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 | ||
acfcbaf1 MB |
77 | /* Returns whether any FS are still pending remounted */ |
78 | static int mark_all_pending_mounts(void) | |
79 | { | |
80 | int unmounted = 1; | |
81 | struct orangefs_sb_info_s *orangefs_sb = NULL; | |
82 | ||
83 | spin_lock(&orangefs_superblocks_lock); | |
84 | list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) { | |
85 | /* All of these file system require a remount */ | |
86 | orangefs_sb->mount_pending = 1; | |
87 | unmounted = 0; | |
88 | } | |
89 | spin_unlock(&orangefs_superblocks_lock); | |
90 | return unmounted; | |
91 | } | |
92 | ||
93 | /* | |
94 | * Determine if a given file system needs to be remounted or not | |
95 | * Returns -1 on error | |
96 | * 0 if already mounted | |
97 | * 1 if needs remount | |
98 | */ | |
99 | static int fs_mount_pending(__s32 fsid) | |
100 | { | |
101 | int mount_pending = -1; | |
102 | struct orangefs_sb_info_s *orangefs_sb = NULL; | |
103 | ||
104 | spin_lock(&orangefs_superblocks_lock); | |
105 | list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) { | |
106 | if (orangefs_sb->fs_id == fsid) { | |
107 | mount_pending = orangefs_sb->mount_pending; | |
108 | break; | |
109 | } | |
110 | } | |
111 | spin_unlock(&orangefs_superblocks_lock); | |
112 | return mount_pending; | |
113 | } | |
114 | ||
8bb8aefd | 115 | static int orangefs_devreq_open(struct inode *inode, struct file *file) |
5db11c21 MM |
116 | { |
117 | int ret = -EINVAL; | |
118 | ||
119 | if (!(file->f_flags & O_NONBLOCK)) { | |
97f10027 MM |
120 | gossip_err("%s: device cannot be opened in blocking mode\n", |
121 | __func__); | |
5db11c21 MM |
122 | goto out; |
123 | } | |
124 | ret = -EACCES; | |
97f10027 | 125 | gossip_debug(GOSSIP_DEV_DEBUG, "client-core: opening device\n"); |
5db11c21 MM |
126 | mutex_lock(&devreq_mutex); |
127 | ||
128 | if (open_access_count == 0) { | |
fee25ce1 | 129 | open_access_count = 1; |
fb6d2526 | 130 | ret = 0; |
5db11c21 MM |
131 | } else { |
132 | DUMP_DEVICE_ERROR(); | |
133 | } | |
134 | mutex_unlock(&devreq_mutex); | |
135 | ||
136 | out: | |
137 | ||
138 | gossip_debug(GOSSIP_DEV_DEBUG, | |
139 | "pvfs2-client-core: open device complete (ret = %d)\n", | |
140 | ret); | |
141 | return ret; | |
142 | } | |
143 | ||
97f10027 | 144 | /* Function for read() callers into the device */ |
8bb8aefd | 145 | static ssize_t orangefs_devreq_read(struct file *file, |
5db11c21 MM |
146 | char __user *buf, |
147 | size_t count, loff_t *offset) | |
148 | { | |
8bb8aefd YL |
149 | struct orangefs_kernel_op_s *op, *temp; |
150 | __s32 proto_ver = ORANGEFS_KERNEL_PROTO_VERSION; | |
151 | static __s32 magic = ORANGEFS_DEVREQ_MAGIC; | |
152 | struct orangefs_kernel_op_s *cur_op = NULL; | |
24c8d080 | 153 | unsigned long ret; |
5db11c21 | 154 | |
24c8d080 | 155 | /* We do not support blocking IO. */ |
5db11c21 | 156 | if (!(file->f_flags & O_NONBLOCK)) { |
97f10027 MM |
157 | gossip_err("%s: blocking read from client-core.\n", |
158 | __func__); | |
5db11c21 | 159 | return -EINVAL; |
24c8d080 MB |
160 | } |
161 | ||
162 | /* | |
a762ae6d | 163 | * The client will do an ioctl to find MAX_DEV_REQ_UPSIZE, then |
24c8d080 MB |
164 | * always read with that size buffer. |
165 | */ | |
a762ae6d | 166 | if (count != MAX_DEV_REQ_UPSIZE) { |
24c8d080 MB |
167 | gossip_err("orangefs: client-core tried to read wrong size\n"); |
168 | return -EINVAL; | |
169 | } | |
170 | ||
ed42fe05 | 171 | restart: |
24c8d080 | 172 | /* Get next op (if any) from top of list. */ |
8bb8aefd YL |
173 | spin_lock(&orangefs_request_list_lock); |
174 | list_for_each_entry_safe(op, temp, &orangefs_request_list, list) { | |
24c8d080 MB |
175 | __s32 fsid; |
176 | /* This lock is held past the end of the loop when we break. */ | |
177 | spin_lock(&op->lock); | |
05a50a5b | 178 | if (unlikely(op_state_purged(op) || op_state_given_up(op))) { |
ed42fe05 AV |
179 | spin_unlock(&op->lock); |
180 | continue; | |
181 | } | |
24c8d080 MB |
182 | |
183 | fsid = fsid_of_op(op); | |
8bb8aefd | 184 | if (fsid != ORANGEFS_FS_ID_NULL) { |
24c8d080 MB |
185 | int ret; |
186 | /* Skip ops whose filesystem needs to be mounted. */ | |
187 | ret = fs_mount_pending(fsid); | |
188 | if (ret == 1) { | |
5db11c21 | 189 | gossip_debug(GOSSIP_DEV_DEBUG, |
5090c967 MM |
190 | "%s: mount pending, skipping op tag " |
191 | "%llu %s\n", | |
192 | __func__, | |
193 | llu(op->tag), | |
194 | get_opname_string(op)); | |
24c8d080 MB |
195 | spin_unlock(&op->lock); |
196 | continue; | |
97f10027 MM |
197 | /* |
198 | * Skip ops whose filesystem we don't know about unless | |
199 | * it is being mounted. | |
200 | */ | |
24c8d080 MB |
201 | /* XXX: is there a better way to detect this? */ |
202 | } else if (ret == -1 && | |
97f10027 MM |
203 | !(op->upcall.type == |
204 | ORANGEFS_VFS_OP_FS_MOUNT || | |
205 | op->upcall.type == | |
206 | ORANGEFS_VFS_OP_GETATTR)) { | |
24c8d080 MB |
207 | gossip_debug(GOSSIP_DEV_DEBUG, |
208 | "orangefs: skipping op tag %llu %s\n", | |
209 | llu(op->tag), get_opname_string(op)); | |
210 | gossip_err( | |
211 | "orangefs: ERROR: fs_mount_pending %d\n", | |
212 | fsid); | |
213 | spin_unlock(&op->lock); | |
5db11c21 | 214 | continue; |
5db11c21 MM |
215 | } |
216 | } | |
24c8d080 MB |
217 | /* |
218 | * Either this op does not pertain to a filesystem, is mounting | |
219 | * a filesystem, or pertains to a mounted filesystem. Let it | |
220 | * through. | |
221 | */ | |
222 | cur_op = op; | |
223 | break; | |
224 | } | |
225 | ||
226 | /* | |
227 | * At this point we either have a valid op and can continue or have not | |
228 | * found an op and must ask the client to try again later. | |
229 | */ | |
230 | if (!cur_op) { | |
8bb8aefd | 231 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 232 | return -EAGAIN; |
5db11c21 MM |
233 | } |
234 | ||
ca9f518e MM |
235 | gossip_debug(GOSSIP_DEV_DEBUG, "%s: reading op tag %llu %s\n", |
236 | __func__, | |
237 | llu(cur_op->tag), | |
238 | get_opname_string(cur_op)); | |
5db11c21 | 239 | |
24c8d080 MB |
240 | /* |
241 | * Such an op should never be on the list in the first place. If so, we | |
242 | * will abort. | |
243 | */ | |
244 | if (op_state_in_progress(cur_op) || op_state_serviced(cur_op)) { | |
245 | gossip_err("orangefs: ERROR: Current op already queued.\n"); | |
05a50a5b | 246 | list_del_init(&cur_op->list); |
5db11c21 | 247 | spin_unlock(&cur_op->lock); |
8bb8aefd | 248 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 249 | return -EAGAIN; |
5db11c21 | 250 | } |
ca9f518e | 251 | |
ed42fe05 | 252 | list_del_init(&cur_op->list); |
8bb8aefd | 253 | spin_unlock(&orangefs_request_list_lock); |
ed42fe05 | 254 | |
24c8d080 MB |
255 | spin_unlock(&cur_op->lock); |
256 | ||
257 | /* Push the upcall out. */ | |
258 | ret = copy_to_user(buf, &proto_ver, sizeof(__s32)); | |
259 | if (ret != 0) | |
260 | goto error; | |
261 | ret = copy_to_user(buf+sizeof(__s32), &magic, sizeof(__s32)); | |
262 | if (ret != 0) | |
263 | goto error; | |
264 | ret = copy_to_user(buf+2 * sizeof(__s32), &cur_op->tag, sizeof(__u64)); | |
265 | if (ret != 0) | |
266 | goto error; | |
267 | ret = copy_to_user(buf+2*sizeof(__s32)+sizeof(__u64), &cur_op->upcall, | |
8bb8aefd | 268 | sizeof(struct orangefs_upcall_s)); |
24c8d080 MB |
269 | if (ret != 0) |
270 | goto error; | |
271 | ||
ed42fe05 AV |
272 | spin_lock(&htable_ops_in_progress_lock); |
273 | spin_lock(&cur_op->lock); | |
274 | if (unlikely(op_state_given_up(cur_op))) { | |
275 | spin_unlock(&cur_op->lock); | |
276 | spin_unlock(&htable_ops_in_progress_lock); | |
05a50a5b | 277 | complete(&cur_op->waitq); |
ed42fe05 AV |
278 | goto restart; |
279 | } | |
280 | ||
281 | /* | |
282 | * Set the operation to be in progress and move it between lists since | |
283 | * it has been sent to the client. | |
284 | */ | |
285 | set_op_state_inprogress(cur_op); | |
9d9e7ba9 MM |
286 | gossip_debug(GOSSIP_DEV_DEBUG, |
287 | "%s: 1 op:%s: op_state:%d: process:%s:\n", | |
288 | __func__, | |
289 | get_opname_string(cur_op), | |
290 | cur_op->op_state, | |
291 | current->comm); | |
ed42fe05 AV |
292 | orangefs_devreq_add_op(cur_op); |
293 | spin_unlock(&cur_op->lock); | |
294 | spin_unlock(&htable_ops_in_progress_lock); | |
ed42fe05 | 295 | |
24c8d080 | 296 | /* The client only asks to read one size buffer. */ |
a762ae6d | 297 | return MAX_DEV_REQ_UPSIZE; |
24c8d080 MB |
298 | error: |
299 | /* | |
300 | * We were unable to copy the op data to the client. Put the op back in | |
301 | * list. If client has crashed, the op will be purged later when the | |
302 | * device is released. | |
303 | */ | |
304 | gossip_err("orangefs: Failed to copy data to user space\n"); | |
8bb8aefd | 305 | spin_lock(&orangefs_request_list_lock); |
24c8d080 | 306 | spin_lock(&cur_op->lock); |
ed42fe05 AV |
307 | if (likely(!op_state_given_up(cur_op))) { |
308 | set_op_state_waiting(cur_op); | |
9d9e7ba9 MM |
309 | gossip_debug(GOSSIP_DEV_DEBUG, |
310 | "%s: 2 op:%s: op_state:%d: process:%s:\n", | |
311 | __func__, | |
312 | get_opname_string(cur_op), | |
313 | cur_op->op_state, | |
314 | current->comm); | |
ed42fe05 | 315 | list_add(&cur_op->list, &orangefs_request_list); |
05a50a5b AV |
316 | spin_unlock(&cur_op->lock); |
317 | } else { | |
318 | spin_unlock(&cur_op->lock); | |
319 | complete(&cur_op->waitq); | |
ed42fe05 | 320 | } |
8bb8aefd | 321 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 322 | return -EFAULT; |
5db11c21 MM |
323 | } |
324 | ||
97f10027 | 325 | /* |
b3ae4755 MM |
326 | * Function for writev() callers into the device. |
327 | * | |
328 | * Userspace should have written: | |
329 | * - __u32 version | |
330 | * - __u32 magic | |
331 | * - __u64 tag | |
332 | * - struct orangefs_downcall_s | |
333 | * - trailer buffer (in the case of READDIR operations) | |
97f10027 | 334 | */ |
b3ae4755 MM |
335 | static ssize_t orangefs_devreq_write_iter(struct kiocb *iocb, |
336 | struct iov_iter *iter) | |
5db11c21 | 337 | { |
b3ae4755 | 338 | ssize_t ret; |
8bb8aefd | 339 | struct orangefs_kernel_op_s *op = NULL; |
b3ae4755 MM |
340 | struct { |
341 | __u32 version; | |
342 | __u32 magic; | |
343 | __u64 tag; | |
344 | } head; | |
345 | int total = ret = iov_iter_count(iter); | |
346 | int n; | |
347 | int downcall_size = sizeof(struct orangefs_downcall_s); | |
348 | int head_size = sizeof(head); | |
349 | ||
350 | gossip_debug(GOSSIP_DEV_DEBUG, "%s: total:%d: ret:%zd:\n", | |
351 | __func__, | |
352 | total, | |
353 | ret); | |
5db11c21 | 354 | |
b3ae4755 | 355 | if (total < MAX_DEV_REQ_DOWNSIZE) { |
cf0c2771 | 356 | gossip_err("%s: total:%d: must be at least:%u:\n", |
b3ae4755 MM |
357 | __func__, |
358 | total, | |
cf0c2771 | 359 | (unsigned int) MAX_DEV_REQ_DOWNSIZE); |
ed42fe05 | 360 | return -EFAULT; |
5db11c21 | 361 | } |
b3ae4755 MM |
362 | |
363 | n = copy_from_iter(&head, head_size, iter); | |
364 | if (n < head_size) { | |
365 | gossip_err("%s: failed to copy head.\n", __func__); | |
ed42fe05 | 366 | return -EFAULT; |
97f10027 | 367 | } |
b3ae4755 MM |
368 | |
369 | if (head.version < ORANGEFS_MINIMUM_USERSPACE_VERSION) { | |
370 | gossip_err("%s: userspace claims version" | |
371 | "%d, minimum version required: %d.\n", | |
372 | __func__, | |
373 | head.version, | |
374 | ORANGEFS_MINIMUM_USERSPACE_VERSION); | |
ed42fe05 | 375 | return -EPROTO; |
5db11c21 | 376 | } |
5db11c21 | 377 | |
b3ae4755 MM |
378 | if (head.magic != ORANGEFS_DEVREQ_MAGIC) { |
379 | gossip_err("Error: Device magic number does not match.\n"); | |
ed42fe05 | 380 | return -EPROTO; |
b3ae4755 | 381 | } |
5db11c21 | 382 | |
ca9f518e | 383 | /* remove the op from the in progress hash table */ |
b3ae4755 MM |
384 | op = orangefs_devreq_remove_op(head.tag); |
385 | if (!op) { | |
386 | gossip_err("WARNING: No one's waiting for tag %llu\n", | |
387 | llu(head.tag)); | |
ed42fe05 | 388 | return ret; |
b3ae4755 | 389 | } |
5db11c21 | 390 | |
b3ae4755 MM |
391 | n = copy_from_iter(&op->downcall, downcall_size, iter); |
392 | if (n != downcall_size) { | |
393 | gossip_err("%s: failed to copy downcall.\n", __func__); | |
5964c1b8 | 394 | goto Efault; |
5db11c21 MM |
395 | } |
396 | ||
b3ae4755 MM |
397 | if (op->downcall.status) |
398 | goto wakeup; | |
97f10027 | 399 | |
b3ae4755 MM |
400 | /* |
401 | * We've successfully peeled off the head and the downcall. | |
402 | * Something has gone awry if total doesn't equal the | |
403 | * sum of head_size, downcall_size and trailer_size. | |
404 | */ | |
405 | if ((head_size + downcall_size + op->downcall.trailer_size) != total) { | |
406 | gossip_err("%s: funky write, head_size:%d" | |
407 | ": downcall_size:%d: trailer_size:%lld" | |
408 | ": total size:%d:\n", | |
409 | __func__, | |
410 | head_size, | |
411 | downcall_size, | |
412 | op->downcall.trailer_size, | |
413 | total); | |
5964c1b8 | 414 | goto Efault; |
b3ae4755 | 415 | } |
97f10027 | 416 | |
b3ae4755 MM |
417 | /* Only READDIR operations should have trailers. */ |
418 | if ((op->downcall.type != ORANGEFS_VFS_OP_READDIR) && | |
419 | (op->downcall.trailer_size != 0)) { | |
420 | gossip_err("%s: %x operation with trailer.", | |
421 | __func__, | |
422 | op->downcall.type); | |
5964c1b8 | 423 | goto Efault; |
b3ae4755 | 424 | } |
97f10027 | 425 | |
b3ae4755 MM |
426 | /* READDIR operations should always have trailers. */ |
427 | if ((op->downcall.type == ORANGEFS_VFS_OP_READDIR) && | |
428 | (op->downcall.trailer_size == 0)) { | |
429 | gossip_err("%s: %x operation with no trailer.", | |
430 | __func__, | |
431 | op->downcall.type); | |
5964c1b8 | 432 | goto Efault; |
b3ae4755 | 433 | } |
97f10027 | 434 | |
b3ae4755 MM |
435 | if (op->downcall.type != ORANGEFS_VFS_OP_READDIR) |
436 | goto wakeup; | |
5db11c21 | 437 | |
b3ae4755 MM |
438 | op->downcall.trailer_buf = |
439 | vmalloc(op->downcall.trailer_size); | |
440 | if (op->downcall.trailer_buf == NULL) { | |
441 | gossip_err("%s: failed trailer vmalloc.\n", | |
442 | __func__); | |
5964c1b8 | 443 | goto Enomem; |
b3ae4755 MM |
444 | } |
445 | memset(op->downcall.trailer_buf, 0, op->downcall.trailer_size); | |
446 | n = copy_from_iter(op->downcall.trailer_buf, | |
447 | op->downcall.trailer_size, | |
448 | iter); | |
449 | if (n != op->downcall.trailer_size) { | |
450 | gossip_err("%s: failed to copy trailer.\n", __func__); | |
451 | vfree(op->downcall.trailer_buf); | |
5964c1b8 | 452 | goto Efault; |
b3ae4755 | 453 | } |
97f10027 | 454 | |
b3ae4755 | 455 | wakeup: |
2a9e5c22 | 456 | /* |
9f08cfe9 MM |
457 | * Return to vfs waitqueue, and back to service_operation |
458 | * through wait_for_matching_downcall. | |
2a9e5c22 AV |
459 | */ |
460 | spin_lock(&op->lock); | |
5964c1b8 | 461 | if (unlikely(op_is_cancel(op))) { |
2a9e5c22 | 462 | spin_unlock(&op->lock); |
78699e29 | 463 | put_cancel(op); |
5964c1b8 AV |
464 | } else if (unlikely(op_state_given_up(op))) { |
465 | spin_unlock(&op->lock); | |
05a50a5b | 466 | complete(&op->waitq); |
5964c1b8 AV |
467 | } else { |
468 | set_op_state_serviced(op); | |
9d9e7ba9 MM |
469 | gossip_debug(GOSSIP_DEV_DEBUG, |
470 | "%s: op:%s: op_state:%d: process:%s:\n", | |
471 | __func__, | |
472 | get_opname_string(op), | |
473 | op->op_state, | |
474 | current->comm); | |
5964c1b8 AV |
475 | spin_unlock(&op->lock); |
476 | } | |
b3ae4755 | 477 | return ret; |
ed42fe05 | 478 | |
5964c1b8 AV |
479 | Efault: |
480 | op->downcall.status = -(ORANGEFS_ERROR_BIT | 9); | |
481 | ret = -EFAULT; | |
482 | goto wakeup; | |
483 | ||
484 | Enomem: | |
485 | op->downcall.status = -(ORANGEFS_ERROR_BIT | 8); | |
486 | ret = -ENOMEM; | |
487 | goto wakeup; | |
5db11c21 MM |
488 | } |
489 | ||
5db11c21 MM |
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; | |
45996492 | 575 | struct orangefs_sb_info_s *orangefs_sb; |
5db11c21 MM |
576 | |
577 | /* mtmoore: add locking here */ | |
578 | ||
579 | switch (command) { | |
8bb8aefd | 580 | case ORANGEFS_DEV_GET_MAGIC: |
5db11c21 MM |
581 | return ((put_user(magic, (__s32 __user *) arg) == -EFAULT) ? |
582 | -EIO : | |
583 | 0); | |
8bb8aefd | 584 | case ORANGEFS_DEV_GET_MAX_UPSIZE: |
5db11c21 MM |
585 | return ((put_user(max_up_size, |
586 | (__s32 __user *) arg) == -EFAULT) ? | |
587 | -EIO : | |
588 | 0); | |
8bb8aefd | 589 | case ORANGEFS_DEV_GET_MAX_DOWNSIZE: |
5db11c21 MM |
590 | return ((put_user(max_down_size, |
591 | (__s32 __user *) arg) == -EFAULT) ? | |
592 | -EIO : | |
593 | 0); | |
8bb8aefd | 594 | case ORANGEFS_DEV_MAP: |
5db11c21 | 595 | ret = copy_from_user(&user_desc, |
8bb8aefd | 596 | (struct ORANGEFS_dev_map_desc __user *) |
5db11c21 | 597 | arg, |
8bb8aefd | 598 | sizeof(struct ORANGEFS_dev_map_desc)); |
ea2c9c9f AV |
599 | /* WTF -EIO and not -EFAULT? */ |
600 | return ret ? -EIO : orangefs_bufmap_initialize(&user_desc); | |
8bb8aefd | 601 | case ORANGEFS_DEV_REMOUNT_ALL: |
5db11c21 | 602 | gossip_debug(GOSSIP_DEV_DEBUG, |
97f10027 MM |
603 | "%s: got ORANGEFS_DEV_REMOUNT_ALL\n", |
604 | __func__); | |
5db11c21 MM |
605 | |
606 | /* | |
8bb8aefd | 607 | * remount all mounted orangefs volumes to regain the lost |
5db11c21 MM |
608 | * dynamic mount tables (if any) -- NOTE: this is done |
609 | * without keeping the superblock list locked due to the | |
adcf34a2 | 610 | * upcall/downcall waiting. also, the request mutex is |
5db11c21 MM |
611 | * used to ensure that no operations will be serviced until |
612 | * all of the remounts are serviced (to avoid ops between | |
613 | * mounts to fail) | |
614 | */ | |
615 | ret = mutex_lock_interruptible(&request_mutex); | |
616 | if (ret < 0) | |
617 | return ret; | |
618 | gossip_debug(GOSSIP_DEV_DEBUG, | |
97f10027 MM |
619 | "%s: priority remount in progress\n", |
620 | __func__); | |
45996492 AV |
621 | spin_lock(&orangefs_superblocks_lock); |
622 | list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) { | |
623 | /* | |
624 | * We have to drop the spinlock, so entries can be | |
625 | * removed. They can't be freed, though, so we just | |
626 | * keep the forward pointers and zero the back ones - | |
627 | * that way we can get to the rest of the list. | |
628 | */ | |
629 | if (!orangefs_sb->list.prev) | |
630 | continue; | |
631 | gossip_debug(GOSSIP_DEV_DEBUG, | |
632 | "%s: Remounting SB %p\n", | |
633 | __func__, | |
634 | orangefs_sb); | |
635 | ||
636 | spin_unlock(&orangefs_superblocks_lock); | |
637 | ret = orangefs_remount(orangefs_sb); | |
638 | spin_lock(&orangefs_superblocks_lock); | |
639 | if (ret) { | |
5db11c21 | 640 | gossip_debug(GOSSIP_DEV_DEBUG, |
45996492 | 641 | "SB %p remount failed\n", |
8bb8aefd | 642 | orangefs_sb); |
45996492 | 643 | break; |
5db11c21 MM |
644 | } |
645 | } | |
45996492 | 646 | spin_unlock(&orangefs_superblocks_lock); |
5db11c21 | 647 | gossip_debug(GOSSIP_DEV_DEBUG, |
97f10027 MM |
648 | "%s: priority remount complete\n", |
649 | __func__); | |
5db11c21 MM |
650 | mutex_unlock(&request_mutex); |
651 | return ret; | |
652 | ||
8bb8aefd | 653 | case ORANGEFS_DEV_UPSTREAM: |
5db11c21 MM |
654 | ret = copy_to_user((void __user *)arg, |
655 | &upstream_kmod, | |
656 | sizeof(upstream_kmod)); | |
657 | ||
658 | if (ret != 0) | |
659 | return -EIO; | |
660 | else | |
661 | return ret; | |
662 | ||
8bb8aefd | 663 | case ORANGEFS_DEV_CLIENT_MASK: |
5db11c21 MM |
664 | ret = copy_from_user(&mask2_info, |
665 | (void __user *)arg, | |
666 | sizeof(struct dev_mask2_info_s)); | |
667 | ||
668 | if (ret != 0) | |
669 | return -EIO; | |
670 | ||
671 | client_debug_mask.mask1 = mask2_info.mask1_value; | |
672 | client_debug_mask.mask2 = mask2_info.mask2_value; | |
673 | ||
674 | pr_info("%s: client debug mask has been been received " | |
675 | ":%llx: :%llx:\n", | |
676 | __func__, | |
677 | (unsigned long long)client_debug_mask.mask1, | |
678 | (unsigned long long)client_debug_mask.mask2); | |
679 | ||
680 | return ret; | |
681 | ||
8bb8aefd | 682 | case ORANGEFS_DEV_CLIENT_STRING: |
5db11c21 MM |
683 | ret = copy_from_user(&client_debug_array_string, |
684 | (void __user *)arg, | |
8bb8aefd | 685 | ORANGEFS_MAX_DEBUG_STRING_LEN); |
53f57fef MM |
686 | /* |
687 | * The real client-core makes an effort to ensure | |
688 | * that actual strings that aren't too long to fit in | |
689 | * this buffer is what we get here. We're going to use | |
690 | * string functions on the stuff we got, so we'll make | |
691 | * this extra effort to try and keep from | |
692 | * flowing out of this buffer when we use the string | |
693 | * functions, even if somehow the stuff we end up | |
694 | * with here is garbage. | |
695 | */ | |
696 | client_debug_array_string[ORANGEFS_MAX_DEBUG_STRING_LEN - 1] = | |
697 | '\0'; | |
698 | ||
5db11c21 | 699 | if (ret != 0) { |
97f10027 | 700 | pr_info("%s: CLIENT_STRING: copy_from_user failed\n", |
5db11c21 MM |
701 | __func__); |
702 | return -EIO; | |
703 | } | |
704 | ||
97f10027 | 705 | pr_info("%s: client debug array string has been received.\n", |
5db11c21 MM |
706 | __func__); |
707 | ||
708 | if (!help_string_initialized) { | |
709 | ||
710 | /* Free the "we don't know yet" default string... */ | |
711 | kfree(debug_help_string); | |
712 | ||
713 | /* build a proper debug help string */ | |
714 | if (orangefs_prepare_debugfs_help_string(0)) { | |
97f10027 | 715 | gossip_err("%s: no debug help string \n", |
5db11c21 MM |
716 | __func__); |
717 | return -EIO; | |
718 | } | |
719 | ||
720 | /* Replace the boilerplate boot-time debug-help file. */ | |
721 | debugfs_remove(help_file_dentry); | |
722 | ||
723 | help_file_dentry = | |
724 | debugfs_create_file( | |
725 | ORANGEFS_KMOD_DEBUG_HELP_FILE, | |
726 | 0444, | |
727 | debug_dir, | |
728 | debug_help_string, | |
729 | &debug_help_fops); | |
730 | ||
731 | if (!help_file_dentry) { | |
732 | gossip_err("%s: debugfs_create_file failed for" | |
733 | " :%s:!\n", | |
734 | __func__, | |
735 | ORANGEFS_KMOD_DEBUG_HELP_FILE); | |
736 | return -EIO; | |
737 | } | |
738 | } | |
739 | ||
740 | debug_mask_to_string(&client_debug_mask, 1); | |
741 | ||
742 | debugfs_remove(client_debug_dentry); | |
743 | ||
8bb8aefd | 744 | orangefs_client_debug_init(); |
5db11c21 MM |
745 | |
746 | help_string_initialized++; | |
747 | ||
748 | return ret; | |
749 | ||
8bb8aefd | 750 | case ORANGEFS_DEV_DEBUG: |
5db11c21 MM |
751 | ret = copy_from_user(&mask_info, |
752 | (void __user *)arg, | |
753 | sizeof(mask_info)); | |
754 | ||
755 | if (ret != 0) | |
756 | return -EIO; | |
757 | ||
758 | if (mask_info.mask_type == KERNEL_MASK) { | |
759 | if ((mask_info.mask_value == 0) | |
760 | && (kernel_mask_set_mod_init)) { | |
761 | /* | |
762 | * the kernel debug mask was set when the | |
763 | * kernel module was loaded; don't override | |
764 | * it if the client-core was started without | |
8bb8aefd | 765 | * a value for ORANGEFS_KMODMASK. |
5db11c21 MM |
766 | */ |
767 | return 0; | |
768 | } | |
769 | debug_mask_to_string(&mask_info.mask_value, | |
770 | mask_info.mask_type); | |
771 | gossip_debug_mask = mask_info.mask_value; | |
97f10027 | 772 | pr_info("%s: kernel debug mask has been modified to " |
5db11c21 | 773 | ":%s: :%llx:\n", |
97f10027 | 774 | __func__, |
5db11c21 MM |
775 | kernel_debug_string, |
776 | (unsigned long long)gossip_debug_mask); | |
777 | } else if (mask_info.mask_type == CLIENT_MASK) { | |
778 | debug_mask_to_string(&mask_info.mask_value, | |
779 | mask_info.mask_type); | |
97f10027 | 780 | pr_info("%s: client debug mask has been modified to" |
5db11c21 | 781 | ":%s: :%llx:\n", |
97f10027 | 782 | __func__, |
5db11c21 MM |
783 | client_debug_string, |
784 | llu(mask_info.mask_value)); | |
785 | } else { | |
786 | gossip_lerr("Invalid mask type....\n"); | |
787 | return -EINVAL; | |
788 | } | |
789 | ||
790 | return ret; | |
791 | ||
792 | default: | |
793 | return -ENOIOCTLCMD; | |
794 | } | |
795 | return -ENOIOCTLCMD; | |
796 | } | |
797 | ||
8bb8aefd | 798 | static long orangefs_devreq_ioctl(struct file *file, |
5db11c21 MM |
799 | unsigned int command, unsigned long arg) |
800 | { | |
801 | long ret; | |
802 | ||
803 | /* Check for properly constructed commands */ | |
804 | ret = check_ioctl_command(command); | |
805 | if (ret < 0) | |
806 | return (int)ret; | |
807 | ||
808 | return (int)dispatch_ioctl_command(command, arg); | |
809 | } | |
810 | ||
811 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
812 | ||
8bb8aefd YL |
813 | /* Compat structure for the ORANGEFS_DEV_MAP ioctl */ |
814 | struct ORANGEFS_dev_map_desc32 { | |
5db11c21 MM |
815 | compat_uptr_t ptr; |
816 | __s32 total_size; | |
817 | __s32 size; | |
818 | __s32 count; | |
819 | }; | |
820 | ||
821 | static unsigned long translate_dev_map26(unsigned long args, long *error) | |
822 | { | |
8bb8aefd | 823 | struct ORANGEFS_dev_map_desc32 __user *p32 = (void __user *)args; |
5db11c21 MM |
824 | /* |
825 | * Depending on the architecture, allocate some space on the | |
826 | * user-call-stack based on our expected layout. | |
827 | */ | |
8bb8aefd | 828 | struct ORANGEFS_dev_map_desc __user *p = |
5db11c21 | 829 | compat_alloc_user_space(sizeof(*p)); |
84d02150 | 830 | compat_uptr_t addr; |
5db11c21 MM |
831 | |
832 | *error = 0; | |
833 | /* get the ptr from the 32 bit user-space */ | |
834 | if (get_user(addr, &p32->ptr)) | |
835 | goto err; | |
836 | /* try to put that into a 64-bit layout */ | |
837 | if (put_user(compat_ptr(addr), &p->ptr)) | |
838 | goto err; | |
839 | /* copy the remaining fields */ | |
840 | if (copy_in_user(&p->total_size, &p32->total_size, sizeof(__s32))) | |
841 | goto err; | |
842 | if (copy_in_user(&p->size, &p32->size, sizeof(__s32))) | |
843 | goto err; | |
844 | if (copy_in_user(&p->count, &p32->count, sizeof(__s32))) | |
845 | goto err; | |
846 | return (unsigned long)p; | |
847 | err: | |
848 | *error = -EFAULT; | |
849 | return 0; | |
850 | } | |
851 | ||
852 | /* | |
853 | * 32 bit user-space apps' ioctl handlers when kernel modules | |
854 | * is compiled as a 64 bit one | |
855 | */ | |
8bb8aefd | 856 | static long orangefs_devreq_compat_ioctl(struct file *filp, unsigned int cmd, |
5db11c21 MM |
857 | unsigned long args) |
858 | { | |
859 | long ret; | |
860 | unsigned long arg = args; | |
861 | ||
862 | /* Check for properly constructed commands */ | |
863 | ret = check_ioctl_command(cmd); | |
864 | if (ret < 0) | |
865 | return ret; | |
8bb8aefd | 866 | if (cmd == ORANGEFS_DEV_MAP) { |
5db11c21 MM |
867 | /* |
868 | * convert the arguments to what we expect internally | |
869 | * in kernel space | |
870 | */ | |
871 | arg = translate_dev_map26(args, &ret); | |
872 | if (ret < 0) { | |
873 | gossip_err("Could not translate dev map\n"); | |
874 | return ret; | |
875 | } | |
876 | } | |
877 | /* no other ioctl requires translation */ | |
878 | return dispatch_ioctl_command(cmd, arg); | |
879 | } | |
880 | ||
2c590d5f MM |
881 | #endif /* CONFIG_COMPAT is in .config */ |
882 | ||
5db11c21 | 883 | /* the assigned character device major number */ |
8bb8aefd | 884 | static int orangefs_dev_major; |
5db11c21 MM |
885 | |
886 | /* | |
8bb8aefd | 887 | * Initialize orangefs device specific state: |
5db11c21 MM |
888 | * Must be called at module load time only |
889 | */ | |
8bb8aefd | 890 | int orangefs_dev_init(void) |
5db11c21 | 891 | { |
8bb8aefd YL |
892 | /* register orangefs-req device */ |
893 | orangefs_dev_major = register_chrdev(0, | |
894 | ORANGEFS_REQDEVICE_NAME, | |
895 | &orangefs_devreq_file_operations); | |
896 | if (orangefs_dev_major < 0) { | |
5db11c21 MM |
897 | gossip_debug(GOSSIP_DEV_DEBUG, |
898 | "Failed to register /dev/%s (error %d)\n", | |
8bb8aefd | 899 | ORANGEFS_REQDEVICE_NAME, orangefs_dev_major); |
8bb8aefd | 900 | return orangefs_dev_major; |
5db11c21 MM |
901 | } |
902 | ||
903 | gossip_debug(GOSSIP_DEV_DEBUG, | |
904 | "*** /dev/%s character device registered ***\n", | |
8bb8aefd | 905 | ORANGEFS_REQDEVICE_NAME); |
5db11c21 | 906 | gossip_debug(GOSSIP_DEV_DEBUG, "'mknod /dev/%s c %d 0'.\n", |
8bb8aefd | 907 | ORANGEFS_REQDEVICE_NAME, orangefs_dev_major); |
5db11c21 MM |
908 | return 0; |
909 | } | |
910 | ||
8bb8aefd | 911 | void orangefs_dev_cleanup(void) |
5db11c21 | 912 | { |
8bb8aefd | 913 | unregister_chrdev(orangefs_dev_major, ORANGEFS_REQDEVICE_NAME); |
5db11c21 MM |
914 | gossip_debug(GOSSIP_DEV_DEBUG, |
915 | "*** /dev/%s character device unregistered ***\n", | |
8bb8aefd | 916 | ORANGEFS_REQDEVICE_NAME); |
5db11c21 MM |
917 | } |
918 | ||
8bb8aefd | 919 | static unsigned int orangefs_devreq_poll(struct file *file, |
5db11c21 MM |
920 | struct poll_table_struct *poll_table) |
921 | { | |
922 | int poll_revent_mask = 0; | |
923 | ||
83595db0 | 924 | poll_wait(file, &orangefs_request_list_waitq, poll_table); |
5db11c21 | 925 | |
83595db0 AV |
926 | if (!list_empty(&orangefs_request_list)) |
927 | poll_revent_mask |= POLL_IN; | |
5db11c21 MM |
928 | return poll_revent_mask; |
929 | } | |
930 | ||
8bb8aefd | 931 | const struct file_operations orangefs_devreq_file_operations = { |
5db11c21 | 932 | .owner = THIS_MODULE, |
8bb8aefd YL |
933 | .read = orangefs_devreq_read, |
934 | .write_iter = orangefs_devreq_write_iter, | |
935 | .open = orangefs_devreq_open, | |
936 | .release = orangefs_devreq_release, | |
937 | .unlocked_ioctl = orangefs_devreq_ioctl, | |
5db11c21 MM |
938 | |
939 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
8bb8aefd | 940 | .compat_ioctl = orangefs_devreq_compat_ioctl, |
5db11c21 | 941 | #endif |
8bb8aefd | 942 | .poll = orangefs_devreq_poll |
5db11c21 | 943 | }; |