589 | return 0; | |
590 | } | |
591 | ||
592 | static long dispatch_ioctl_command(unsigned int command, unsigned long arg) | |
593 | { | |
8bb8aefd | 594 | static __s32 magic = ORANGEFS_DEVREQ_MAGIC; |
a762ae6d MB |
595 | static __s32 max_up_size = MAX_DEV_REQ_UPSIZE; |
596 | static __s32 max_down_size = MAX_DEV_REQ_DOWNSIZE; | |
8bb8aefd | 597 | struct ORANGEFS_dev_map_desc user_desc; |
5db11c21 MM |
598 | int ret = 0; |
599 | struct dev_mask_info_s mask_info = { 0 }; | |
600 | struct dev_mask2_info_s mask2_info = { 0, 0 }; | |
601 | int upstream_kmod = 1; | |
602 | struct list_head *tmp = NULL; | |
8bb8aefd | 603 | struct orangefs_sb_info_s *orangefs_sb = NULL; |
5db11c21 MM |
604 | |
605 | /* mtmoore: add locking here */ | |
606 | ||
607 | switch (command) { | |
8bb8aefd | 608 | case ORANGEFS_DEV_GET_MAGIC: |
5db11c21 MM |
609 | return ((put_user(magic, (__s32 __user *) arg) == -EFAULT) ? |
610 | -EIO : | |
611 | 0); | |
8bb8aefd | 612 | case ORANGEFS_DEV_GET_MAX_UPSIZE: |
5db11c21 MM |
613 | return ((put_user(max_up_size, |
614 | (__s32 __user *) arg) == -EFAULT) ? | |
615 | -EIO : | |
616 | 0); | |
8bb8aefd | 617 | case ORANGEFS_DEV_GET_MAX_DOWNSIZE: |
5db11c21 MM |
618 | return ((put_user(max_down_size, |
619 | (__s32 __user *) arg) == -EFAULT) ? | |
620 | -EIO : | |
621 | 0); | |
8bb8aefd | 622 | case ORANGEFS_DEV_MAP: |
5db11c21 | 623 | ret = copy_from_user(&user_desc, |
8bb8aefd | 624 | (struct ORANGEFS_dev_map_desc __user *) |
5db11c21 | 625 | arg, |
8bb8aefd | 626 | sizeof(struct ORANGEFS_dev_map_desc)); |
7d221485 | 627 | if (orangefs_get_bufmap_init()) { |
90d26aa8 MB |
628 | return -EINVAL; |
629 | } else { | |
630 | return ret ? | |
631 | -EIO : | |
632 | orangefs_bufmap_initialize(&user_desc); | |
633 | } | |
8bb8aefd | 634 | case ORANGEFS_DEV_REMOUNT_ALL: |
5db11c21 | 635 | gossip_debug(GOSSIP_DEV_DEBUG, |
97f10027 MM |
636 | "%s: got ORANGEFS_DEV_REMOUNT_ALL\n", |
637 | __func__); | |
5db11c21 MM |
638 | |
639 | /* | |
8bb8aefd | 640 | * remount all mounted orangefs volumes to regain the lost |
5db11c21 MM |
641 | * dynamic mount tables (if any) -- NOTE: this is done |
642 | * without keeping the superblock list locked due to the | |
643 | * upcall/downcall waiting. also, the request semaphore is | |
644 | * used to ensure that no operations will be serviced until | |
645 | * all of the remounts are serviced (to avoid ops between | |
646 | * mounts to fail) | |
647 | */ | |
648 | ret = mutex_lock_interruptible(&request_mutex); | |
649 | if (ret < 0) | |
650 | return ret; | |
651 | gossip_debug(GOSSIP_DEV_DEBUG, | |
97f10027 MM |
652 | "%s: priority remount in progress\n", |
653 | __func__); | |
8bb8aefd YL |
654 | list_for_each(tmp, &orangefs_superblocks) { |
655 | orangefs_sb = | |
97f10027 MM |
656 | list_entry(tmp, |
657 | struct orangefs_sb_info_s, | |
658 | list); | |
8bb8aefd | 659 | if (orangefs_sb && (orangefs_sb->sb)) { |
5db11c21 | 660 | gossip_debug(GOSSIP_DEV_DEBUG, |
97f10027 MM |
661 | "%s: Remounting SB %p\n", |
662 | __func__, | |
8bb8aefd | 663 | orangefs_sb); |
5db11c21 | 664 | |
8bb8aefd | 665 | ret = orangefs_remount(orangefs_sb->sb); |
5db11c21 MM |
666 | if (ret) { |
667 | gossip_debug(GOSSIP_DEV_DEBUG, | |
668 | "SB %p remount failed\n", | |
8bb8aefd | 669 | orangefs_sb); |
97f10027 | 670 | break; |
5db11c21 MM |
671 | } |
672 | } | |
673 | } | |
674 | gossip_debug(GOSSIP_DEV_DEBUG, | |
97f10027 MM |
675 | "%s: priority remount complete\n", |
676 | __func__); | |
5db11c21 MM |
677 | mutex_unlock(&request_mutex); |
678 | return ret; | |
679 | ||
8bb8aefd | 680 | case ORANGEFS_DEV_UPSTREAM: |
5db11c21 MM |
681 | ret = copy_to_user((void __user *)arg, |
682 | &upstream_kmod, | |
683 | sizeof(upstream_kmod)); | |
684 | ||
685 | if (ret != 0) | |
686 | return -EIO; | |
687 | else | |
688 | return ret; | |
689 | ||
8bb8aefd | 690 | case ORANGEFS_DEV_CLIENT_MASK: |
5db11c21 MM |
691 | ret = copy_from_user(&mask2_info, |
692 | (void __user *)arg, | |
693 | sizeof(struct dev_mask2_info_s)); | |
694 | ||
695 | if (ret != 0) | |
696 | return -EIO; | |
697 | ||
698 | client_debug_mask.mask1 = mask2_info.mask1_value; | |
699 | client_debug_mask.mask2 = mask2_info.mask2_value; | |
700 | ||
701 | pr_info("%s: client debug mask has been been received " | |
702 | ":%llx: :%llx:\n", | |
703 | __func__, | |
704 | (unsigned long long)client_debug_mask.mask1, | |
705 | (unsigned long long)client_debug_mask.mask2); | |
706 | ||
707 | return ret; | |
708 | ||
8bb8aefd | 709 | case ORANGEFS_DEV_CLIENT_STRING: |
5db11c21 MM |
710 | ret = copy_from_user(&client_debug_array_string, |
711 | (void __user *)arg, | |
8bb8aefd | 712 | ORANGEFS_MAX_DEBUG_STRING_LEN); |
5db11c21 | 713 | if (ret != 0) { |
97f10027 | 714 | pr_info("%s: CLIENT_STRING: copy_from_user failed\n", |
5db11c21 MM |
715 | __func__); |
716 | return -EIO; | |
717 | } | |
718 | ||
97f10027 | 719 | pr_info("%s: client debug array string has been received.\n", |
5db11c21 MM |
720 | __func__); |
721 | ||
722 | if (!help_string_initialized) { | |
723 | ||
724 | /* Free the "we don't know yet" default string... */ | |
725 | kfree(debug_help_string); | |
726 | ||
727 | /* build a proper debug help string */ | |
728 | if (orangefs_prepare_debugfs_help_string(0)) { | |
97f10027 | 729 | gossip_err("%s: no debug help string \n", |
5db11c21 MM |
730 | __func__); |
731 | return -EIO; | |
732 | } | |
733 | ||
734 | /* Replace the boilerplate boot-time debug-help file. */ | |
735 | debugfs_remove(help_file_dentry); | |
736 | ||
737 | help_file_dentry = | |
738 | debugfs_create_file( | |
739 | ORANGEFS_KMOD_DEBUG_HELP_FILE, | |
740 | 0444, | |
741 | debug_dir, | |
742 | debug_help_string, | |
743 | &debug_help_fops); | |
744 | ||
745 | if (!help_file_dentry) { | |
746 | gossip_err("%s: debugfs_create_file failed for" | |
747 | " :%s:!\n", | |
748 | __func__, | |
749 | ORANGEFS_KMOD_DEBUG_HELP_FILE); | |
750 | return -EIO; | |
751 | } | |
752 | } | |
753 | ||
754 | debug_mask_to_string(&client_debug_mask, 1); | |
755 | ||
756 | debugfs_remove(client_debug_dentry); | |
757 | ||
8bb8aefd | 758 | orangefs_client_debug_init(); |
5db11c21 MM |
759 | |
760 | help_string_initialized++; | |
761 | ||
762 | return ret; | |
763 | ||
8bb8aefd | 764 | case ORANGEFS_DEV_DEBUG: |
5db11c21 MM |
765 | ret = copy_from_user(&mask_info, |
766 | (void __user *)arg, | |
767 | sizeof(mask_info)); | |
768 | ||
769 | if (ret != 0) | |
770 | return -EIO; | |
771 | ||
772 | if (mask_info.mask_type == KERNEL_MASK) { | |
773 | if ((mask_info.mask_value == 0) | |
774 | && (kernel_mask_set_mod_init)) { | |
775 | /* | |
776 | * the kernel debug mask was set when the | |
777 | * kernel module was loaded; don't override | |
778 | * it if the client-core was started without | |
8bb8aefd | 779 | * a value for ORANGEFS_KMODMASK. |
5db11c21 MM |
780 | */ |
781 | return 0; | |
782 | } | |
783 | debug_mask_to_string(&mask_info.mask_value, | |
784 | mask_info.mask_type); | |
785 | gossip_debug_mask = mask_info.mask_value; | |
97f10027 | 786 | pr_info("%s: kernel debug mask has been modified to " |
5db11c21 | 787 | ":%s: :%llx:\n", |
97f10027 | 788 | __func__, |
5db11c21 MM |
789 | kernel_debug_string, |
790 | (unsigned long long)gossip_debug_mask); | |
791 | } else if (mask_info.mask_type == CLIENT_MASK) { | |
792 | debug_mask_to_string(&mask_info.mask_value, | |
793 | mask_info.mask_type); | |
97f10027 | 794 | pr_info("%s: client debug mask has been modified to" |
5db11c21 | 795 | ":%s: :%llx:\n", |
97f10027 | 796 | __func__, |
5db11c21 MM |
797 | client_debug_string, |
798 | llu(mask_info.mask_value)); | |
799 | } else { | |
800 | gossip_lerr("Invalid mask type....\n"); | |
801 | return -EINVAL; | |
802 | } | |
803 | ||
804 | return ret; | |
805 | ||
806 | default: | |
807 | return -ENOIOCTLCMD; | |
808 | } | |
809 | return -ENOIOCTLCMD; | |
810 | } | |
811 | ||
8bb8aefd | 812 | static long orangefs_devreq_ioctl(struct file *file, |
5db11c21 MM |
813 | unsigned int command, unsigned long arg) |
814 | { | |
815 | long ret; | |
816 | ||
817 | /* Check for properly constructed commands */ | |
818 | ret = check_ioctl_command(command); | |
819 | if (ret < 0) | |
820 | return (int)ret; | |
821 | ||
822 | return (int)dispatch_ioctl_command(command, arg); | |
823 | } | |
824 | ||
825 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
826 | ||
8bb8aefd YL |
827 | /* Compat structure for the ORANGEFS_DEV_MAP ioctl */ |
828 | struct ORANGEFS_dev_map_desc32 { | |
5db11c21 MM |
829 | compat_uptr_t ptr; |
830 | __s32 total_size; | |
831 | __s32 size; | |
832 | __s32 count; | |
833 | }; | |
834 | ||
835 | static unsigned long translate_dev_map26(unsigned long args, long *error) | |
836 | { | |
8bb8aefd | 837 | struct ORANGEFS_dev_map_desc32 __user *p32 = (void __user *)args; |
5db11c21 MM |
838 | /* |
839 | * Depending on the architecture, allocate some space on the | |
840 | * user-call-stack based on our expected layout. | |
841 | */ | |
8bb8aefd | 842 | struct ORANGEFS_dev_map_desc __user *p = |
5db11c21 | 843 | compat_alloc_user_space(sizeof(*p)); |
84d02150 | 844 | compat_uptr_t addr; |
5db11c21 MM |
845 | |
846 | *error = 0; | |
847 | /* get the ptr from the 32 bit user-space */ | |
848 | if (get_user(addr, &p32->ptr)) | |
849 | goto err; | |
850 | /* try to put that into a 64-bit layout */ | |
851 | if (put_user(compat_ptr(addr), &p->ptr)) | |
852 | goto err; | |
853 | /* copy the remaining fields */ | |
854 | if (copy_in_user(&p->total_size, &p32->total_size, sizeof(__s32))) | |
855 | goto err; | |
856 | if (copy_in_user(&p->size, &p32->size, sizeof(__s32))) | |
857 | goto err; | |
858 | if (copy_in_user(&p->count, &p32->count, sizeof(__s32))) | |
859 | goto err; | |
860 | return (unsigned long)p; | |
861 | err: | |
862 | *error = -EFAULT; | |
863 | return 0; | |
864 | } | |
865 | ||
866 | /* | |
867 | * 32 bit user-space apps' ioctl handlers when kernel modules | |
868 | * is compiled as a 64 bit one | |
869 | */ | |
8bb8aefd | 870 | static long orangefs_devreq_compat_ioctl(struct file *filp, unsigned int cmd, |
5db11c21 MM |
871 | unsigned long args) |
872 | { | |
873 | long ret; | |
874 | unsigned long arg = args; | |
875 | ||
876 | /* Check for properly constructed commands */ | |
877 | ret = check_ioctl_command(cmd); | |
878 | if (ret < 0) | |
879 | return ret; | |
8bb8aefd | 880 | if (cmd == ORANGEFS_DEV_MAP) { |
5db11c21 MM |
881 | /* |
882 | * convert the arguments to what we expect internally | |
883 | * in kernel space | |
884 | */ | |
885 | arg = translate_dev_map26(args, &ret); | |
886 | if (ret < 0) { | |
887 | gossip_err("Could not translate dev map\n"); | |
888 | return ret; | |
889 | } | |
890 | } | |
891 | /* no other ioctl requires translation */ | |
892 | return dispatch_ioctl_command(cmd, arg); | |
893 | } | |
894 | ||
2c590d5f MM |
895 | #endif /* CONFIG_COMPAT is in .config */ |
896 | ||
5db11c21 | 897 | /* the assigned character device major number */ |
8bb8aefd | 898 | static int orangefs_dev_major; |
5db11c21 MM |
899 | |
900 | /* | |
8bb8aefd | 901 | * Initialize orangefs device specific state: |
5db11c21 MM |
902 | * Must be called at module load time only |
903 | */ | |
8bb8aefd | 904 | int orangefs_dev_init(void) |
5db11c21 | 905 | { |
8bb8aefd YL |
906 | /* register orangefs-req device */ |
907 | orangefs_dev_major = register_chrdev(0, | |
908 | ORANGEFS_REQDEVICE_NAME, | |
909 | &orangefs_devreq_file_operations); | |
910 | if (orangefs_dev_major < 0) { | |
5db11c21 MM |
911 | gossip_debug(GOSSIP_DEV_DEBUG, |
912 | "Failed to register /dev/%s (error %d)\n", | |
8bb8aefd | 913 | ORANGEFS_REQDEVICE_NAME, orangefs_dev_major); |
8bb8aefd | 914 | return orangefs_dev_major; |
5db11c21 MM |
915 | } |
916 | ||
917 | gossip_debug(GOSSIP_DEV_DEBUG, | |
918 | "*** /dev/%s character device registered ***\n", | |
8bb8aefd | 919 | ORANGEFS_REQDEVICE_NAME); |
5db11c21 | 920 | gossip_debug(GOSSIP_DEV_DEBUG, "'mknod /dev/%s c %d 0'.\n", |
8bb8aefd | 921 | ORANGEFS_REQDEVICE_NAME, orangefs_dev_major); |
5db11c21 MM |
922 | return 0; |
923 | } | |
924 | ||
8bb8aefd | 925 | void orangefs_dev_cleanup(void) |
5db11c21 | 926 | { |
8bb8aefd | 927 | unregister_chrdev(orangefs_dev_major, ORANGEFS_REQDEVICE_NAME); |
5db11c21 MM |
928 | gossip_debug(GOSSIP_DEV_DEBUG, |
929 | "*** /dev/%s character device unregistered ***\n", | |
8bb8aefd | 930 | ORANGEFS_REQDEVICE_NAME); |
5db11c21 MM |
931 | } |
932 | ||
8bb8aefd | 933 | static unsigned int orangefs_devreq_poll(struct file *file, |
5db11c21 MM |
934 | struct poll_table_struct *poll_table) |
935 | { | |
936 | int poll_revent_mask = 0; | |
937 | ||
83595db0 | 938 | poll_wait(file, &orangefs_request_list_waitq, poll_table); |
5db11c21 | 939 | |
83595db0 AV |
940 | if (!list_empty(&orangefs_request_list)) |
941 | poll_revent_mask |= POLL_IN; | |
5db11c21 MM |
942 | return poll_revent_mask; |
943 | } | |
944 | ||
8bb8aefd | 945 | const struct file_operations orangefs_devreq_file_operations = { |
5db11c21 | 946 | .owner = THIS_MODULE, |
8bb8aefd YL |
947 | .read = orangefs_devreq_read, |
948 | .write_iter = orangefs_devreq_write_iter, | |
949 | .open = orangefs_devreq_open, | |
950 | .release = orangefs_devreq_release, | |
951 | .unlocked_ioctl = orangefs_devreq_ioctl, | |
5db11c21 MM |
952 | |
953 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
8bb8aefd | 954 | .compat_ioctl = orangefs_devreq_compat_ioctl, |
5db11c21 | 955 | #endif |
8bb8aefd | 956 | .poll = orangefs_devreq_poll |
5db11c21 | 957 | }; |