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 | ||
ca9f518e MM |
49 | /* |
50 | * find the op with this tag and remove it from the in progress | |
51 | * hash table. | |
52 | */ | |
8bb8aefd | 53 | static struct orangefs_kernel_op_s *orangefs_devreq_remove_op(__u64 tag) |
5db11c21 | 54 | { |
8bb8aefd | 55 | struct orangefs_kernel_op_s *op, *next; |
5db11c21 MM |
56 | int index; |
57 | ||
58 | index = hash_func(tag, hash_table_size); | |
59 | ||
60 | spin_lock(&htable_ops_in_progress_lock); | |
61 | list_for_each_entry_safe(op, | |
62 | next, | |
63 | &htable_ops_in_progress[index], | |
64 | list) { | |
05a50a5b AV |
65 | if (op->tag == tag && !op_state_purged(op) && |
66 | !op_state_given_up(op)) { | |
ed42fe05 | 67 | list_del_init(&op->list); |
5db11c21 MM |
68 | spin_unlock(&htable_ops_in_progress_lock); |
69 | return op; | |
70 | } | |
71 | } | |
72 | ||
73 | spin_unlock(&htable_ops_in_progress_lock); | |
74 | return NULL; | |
75 | } | |
76 | ||
8bb8aefd | 77 | static int orangefs_devreq_open(struct inode *inode, struct file *file) |
5db11c21 MM |
78 | { |
79 | int ret = -EINVAL; | |
80 | ||
81 | if (!(file->f_flags & O_NONBLOCK)) { | |
97f10027 MM |
82 | gossip_err("%s: device cannot be opened in blocking mode\n", |
83 | __func__); | |
5db11c21 MM |
84 | goto out; |
85 | } | |
86 | ret = -EACCES; | |
97f10027 | 87 | gossip_debug(GOSSIP_DEV_DEBUG, "client-core: opening device\n"); |
5db11c21 MM |
88 | mutex_lock(&devreq_mutex); |
89 | ||
90 | if (open_access_count == 0) { | |
fee25ce1 | 91 | open_access_count = 1; |
fb6d2526 | 92 | ret = 0; |
5db11c21 MM |
93 | } else { |
94 | DUMP_DEVICE_ERROR(); | |
95 | } | |
96 | mutex_unlock(&devreq_mutex); | |
97 | ||
98 | out: | |
99 | ||
100 | gossip_debug(GOSSIP_DEV_DEBUG, | |
101 | "pvfs2-client-core: open device complete (ret = %d)\n", | |
102 | ret); | |
103 | return ret; | |
104 | } | |
105 | ||
97f10027 | 106 | /* Function for read() callers into the device */ |
8bb8aefd | 107 | static ssize_t orangefs_devreq_read(struct file *file, |
5db11c21 MM |
108 | char __user *buf, |
109 | size_t count, loff_t *offset) | |
110 | { | |
8bb8aefd YL |
111 | struct orangefs_kernel_op_s *op, *temp; |
112 | __s32 proto_ver = ORANGEFS_KERNEL_PROTO_VERSION; | |
113 | static __s32 magic = ORANGEFS_DEVREQ_MAGIC; | |
114 | struct orangefs_kernel_op_s *cur_op = NULL; | |
24c8d080 | 115 | unsigned long ret; |
5db11c21 | 116 | |
24c8d080 | 117 | /* We do not support blocking IO. */ |
5db11c21 | 118 | if (!(file->f_flags & O_NONBLOCK)) { |
97f10027 MM |
119 | gossip_err("%s: blocking read from client-core.\n", |
120 | __func__); | |
5db11c21 | 121 | return -EINVAL; |
24c8d080 MB |
122 | } |
123 | ||
124 | /* | |
a762ae6d | 125 | * The client will do an ioctl to find MAX_DEV_REQ_UPSIZE, then |
24c8d080 MB |
126 | * always read with that size buffer. |
127 | */ | |
a762ae6d | 128 | if (count != MAX_DEV_REQ_UPSIZE) { |
24c8d080 MB |
129 | gossip_err("orangefs: client-core tried to read wrong size\n"); |
130 | return -EINVAL; | |
131 | } | |
132 | ||
ed42fe05 | 133 | restart: |
24c8d080 | 134 | /* Get next op (if any) from top of list. */ |
8bb8aefd YL |
135 | spin_lock(&orangefs_request_list_lock); |
136 | list_for_each_entry_safe(op, temp, &orangefs_request_list, list) { | |
24c8d080 MB |
137 | __s32 fsid; |
138 | /* This lock is held past the end of the loop when we break. */ | |
139 | spin_lock(&op->lock); | |
05a50a5b | 140 | if (unlikely(op_state_purged(op) || op_state_given_up(op))) { |
ed42fe05 AV |
141 | spin_unlock(&op->lock); |
142 | continue; | |
143 | } | |
24c8d080 MB |
144 | |
145 | fsid = fsid_of_op(op); | |
8bb8aefd | 146 | if (fsid != ORANGEFS_FS_ID_NULL) { |
24c8d080 MB |
147 | int ret; |
148 | /* Skip ops whose filesystem needs to be mounted. */ | |
149 | ret = fs_mount_pending(fsid); | |
150 | if (ret == 1) { | |
5db11c21 | 151 | gossip_debug(GOSSIP_DEV_DEBUG, |
5090c967 MM |
152 | "%s: mount pending, skipping op tag " |
153 | "%llu %s\n", | |
154 | __func__, | |
155 | llu(op->tag), | |
156 | get_opname_string(op)); | |
24c8d080 MB |
157 | spin_unlock(&op->lock); |
158 | continue; | |
97f10027 MM |
159 | /* |
160 | * Skip ops whose filesystem we don't know about unless | |
161 | * it is being mounted. | |
162 | */ | |
24c8d080 MB |
163 | /* XXX: is there a better way to detect this? */ |
164 | } else if (ret == -1 && | |
97f10027 MM |
165 | !(op->upcall.type == |
166 | ORANGEFS_VFS_OP_FS_MOUNT || | |
167 | op->upcall.type == | |
168 | ORANGEFS_VFS_OP_GETATTR)) { | |
24c8d080 MB |
169 | gossip_debug(GOSSIP_DEV_DEBUG, |
170 | "orangefs: skipping op tag %llu %s\n", | |
171 | llu(op->tag), get_opname_string(op)); | |
172 | gossip_err( | |
173 | "orangefs: ERROR: fs_mount_pending %d\n", | |
174 | fsid); | |
175 | spin_unlock(&op->lock); | |
5db11c21 | 176 | continue; |
5db11c21 MM |
177 | } |
178 | } | |
24c8d080 MB |
179 | /* |
180 | * Either this op does not pertain to a filesystem, is mounting | |
181 | * a filesystem, or pertains to a mounted filesystem. Let it | |
182 | * through. | |
183 | */ | |
184 | cur_op = op; | |
185 | break; | |
186 | } | |
187 | ||
188 | /* | |
189 | * At this point we either have a valid op and can continue or have not | |
190 | * found an op and must ask the client to try again later. | |
191 | */ | |
192 | if (!cur_op) { | |
8bb8aefd | 193 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 194 | return -EAGAIN; |
5db11c21 MM |
195 | } |
196 | ||
ca9f518e MM |
197 | gossip_debug(GOSSIP_DEV_DEBUG, "%s: reading op tag %llu %s\n", |
198 | __func__, | |
199 | llu(cur_op->tag), | |
200 | get_opname_string(cur_op)); | |
5db11c21 | 201 | |
24c8d080 MB |
202 | /* |
203 | * Such an op should never be on the list in the first place. If so, we | |
204 | * will abort. | |
205 | */ | |
206 | if (op_state_in_progress(cur_op) || op_state_serviced(cur_op)) { | |
207 | gossip_err("orangefs: ERROR: Current op already queued.\n"); | |
05a50a5b | 208 | list_del_init(&cur_op->list); |
5db11c21 | 209 | spin_unlock(&cur_op->lock); |
8bb8aefd | 210 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 211 | return -EAGAIN; |
5db11c21 | 212 | } |
ca9f518e | 213 | |
ed42fe05 | 214 | list_del_init(&cur_op->list); |
8bb8aefd | 215 | spin_unlock(&orangefs_request_list_lock); |
ed42fe05 | 216 | |
24c8d080 MB |
217 | spin_unlock(&cur_op->lock); |
218 | ||
219 | /* Push the upcall out. */ | |
220 | ret = copy_to_user(buf, &proto_ver, sizeof(__s32)); | |
221 | if (ret != 0) | |
222 | goto error; | |
223 | ret = copy_to_user(buf+sizeof(__s32), &magic, sizeof(__s32)); | |
224 | if (ret != 0) | |
225 | goto error; | |
226 | ret = copy_to_user(buf+2 * sizeof(__s32), &cur_op->tag, sizeof(__u64)); | |
227 | if (ret != 0) | |
228 | goto error; | |
229 | ret = copy_to_user(buf+2*sizeof(__s32)+sizeof(__u64), &cur_op->upcall, | |
8bb8aefd | 230 | sizeof(struct orangefs_upcall_s)); |
24c8d080 MB |
231 | if (ret != 0) |
232 | goto error; | |
233 | ||
ed42fe05 AV |
234 | spin_lock(&htable_ops_in_progress_lock); |
235 | spin_lock(&cur_op->lock); | |
236 | if (unlikely(op_state_given_up(cur_op))) { | |
237 | spin_unlock(&cur_op->lock); | |
238 | spin_unlock(&htable_ops_in_progress_lock); | |
05a50a5b | 239 | complete(&cur_op->waitq); |
ed42fe05 AV |
240 | goto restart; |
241 | } | |
242 | ||
243 | /* | |
244 | * Set the operation to be in progress and move it between lists since | |
245 | * it has been sent to the client. | |
246 | */ | |
247 | set_op_state_inprogress(cur_op); | |
9d9e7ba9 MM |
248 | gossip_debug(GOSSIP_DEV_DEBUG, |
249 | "%s: 1 op:%s: op_state:%d: process:%s:\n", | |
250 | __func__, | |
251 | get_opname_string(cur_op), | |
252 | cur_op->op_state, | |
253 | current->comm); | |
ed42fe05 AV |
254 | orangefs_devreq_add_op(cur_op); |
255 | spin_unlock(&cur_op->lock); | |
256 | spin_unlock(&htable_ops_in_progress_lock); | |
ed42fe05 | 257 | |
24c8d080 | 258 | /* The client only asks to read one size buffer. */ |
a762ae6d | 259 | return MAX_DEV_REQ_UPSIZE; |
24c8d080 MB |
260 | error: |
261 | /* | |
262 | * We were unable to copy the op data to the client. Put the op back in | |
263 | * list. If client has crashed, the op will be purged later when the | |
264 | * device is released. | |
265 | */ | |
266 | gossip_err("orangefs: Failed to copy data to user space\n"); | |
8bb8aefd | 267 | spin_lock(&orangefs_request_list_lock); |
24c8d080 | 268 | spin_lock(&cur_op->lock); |
ed42fe05 AV |
269 | if (likely(!op_state_given_up(cur_op))) { |
270 | set_op_state_waiting(cur_op); | |
9d9e7ba9 MM |
271 | gossip_debug(GOSSIP_DEV_DEBUG, |
272 | "%s: 2 op:%s: op_state:%d: process:%s:\n", | |
273 | __func__, | |
274 | get_opname_string(cur_op), | |
275 | cur_op->op_state, | |
276 | current->comm); | |
ed42fe05 | 277 | list_add(&cur_op->list, &orangefs_request_list); |
05a50a5b AV |
278 | spin_unlock(&cur_op->lock); |
279 | } else { | |
280 | spin_unlock(&cur_op->lock); | |
281 | complete(&cur_op->waitq); | |
ed42fe05 | 282 | } |
8bb8aefd | 283 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 284 | return -EFAULT; |
5db11c21 MM |
285 | } |
286 | ||
97f10027 | 287 | /* |
b3ae4755 MM |
288 | * Function for writev() callers into the device. |
289 | * | |
290 | * Userspace should have written: | |
291 | * - __u32 version | |
292 | * - __u32 magic | |
293 | * - __u64 tag | |
294 | * - struct orangefs_downcall_s | |
295 | * - trailer buffer (in the case of READDIR operations) | |
97f10027 | 296 | */ |
b3ae4755 MM |
297 | static ssize_t orangefs_devreq_write_iter(struct kiocb *iocb, |
298 | struct iov_iter *iter) | |
5db11c21 | 299 | { |
b3ae4755 | 300 | ssize_t ret; |
8bb8aefd | 301 | struct orangefs_kernel_op_s *op = NULL; |
b3ae4755 MM |
302 | struct { |
303 | __u32 version; | |
304 | __u32 magic; | |
305 | __u64 tag; | |
306 | } head; | |
307 | int total = ret = iov_iter_count(iter); | |
308 | int n; | |
309 | int downcall_size = sizeof(struct orangefs_downcall_s); | |
310 | int head_size = sizeof(head); | |
311 | ||
312 | gossip_debug(GOSSIP_DEV_DEBUG, "%s: total:%d: ret:%zd:\n", | |
313 | __func__, | |
314 | total, | |
315 | ret); | |
5db11c21 | 316 | |
b3ae4755 | 317 | if (total < MAX_DEV_REQ_DOWNSIZE) { |
cf0c2771 | 318 | gossip_err("%s: total:%d: must be at least:%u:\n", |
b3ae4755 MM |
319 | __func__, |
320 | total, | |
cf0c2771 | 321 | (unsigned int) MAX_DEV_REQ_DOWNSIZE); |
ed42fe05 | 322 | return -EFAULT; |
5db11c21 | 323 | } |
b3ae4755 MM |
324 | |
325 | n = copy_from_iter(&head, head_size, iter); | |
326 | if (n < head_size) { | |
327 | gossip_err("%s: failed to copy head.\n", __func__); | |
ed42fe05 | 328 | return -EFAULT; |
97f10027 | 329 | } |
b3ae4755 MM |
330 | |
331 | if (head.version < ORANGEFS_MINIMUM_USERSPACE_VERSION) { | |
332 | gossip_err("%s: userspace claims version" | |
333 | "%d, minimum version required: %d.\n", | |
334 | __func__, | |
335 | head.version, | |
336 | ORANGEFS_MINIMUM_USERSPACE_VERSION); | |
ed42fe05 | 337 | return -EPROTO; |
5db11c21 | 338 | } |
5db11c21 | 339 | |
b3ae4755 MM |
340 | if (head.magic != ORANGEFS_DEVREQ_MAGIC) { |
341 | gossip_err("Error: Device magic number does not match.\n"); | |
ed42fe05 | 342 | return -EPROTO; |
b3ae4755 | 343 | } |
5db11c21 | 344 | |
ca9f518e | 345 | /* remove the op from the in progress hash table */ |
b3ae4755 MM |
346 | op = orangefs_devreq_remove_op(head.tag); |
347 | if (!op) { | |
348 | gossip_err("WARNING: No one's waiting for tag %llu\n", | |
349 | llu(head.tag)); | |
ed42fe05 | 350 | return ret; |
b3ae4755 | 351 | } |
5db11c21 | 352 | |
b3ae4755 MM |
353 | n = copy_from_iter(&op->downcall, downcall_size, iter); |
354 | if (n != downcall_size) { | |
355 | gossip_err("%s: failed to copy downcall.\n", __func__); | |
5964c1b8 | 356 | goto Efault; |
5db11c21 MM |
357 | } |
358 | ||
b3ae4755 MM |
359 | if (op->downcall.status) |
360 | goto wakeup; | |
97f10027 | 361 | |
b3ae4755 MM |
362 | /* |
363 | * We've successfully peeled off the head and the downcall. | |
364 | * Something has gone awry if total doesn't equal the | |
365 | * sum of head_size, downcall_size and trailer_size. | |
366 | */ | |
367 | if ((head_size + downcall_size + op->downcall.trailer_size) != total) { | |
368 | gossip_err("%s: funky write, head_size:%d" | |
369 | ": downcall_size:%d: trailer_size:%lld" | |
370 | ": total size:%d:\n", | |
371 | __func__, | |
372 | head_size, | |
373 | downcall_size, | |
374 | op->downcall.trailer_size, | |
375 | total); | |
5964c1b8 | 376 | goto Efault; |
b3ae4755 | 377 | } |
97f10027 | 378 | |
b3ae4755 MM |
379 | /* Only READDIR operations should have trailers. */ |
380 | if ((op->downcall.type != ORANGEFS_VFS_OP_READDIR) && | |
381 | (op->downcall.trailer_size != 0)) { | |
382 | gossip_err("%s: %x operation with trailer.", | |
383 | __func__, | |
384 | op->downcall.type); | |
5964c1b8 | 385 | goto Efault; |
b3ae4755 | 386 | } |
97f10027 | 387 | |
b3ae4755 MM |
388 | /* READDIR operations should always have trailers. */ |
389 | if ((op->downcall.type == ORANGEFS_VFS_OP_READDIR) && | |
390 | (op->downcall.trailer_size == 0)) { | |
391 | gossip_err("%s: %x operation with no trailer.", | |
392 | __func__, | |
393 | op->downcall.type); | |
5964c1b8 | 394 | goto Efault; |
b3ae4755 | 395 | } |
97f10027 | 396 | |
b3ae4755 MM |
397 | if (op->downcall.type != ORANGEFS_VFS_OP_READDIR) |
398 | goto wakeup; | |
5db11c21 | 399 | |
b3ae4755 MM |
400 | op->downcall.trailer_buf = |
401 | vmalloc(op->downcall.trailer_size); | |
402 | if (op->downcall.trailer_buf == NULL) { | |
403 | gossip_err("%s: failed trailer vmalloc.\n", | |
404 | __func__); | |
5964c1b8 | 405 | goto Enomem; |
b3ae4755 MM |
406 | } |
407 | memset(op->downcall.trailer_buf, 0, op->downcall.trailer_size); | |
408 | n = copy_from_iter(op->downcall.trailer_buf, | |
409 | op->downcall.trailer_size, | |
410 | iter); | |
411 | if (n != op->downcall.trailer_size) { | |
412 | gossip_err("%s: failed to copy trailer.\n", __func__); | |
413 | vfree(op->downcall.trailer_buf); | |
5964c1b8 | 414 | goto Efault; |
b3ae4755 | 415 | } |
97f10027 | 416 | |
b3ae4755 | 417 | wakeup: |
2a9e5c22 | 418 | /* |
9f08cfe9 MM |
419 | * Return to vfs waitqueue, and back to service_operation |
420 | * through wait_for_matching_downcall. | |
2a9e5c22 AV |
421 | */ |
422 | spin_lock(&op->lock); | |
5964c1b8 | 423 | if (unlikely(op_is_cancel(op))) { |
2a9e5c22 | 424 | spin_unlock(&op->lock); |
78699e29 | 425 | put_cancel(op); |
5964c1b8 AV |
426 | } else if (unlikely(op_state_given_up(op))) { |
427 | spin_unlock(&op->lock); | |
05a50a5b | 428 | complete(&op->waitq); |
5964c1b8 AV |
429 | } else { |
430 | set_op_state_serviced(op); | |
9d9e7ba9 MM |
431 | gossip_debug(GOSSIP_DEV_DEBUG, |
432 | "%s: op:%s: op_state:%d: process:%s:\n", | |
433 | __func__, | |
434 | get_opname_string(op), | |
435 | op->op_state, | |
436 | current->comm); | |
5964c1b8 AV |
437 | spin_unlock(&op->lock); |
438 | } | |
b3ae4755 | 439 | return ret; |
ed42fe05 | 440 | |
5964c1b8 AV |
441 | Efault: |
442 | op->downcall.status = -(ORANGEFS_ERROR_BIT | 9); | |
443 | ret = -EFAULT; | |
444 | goto wakeup; | |
445 | ||
446 | Enomem: | |
447 | op->downcall.status = -(ORANGEFS_ERROR_BIT | 8); | |
448 | ret = -ENOMEM; | |
449 | goto wakeup; | |
5db11c21 MM |
450 | } |
451 | ||
452 | /* Returns whether any FS are still pending remounted */ | |
453 | static int mark_all_pending_mounts(void) | |
454 | { | |
455 | int unmounted = 1; | |
8bb8aefd | 456 | struct orangefs_sb_info_s *orangefs_sb = NULL; |
5db11c21 | 457 | |
8bb8aefd YL |
458 | spin_lock(&orangefs_superblocks_lock); |
459 | list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) { | |
5db11c21 | 460 | /* All of these file system require a remount */ |
8bb8aefd | 461 | orangefs_sb->mount_pending = 1; |
5db11c21 MM |
462 | unmounted = 0; |
463 | } | |
8bb8aefd | 464 | spin_unlock(&orangefs_superblocks_lock); |
5db11c21 MM |
465 | return unmounted; |
466 | } | |
467 | ||
468 | /* | |
469 | * Determine if a given file system needs to be remounted or not | |
470 | * Returns -1 on error | |
471 | * 0 if already mounted | |
472 | * 1 if needs remount | |
473 | */ | |
474 | int fs_mount_pending(__s32 fsid) | |
475 | { | |
476 | int mount_pending = -1; | |
8bb8aefd | 477 | struct orangefs_sb_info_s *orangefs_sb = NULL; |
5db11c21 | 478 | |
8bb8aefd YL |
479 | spin_lock(&orangefs_superblocks_lock); |
480 | list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) { | |
481 | if (orangefs_sb->fs_id == fsid) { | |
482 | mount_pending = orangefs_sb->mount_pending; | |
5db11c21 MM |
483 | break; |
484 | } | |
485 | } | |
8bb8aefd | 486 | spin_unlock(&orangefs_superblocks_lock); |
5db11c21 MM |
487 | return mount_pending; |
488 | } | |
489 | ||
490 | /* | |
491 | * NOTE: gets called when the last reference to this device is dropped. | |
492 | * Using the open_access_count variable, we enforce a reference count | |
493 | * on this file so that it can be opened by only one process at a time. | |
494 | * the devreq_mutex is used to make sure all i/o has completed | |
8bb8aefd | 495 | * before we call orangefs_bufmap_finalize, and similar such tricky |
5db11c21 MM |
496 | * situations |
497 | */ | |
8bb8aefd | 498 | static int orangefs_devreq_release(struct inode *inode, struct file *file) |
5db11c21 MM |
499 | { |
500 | int unmounted = 0; | |
501 | ||
502 | gossip_debug(GOSSIP_DEV_DEBUG, | |
503 | "%s:pvfs2-client-core: exiting, closing device\n", | |
504 | __func__); | |
505 | ||
506 | mutex_lock(&devreq_mutex); | |
ea2c9c9f | 507 | orangefs_bufmap_finalize(); |
5db11c21 | 508 | |
fee25ce1 | 509 | open_access_count = -1; |
5db11c21 MM |
510 | |
511 | unmounted = mark_all_pending_mounts(); | |
8bb8aefd | 512 | gossip_debug(GOSSIP_DEV_DEBUG, "ORANGEFS Device Close: Filesystem(s) %s\n", |
5db11c21 | 513 | (unmounted ? "UNMOUNTED" : "MOUNTED")); |
5db11c21 | 514 | |
5db11c21 | 515 | purge_waiting_ops(); |
5db11c21 | 516 | purge_inprogress_ops(); |
ea2c9c9f AV |
517 | |
518 | orangefs_bufmap_run_down(); | |
519 | ||
5db11c21 MM |
520 | gossip_debug(GOSSIP_DEV_DEBUG, |
521 | "pvfs2-client-core: device close complete\n"); | |
fee25ce1 AV |
522 | open_access_count = 0; |
523 | mutex_unlock(&devreq_mutex); | |
5db11c21 MM |
524 | return 0; |
525 | } | |
526 | ||
527 | int is_daemon_in_service(void) | |
528 | { | |
529 | int in_service; | |
530 | ||
531 | /* | |
532 | * What this function does is checks if client-core is alive | |
533 | * based on the access count we maintain on the device. | |
534 | */ | |
535 | mutex_lock(&devreq_mutex); | |
536 | in_service = open_access_count == 1 ? 0 : -EIO; | |
537 | mutex_unlock(&devreq_mutex); | |
538 | return in_service; | |
539 | } | |
540 | ||
78699e29 AV |
541 | bool __is_daemon_in_service(void) |
542 | { | |
543 | return open_access_count == 1; | |
544 | } | |
545 | ||
5db11c21 MM |
546 | static inline long check_ioctl_command(unsigned int command) |
547 | { | |
548 | /* Check for valid ioctl codes */ | |
8bb8aefd | 549 | if (_IOC_TYPE(command) != ORANGEFS_DEV_MAGIC) { |
5db11c21 MM |
550 | gossip_err("device ioctl magic numbers don't match! Did you rebuild pvfs2-client-core/libpvfs2? [cmd %x, magic %x != %x]\n", |
551 | command, | |
552 | _IOC_TYPE(command), | |
8bb8aefd | 553 | ORANGEFS_DEV_MAGIC); |
5db11c21 MM |
554 | return -EINVAL; |
555 | } | |
556 | /* and valid ioctl commands */ | |
8bb8aefd | 557 | if (_IOC_NR(command) >= ORANGEFS_DEV_MAXNR || _IOC_NR(command) <= 0) { |
5db11c21 | 558 | gossip_err("Invalid ioctl command number [%d >= %d]\n", |
8bb8aefd | 559 | _IOC_NR(command), ORANGEFS_DEV_MAXNR); |
5db11c21 MM |
560 | return -ENOIOCTLCMD; |
561 | } | |
562 | return 0; | |
563 | } | |
564 | ||
565 | static long dispatch_ioctl_command(unsigned int command, unsigned long arg) | |
566 | { | |
8bb8aefd | 567 | static __s32 magic = ORANGEFS_DEVREQ_MAGIC; |
a762ae6d MB |
568 | static __s32 max_up_size = MAX_DEV_REQ_UPSIZE; |
569 | static __s32 max_down_size = MAX_DEV_REQ_DOWNSIZE; | |
8bb8aefd | 570 | struct ORANGEFS_dev_map_desc user_desc; |
5db11c21 MM |
571 | int ret = 0; |
572 | struct dev_mask_info_s mask_info = { 0 }; | |
573 | struct dev_mask2_info_s mask2_info = { 0, 0 }; | |
574 | int upstream_kmod = 1; | |
575 | struct list_head *tmp = NULL; | |
8bb8aefd | 576 | struct orangefs_sb_info_s *orangefs_sb = NULL; |
5db11c21 MM |
577 | |
578 | /* mtmoore: add locking here */ | |
579 | ||
580 | switch (command) { | |
8bb8aefd | 581 | case ORANGEFS_DEV_GET_MAGIC: |
5db11c21 MM |
582 | return ((put_user(magic, (__s32 __user *) arg) == -EFAULT) ? |
583 | -EIO : | |
584 | 0); | |
8bb8aefd | 585 | case ORANGEFS_DEV_GET_MAX_UPSIZE: |
5db11c21 MM |
586 | return ((put_user(max_up_size, |
587 | (__s32 __user *) arg) == -EFAULT) ? | |
588 | -EIO : | |
589 | 0); | |
8bb8aefd | 590 | case ORANGEFS_DEV_GET_MAX_DOWNSIZE: |
5db11c21 MM |
591 | return ((put_user(max_down_size, |
592 | (__s32 __user *) arg) == -EFAULT) ? | |
593 | -EIO : | |
594 | 0); | |
8bb8aefd | 595 | case ORANGEFS_DEV_MAP: |
5db11c21 | 596 | ret = copy_from_user(&user_desc, |
8bb8aefd | 597 | (struct ORANGEFS_dev_map_desc __user *) |
5db11c21 | 598 | arg, |
8bb8aefd | 599 | sizeof(struct ORANGEFS_dev_map_desc)); |
ea2c9c9f AV |
600 | /* WTF -EIO and not -EFAULT? */ |
601 | return ret ? -EIO : orangefs_bufmap_initialize(&user_desc); | |
8bb8aefd | 602 | case ORANGEFS_DEV_REMOUNT_ALL: |
5db11c21 | 603 | gossip_debug(GOSSIP_DEV_DEBUG, |
97f10027 MM |
604 | "%s: got ORANGEFS_DEV_REMOUNT_ALL\n", |
605 | __func__); | |
5db11c21 MM |
606 | |
607 | /* | |
8bb8aefd | 608 | * remount all mounted orangefs volumes to regain the lost |
5db11c21 MM |
609 | * dynamic mount tables (if any) -- NOTE: this is done |
610 | * without keeping the superblock list locked due to the | |
adcf34a2 | 611 | * upcall/downcall waiting. also, the request mutex is |
5db11c21 MM |
612 | * used to ensure that no operations will be serviced until |
613 | * all of the remounts are serviced (to avoid ops between | |
614 | * mounts to fail) | |
615 | */ | |
616 | ret = mutex_lock_interruptible(&request_mutex); | |
617 | if (ret < 0) | |
618 | return ret; | |
619 | gossip_debug(GOSSIP_DEV_DEBUG, | |
97f10027 MM |
620 | "%s: priority remount in progress\n", |
621 | __func__); | |
8bb8aefd YL |
622 | list_for_each(tmp, &orangefs_superblocks) { |
623 | orangefs_sb = | |
97f10027 MM |
624 | list_entry(tmp, |
625 | struct orangefs_sb_info_s, | |
626 | list); | |
8bb8aefd | 627 | if (orangefs_sb && (orangefs_sb->sb)) { |
5db11c21 | 628 | gossip_debug(GOSSIP_DEV_DEBUG, |
97f10027 MM |
629 | "%s: Remounting SB %p\n", |
630 | __func__, | |
8bb8aefd | 631 | orangefs_sb); |
5db11c21 | 632 | |
8bb8aefd | 633 | ret = orangefs_remount(orangefs_sb->sb); |
5db11c21 MM |
634 | if (ret) { |
635 | gossip_debug(GOSSIP_DEV_DEBUG, | |
636 | "SB %p remount failed\n", | |
8bb8aefd | 637 | orangefs_sb); |
97f10027 | 638 | break; |
5db11c21 MM |
639 | } |
640 | } | |
641 | } | |
642 | gossip_debug(GOSSIP_DEV_DEBUG, | |
97f10027 MM |
643 | "%s: priority remount complete\n", |
644 | __func__); | |
5db11c21 MM |
645 | mutex_unlock(&request_mutex); |
646 | return ret; | |
647 | ||
8bb8aefd | 648 | case ORANGEFS_DEV_UPSTREAM: |
5db11c21 MM |
649 | ret = copy_to_user((void __user *)arg, |
650 | &upstream_kmod, | |
651 | sizeof(upstream_kmod)); | |
652 | ||
653 | if (ret != 0) | |
654 | return -EIO; | |
655 | else | |
656 | return ret; | |
657 | ||
8bb8aefd | 658 | case ORANGEFS_DEV_CLIENT_MASK: |
5db11c21 MM |
659 | ret = copy_from_user(&mask2_info, |
660 | (void __user *)arg, | |
661 | sizeof(struct dev_mask2_info_s)); | |
662 | ||
663 | if (ret != 0) | |
664 | return -EIO; | |
665 | ||
666 | client_debug_mask.mask1 = mask2_info.mask1_value; | |
667 | client_debug_mask.mask2 = mask2_info.mask2_value; | |
668 | ||
669 | pr_info("%s: client debug mask has been been received " | |
670 | ":%llx: :%llx:\n", | |
671 | __func__, | |
672 | (unsigned long long)client_debug_mask.mask1, | |
673 | (unsigned long long)client_debug_mask.mask2); | |
674 | ||
675 | return ret; | |
676 | ||
8bb8aefd | 677 | case ORANGEFS_DEV_CLIENT_STRING: |
5db11c21 MM |
678 | ret = copy_from_user(&client_debug_array_string, |
679 | (void __user *)arg, | |
8bb8aefd | 680 | ORANGEFS_MAX_DEBUG_STRING_LEN); |
5db11c21 | 681 | if (ret != 0) { |
97f10027 | 682 | pr_info("%s: CLIENT_STRING: copy_from_user failed\n", |
5db11c21 MM |
683 | __func__); |
684 | return -EIO; | |
685 | } | |
686 | ||
97f10027 | 687 | pr_info("%s: client debug array string has been received.\n", |
5db11c21 MM |
688 | __func__); |
689 | ||
690 | if (!help_string_initialized) { | |
691 | ||
692 | /* Free the "we don't know yet" default string... */ | |
693 | kfree(debug_help_string); | |
694 | ||
695 | /* build a proper debug help string */ | |
696 | if (orangefs_prepare_debugfs_help_string(0)) { | |
97f10027 | 697 | gossip_err("%s: no debug help string \n", |
5db11c21 MM |
698 | __func__); |
699 | return -EIO; | |
700 | } | |
701 | ||
702 | /* Replace the boilerplate boot-time debug-help file. */ | |
703 | debugfs_remove(help_file_dentry); | |
704 | ||
705 | help_file_dentry = | |
706 | debugfs_create_file( | |
707 | ORANGEFS_KMOD_DEBUG_HELP_FILE, | |
708 | 0444, | |
709 | debug_dir, | |
710 | debug_help_string, | |
711 | &debug_help_fops); | |
712 | ||
713 | if (!help_file_dentry) { | |
714 | gossip_err("%s: debugfs_create_file failed for" | |
715 | " :%s:!\n", | |
716 | __func__, | |
717 | ORANGEFS_KMOD_DEBUG_HELP_FILE); | |
718 | return -EIO; | |
719 | } | |
720 | } | |
721 | ||
722 | debug_mask_to_string(&client_debug_mask, 1); | |
723 | ||
724 | debugfs_remove(client_debug_dentry); | |
725 | ||
8bb8aefd | 726 | orangefs_client_debug_init(); |
5db11c21 MM |
727 | |
728 | help_string_initialized++; | |
729 | ||
730 | return ret; | |
731 | ||
8bb8aefd | 732 | case ORANGEFS_DEV_DEBUG: |
5db11c21 MM |
733 | ret = copy_from_user(&mask_info, |
734 | (void __user *)arg, | |
735 | sizeof(mask_info)); | |
736 | ||
737 | if (ret != 0) | |
738 | return -EIO; | |
739 | ||
740 | if (mask_info.mask_type == KERNEL_MASK) { | |
741 | if ((mask_info.mask_value == 0) | |
742 | && (kernel_mask_set_mod_init)) { | |
743 | /* | |
744 | * the kernel debug mask was set when the | |
745 | * kernel module was loaded; don't override | |
746 | * it if the client-core was started without | |
8bb8aefd | 747 | * a value for ORANGEFS_KMODMASK. |
5db11c21 MM |
748 | */ |
749 | return 0; | |
750 | } | |
751 | debug_mask_to_string(&mask_info.mask_value, | |
752 | mask_info.mask_type); | |
753 | gossip_debug_mask = mask_info.mask_value; | |
97f10027 | 754 | pr_info("%s: kernel debug mask has been modified to " |
5db11c21 | 755 | ":%s: :%llx:\n", |
97f10027 | 756 | __func__, |
5db11c21 MM |
757 | kernel_debug_string, |
758 | (unsigned long long)gossip_debug_mask); | |
759 | } else if (mask_info.mask_type == CLIENT_MASK) { | |
760 | debug_mask_to_string(&mask_info.mask_value, | |
761 | mask_info.mask_type); | |
97f10027 | 762 | pr_info("%s: client debug mask has been modified to" |
5db11c21 | 763 | ":%s: :%llx:\n", |
97f10027 | 764 | __func__, |
5db11c21 MM |
765 | client_debug_string, |
766 | llu(mask_info.mask_value)); | |
767 | } else { | |
768 | gossip_lerr("Invalid mask type....\n"); | |
769 | return -EINVAL; | |
770 | } | |
771 | ||
772 | return ret; | |
773 | ||
774 | default: | |
775 | return -ENOIOCTLCMD; | |
776 | } | |
777 | return -ENOIOCTLCMD; | |
778 | } | |
779 | ||
8bb8aefd | 780 | static long orangefs_devreq_ioctl(struct file *file, |
5db11c21 MM |
781 | unsigned int command, unsigned long arg) |
782 | { | |
783 | long ret; | |
784 | ||
785 | /* Check for properly constructed commands */ | |
786 | ret = check_ioctl_command(command); | |
787 | if (ret < 0) | |
788 | return (int)ret; | |
789 | ||
790 | return (int)dispatch_ioctl_command(command, arg); | |
791 | } | |
792 | ||
793 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
794 | ||
8bb8aefd YL |
795 | /* Compat structure for the ORANGEFS_DEV_MAP ioctl */ |
796 | struct ORANGEFS_dev_map_desc32 { | |
5db11c21 MM |
797 | compat_uptr_t ptr; |
798 | __s32 total_size; | |
799 | __s32 size; | |
800 | __s32 count; | |
801 | }; | |
802 | ||
803 | static unsigned long translate_dev_map26(unsigned long args, long *error) | |
804 | { | |
8bb8aefd | 805 | struct ORANGEFS_dev_map_desc32 __user *p32 = (void __user *)args; |
5db11c21 MM |
806 | /* |
807 | * Depending on the architecture, allocate some space on the | |
808 | * user-call-stack based on our expected layout. | |
809 | */ | |
8bb8aefd | 810 | struct ORANGEFS_dev_map_desc __user *p = |
5db11c21 | 811 | compat_alloc_user_space(sizeof(*p)); |
84d02150 | 812 | compat_uptr_t addr; |
5db11c21 MM |
813 | |
814 | *error = 0; | |
815 | /* get the ptr from the 32 bit user-space */ | |
816 | if (get_user(addr, &p32->ptr)) | |
817 | goto err; | |
818 | /* try to put that into a 64-bit layout */ | |
819 | if (put_user(compat_ptr(addr), &p->ptr)) | |
820 | goto err; | |
821 | /* copy the remaining fields */ | |
822 | if (copy_in_user(&p->total_size, &p32->total_size, sizeof(__s32))) | |
823 | goto err; | |
824 | if (copy_in_user(&p->size, &p32->size, sizeof(__s32))) | |
825 | goto err; | |
826 | if (copy_in_user(&p->count, &p32->count, sizeof(__s32))) | |
827 | goto err; | |
828 | return (unsigned long)p; | |
829 | err: | |
830 | *error = -EFAULT; | |
831 | return 0; | |
832 | } | |
833 | ||
834 | /* | |
835 | * 32 bit user-space apps' ioctl handlers when kernel modules | |
836 | * is compiled as a 64 bit one | |
837 | */ | |
8bb8aefd | 838 | static long orangefs_devreq_compat_ioctl(struct file *filp, unsigned int cmd, |
5db11c21 MM |
839 | unsigned long args) |
840 | { | |
841 | long ret; | |
842 | unsigned long arg = args; | |
843 | ||
844 | /* Check for properly constructed commands */ | |
845 | ret = check_ioctl_command(cmd); | |
846 | if (ret < 0) | |
847 | return ret; | |
8bb8aefd | 848 | if (cmd == ORANGEFS_DEV_MAP) { |
5db11c21 MM |
849 | /* |
850 | * convert the arguments to what we expect internally | |
851 | * in kernel space | |
852 | */ | |
853 | arg = translate_dev_map26(args, &ret); | |
854 | if (ret < 0) { | |
855 | gossip_err("Could not translate dev map\n"); | |
856 | return ret; | |
857 | } | |
858 | } | |
859 | /* no other ioctl requires translation */ | |
860 | return dispatch_ioctl_command(cmd, arg); | |
861 | } | |
862 | ||
2c590d5f MM |
863 | #endif /* CONFIG_COMPAT is in .config */ |
864 | ||
5db11c21 | 865 | /* the assigned character device major number */ |
8bb8aefd | 866 | static int orangefs_dev_major; |
5db11c21 MM |
867 | |
868 | /* | |
8bb8aefd | 869 | * Initialize orangefs device specific state: |
5db11c21 MM |
870 | * Must be called at module load time only |
871 | */ | |
8bb8aefd | 872 | int orangefs_dev_init(void) |
5db11c21 | 873 | { |
8bb8aefd YL |
874 | /* register orangefs-req device */ |
875 | orangefs_dev_major = register_chrdev(0, | |
876 | ORANGEFS_REQDEVICE_NAME, | |
877 | &orangefs_devreq_file_operations); | |
878 | if (orangefs_dev_major < 0) { | |
5db11c21 MM |
879 | gossip_debug(GOSSIP_DEV_DEBUG, |
880 | "Failed to register /dev/%s (error %d)\n", | |
8bb8aefd | 881 | ORANGEFS_REQDEVICE_NAME, orangefs_dev_major); |
8bb8aefd | 882 | return orangefs_dev_major; |
5db11c21 MM |
883 | } |
884 | ||
885 | gossip_debug(GOSSIP_DEV_DEBUG, | |
886 | "*** /dev/%s character device registered ***\n", | |
8bb8aefd | 887 | ORANGEFS_REQDEVICE_NAME); |
5db11c21 | 888 | gossip_debug(GOSSIP_DEV_DEBUG, "'mknod /dev/%s c %d 0'.\n", |
8bb8aefd | 889 | ORANGEFS_REQDEVICE_NAME, orangefs_dev_major); |
5db11c21 MM |
890 | return 0; |
891 | } | |
892 | ||
8bb8aefd | 893 | void orangefs_dev_cleanup(void) |
5db11c21 | 894 | { |
8bb8aefd | 895 | unregister_chrdev(orangefs_dev_major, ORANGEFS_REQDEVICE_NAME); |
5db11c21 MM |
896 | gossip_debug(GOSSIP_DEV_DEBUG, |
897 | "*** /dev/%s character device unregistered ***\n", | |
8bb8aefd | 898 | ORANGEFS_REQDEVICE_NAME); |
5db11c21 MM |
899 | } |
900 | ||
8bb8aefd | 901 | static unsigned int orangefs_devreq_poll(struct file *file, |
5db11c21 MM |
902 | struct poll_table_struct *poll_table) |
903 | { | |
904 | int poll_revent_mask = 0; | |
905 | ||
83595db0 | 906 | poll_wait(file, &orangefs_request_list_waitq, poll_table); |
5db11c21 | 907 | |
83595db0 AV |
908 | if (!list_empty(&orangefs_request_list)) |
909 | poll_revent_mask |= POLL_IN; | |
5db11c21 MM |
910 | return poll_revent_mask; |
911 | } | |
912 | ||
8bb8aefd | 913 | const struct file_operations orangefs_devreq_file_operations = { |
5db11c21 | 914 | .owner = THIS_MODULE, |
8bb8aefd YL |
915 | .read = orangefs_devreq_read, |
916 | .write_iter = orangefs_devreq_write_iter, | |
917 | .open = orangefs_devreq_open, | |
918 | .release = orangefs_devreq_release, | |
919 | .unlocked_ioctl = orangefs_devreq_ioctl, | |
5db11c21 MM |
920 | |
921 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
8bb8aefd | 922 | .compat_ioctl = orangefs_devreq_compat_ioctl, |
5db11c21 | 923 | #endif |
8bb8aefd | 924 | .poll = orangefs_devreq_poll |
5db11c21 | 925 | }; |