Commit | Line | Data |
---|---|---|
5db11c21 MM |
1 | /* |
2 | * (C) 2001 Clemson University and The University of Chicago | |
3 | * | |
4 | * Changes by Acxiom Corporation to add protocol version to kernel | |
5 | * communication, Copyright Acxiom Corporation, 2005. | |
6 | * | |
7 | * See COPYING in top-level directory. | |
8 | */ | |
9 | ||
10 | #include "protocol.h" | |
575e9461 MM |
11 | #include "orangefs-kernel.h" |
12 | #include "orangefs-dev-proto.h" | |
13 | #include "orangefs-bufmap.h" | |
5db11c21 MM |
14 | |
15 | #include <linux/debugfs.h> | |
16 | #include <linux/slab.h> | |
17 | ||
18 | /* this file implements the /dev/pvfs2-req device node */ | |
19 | ||
20 | static int open_access_count; | |
21 | ||
22 | #define DUMP_DEVICE_ERROR() \ | |
23 | do { \ | |
24 | gossip_err("*****************************************************\n");\ | |
8bb8aefd | 25 | gossip_err("ORANGEFS Device Error: You cannot open the device file "); \ |
5db11c21 | 26 | gossip_err("\n/dev/%s more than once. Please make sure that\nthere " \ |
8bb8aefd | 27 | "are no ", ORANGEFS_REQDEVICE_NAME); \ |
5db11c21 MM |
28 | gossip_err("instances of a program using this device\ncurrently " \ |
29 | "running. (You must verify this!)\n"); \ | |
30 | gossip_err("For example, you can use the lsof program as follows:\n");\ | |
31 | gossip_err("'lsof | grep %s' (run this as root)\n", \ | |
8bb8aefd | 32 | ORANGEFS_REQDEVICE_NAME); \ |
5db11c21 MM |
33 | gossip_err(" open_access_count = %d\n", open_access_count); \ |
34 | gossip_err("*****************************************************\n");\ | |
35 | } while (0) | |
36 | ||
37 | static int hash_func(__u64 tag, int table_size) | |
38 | { | |
2c590d5f | 39 | return do_div(tag, (unsigned int)table_size); |
5db11c21 MM |
40 | } |
41 | ||
8bb8aefd | 42 | static void orangefs_devreq_add_op(struct orangefs_kernel_op_s *op) |
5db11c21 MM |
43 | { |
44 | int index = hash_func(op->tag, hash_table_size); | |
45 | ||
5db11c21 | 46 | list_add_tail(&op->list, &htable_ops_in_progress[index]); |
5db11c21 MM |
47 | } |
48 | ||
8bb8aefd | 49 | static struct orangefs_kernel_op_s *orangefs_devreq_remove_op(__u64 tag) |
5db11c21 | 50 | { |
8bb8aefd | 51 | struct orangefs_kernel_op_s *op, *next; |
5db11c21 MM |
52 | int index; |
53 | ||
54 | index = hash_func(tag, hash_table_size); | |
55 | ||
56 | spin_lock(&htable_ops_in_progress_lock); | |
57 | list_for_each_entry_safe(op, | |
58 | next, | |
59 | &htable_ops_in_progress[index], | |
60 | list) { | |
ed42fe05 AV |
61 | if (op->tag == tag && !op_state_purged(op)) { |
62 | list_del_init(&op->list); | |
63 | get_op(op); /* increase ref count. */ | |
5db11c21 MM |
64 | spin_unlock(&htable_ops_in_progress_lock); |
65 | return op; | |
66 | } | |
67 | } | |
68 | ||
69 | spin_unlock(&htable_ops_in_progress_lock); | |
70 | return NULL; | |
71 | } | |
72 | ||
8bb8aefd | 73 | static int orangefs_devreq_open(struct inode *inode, struct file *file) |
5db11c21 MM |
74 | { |
75 | int ret = -EINVAL; | |
76 | ||
77 | if (!(file->f_flags & O_NONBLOCK)) { | |
97f10027 MM |
78 | gossip_err("%s: device cannot be opened in blocking mode\n", |
79 | __func__); | |
5db11c21 MM |
80 | goto out; |
81 | } | |
82 | ret = -EACCES; | |
97f10027 | 83 | gossip_debug(GOSSIP_DEV_DEBUG, "client-core: opening device\n"); |
5db11c21 MM |
84 | mutex_lock(&devreq_mutex); |
85 | ||
86 | if (open_access_count == 0) { | |
fee25ce1 | 87 | open_access_count = 1; |
fb6d2526 | 88 | ret = 0; |
5db11c21 MM |
89 | } else { |
90 | DUMP_DEVICE_ERROR(); | |
91 | } | |
92 | mutex_unlock(&devreq_mutex); | |
93 | ||
94 | out: | |
95 | ||
96 | gossip_debug(GOSSIP_DEV_DEBUG, | |
97 | "pvfs2-client-core: open device complete (ret = %d)\n", | |
98 | ret); | |
99 | return ret; | |
100 | } | |
101 | ||
97f10027 | 102 | /* Function for read() callers into the device */ |
8bb8aefd | 103 | static ssize_t orangefs_devreq_read(struct file *file, |
5db11c21 MM |
104 | char __user *buf, |
105 | size_t count, loff_t *offset) | |
106 | { | |
8bb8aefd YL |
107 | struct orangefs_kernel_op_s *op, *temp; |
108 | __s32 proto_ver = ORANGEFS_KERNEL_PROTO_VERSION; | |
109 | static __s32 magic = ORANGEFS_DEVREQ_MAGIC; | |
110 | struct orangefs_kernel_op_s *cur_op = NULL; | |
24c8d080 | 111 | unsigned long ret; |
5db11c21 | 112 | |
24c8d080 | 113 | /* We do not support blocking IO. */ |
5db11c21 | 114 | if (!(file->f_flags & O_NONBLOCK)) { |
97f10027 MM |
115 | gossip_err("%s: blocking read from client-core.\n", |
116 | __func__); | |
5db11c21 | 117 | return -EINVAL; |
24c8d080 MB |
118 | } |
119 | ||
120 | /* | |
a762ae6d | 121 | * The client will do an ioctl to find MAX_DEV_REQ_UPSIZE, then |
24c8d080 MB |
122 | * always read with that size buffer. |
123 | */ | |
a762ae6d | 124 | if (count != MAX_DEV_REQ_UPSIZE) { |
24c8d080 MB |
125 | gossip_err("orangefs: client-core tried to read wrong size\n"); |
126 | return -EINVAL; | |
127 | } | |
128 | ||
ed42fe05 | 129 | restart: |
24c8d080 | 130 | /* Get next op (if any) from top of list. */ |
8bb8aefd YL |
131 | spin_lock(&orangefs_request_list_lock); |
132 | list_for_each_entry_safe(op, temp, &orangefs_request_list, list) { | |
24c8d080 MB |
133 | __s32 fsid; |
134 | /* This lock is held past the end of the loop when we break. */ | |
135 | spin_lock(&op->lock); | |
ed42fe05 AV |
136 | if (unlikely(op_state_purged(op))) { |
137 | spin_unlock(&op->lock); | |
138 | continue; | |
139 | } | |
24c8d080 MB |
140 | |
141 | fsid = fsid_of_op(op); | |
8bb8aefd | 142 | if (fsid != ORANGEFS_FS_ID_NULL) { |
24c8d080 MB |
143 | int ret; |
144 | /* Skip ops whose filesystem needs to be mounted. */ | |
145 | ret = fs_mount_pending(fsid); | |
146 | if (ret == 1) { | |
5db11c21 | 147 | gossip_debug(GOSSIP_DEV_DEBUG, |
5090c967 MM |
148 | "%s: mount pending, skipping op tag " |
149 | "%llu %s\n", | |
150 | __func__, | |
151 | llu(op->tag), | |
152 | get_opname_string(op)); | |
24c8d080 MB |
153 | spin_unlock(&op->lock); |
154 | continue; | |
97f10027 MM |
155 | /* |
156 | * Skip ops whose filesystem we don't know about unless | |
157 | * it is being mounted. | |
158 | */ | |
24c8d080 MB |
159 | /* XXX: is there a better way to detect this? */ |
160 | } else if (ret == -1 && | |
97f10027 MM |
161 | !(op->upcall.type == |
162 | ORANGEFS_VFS_OP_FS_MOUNT || | |
163 | op->upcall.type == | |
164 | ORANGEFS_VFS_OP_GETATTR)) { | |
24c8d080 MB |
165 | gossip_debug(GOSSIP_DEV_DEBUG, |
166 | "orangefs: skipping op tag %llu %s\n", | |
167 | llu(op->tag), get_opname_string(op)); | |
168 | gossip_err( | |
169 | "orangefs: ERROR: fs_mount_pending %d\n", | |
170 | fsid); | |
171 | spin_unlock(&op->lock); | |
5db11c21 | 172 | continue; |
5db11c21 MM |
173 | } |
174 | } | |
24c8d080 MB |
175 | /* |
176 | * Either this op does not pertain to a filesystem, is mounting | |
177 | * a filesystem, or pertains to a mounted filesystem. Let it | |
178 | * through. | |
179 | */ | |
180 | cur_op = op; | |
181 | break; | |
182 | } | |
183 | ||
184 | /* | |
185 | * At this point we either have a valid op and can continue or have not | |
186 | * found an op and must ask the client to try again later. | |
187 | */ | |
188 | if (!cur_op) { | |
8bb8aefd | 189 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 190 | return -EAGAIN; |
5db11c21 MM |
191 | } |
192 | ||
24c8d080 MB |
193 | gossip_debug(GOSSIP_DEV_DEBUG, "orangefs: reading op tag %llu %s\n", |
194 | llu(cur_op->tag), get_opname_string(cur_op)); | |
5db11c21 | 195 | |
24c8d080 MB |
196 | /* |
197 | * Such an op should never be on the list in the first place. If so, we | |
198 | * will abort. | |
199 | */ | |
200 | if (op_state_in_progress(cur_op) || op_state_serviced(cur_op)) { | |
201 | gossip_err("orangefs: ERROR: Current op already queued.\n"); | |
202 | list_del(&cur_op->list); | |
5db11c21 | 203 | spin_unlock(&cur_op->lock); |
8bb8aefd | 204 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 205 | return -EAGAIN; |
5db11c21 | 206 | } |
ed42fe05 AV |
207 | list_del_init(&cur_op->list); |
208 | get_op(op); | |
8bb8aefd | 209 | spin_unlock(&orangefs_request_list_lock); |
ed42fe05 | 210 | |
24c8d080 MB |
211 | spin_unlock(&cur_op->lock); |
212 | ||
213 | /* Push the upcall out. */ | |
214 | ret = copy_to_user(buf, &proto_ver, sizeof(__s32)); | |
215 | if (ret != 0) | |
216 | goto error; | |
217 | ret = copy_to_user(buf+sizeof(__s32), &magic, sizeof(__s32)); | |
218 | if (ret != 0) | |
219 | goto error; | |
220 | ret = copy_to_user(buf+2 * sizeof(__s32), &cur_op->tag, sizeof(__u64)); | |
221 | if (ret != 0) | |
222 | goto error; | |
223 | ret = copy_to_user(buf+2*sizeof(__s32)+sizeof(__u64), &cur_op->upcall, | |
8bb8aefd | 224 | sizeof(struct orangefs_upcall_s)); |
24c8d080 MB |
225 | if (ret != 0) |
226 | goto error; | |
227 | ||
ed42fe05 AV |
228 | spin_lock(&htable_ops_in_progress_lock); |
229 | spin_lock(&cur_op->lock); | |
230 | if (unlikely(op_state_given_up(cur_op))) { | |
231 | spin_unlock(&cur_op->lock); | |
232 | spin_unlock(&htable_ops_in_progress_lock); | |
233 | op_release(cur_op); | |
234 | goto restart; | |
235 | } | |
236 | ||
237 | /* | |
238 | * Set the operation to be in progress and move it between lists since | |
239 | * it has been sent to the client. | |
240 | */ | |
241 | set_op_state_inprogress(cur_op); | |
242 | orangefs_devreq_add_op(cur_op); | |
243 | spin_unlock(&cur_op->lock); | |
244 | spin_unlock(&htable_ops_in_progress_lock); | |
245 | op_release(cur_op); | |
246 | ||
24c8d080 | 247 | /* The client only asks to read one size buffer. */ |
a762ae6d | 248 | return MAX_DEV_REQ_UPSIZE; |
24c8d080 MB |
249 | error: |
250 | /* | |
251 | * We were unable to copy the op data to the client. Put the op back in | |
252 | * list. If client has crashed, the op will be purged later when the | |
253 | * device is released. | |
254 | */ | |
255 | gossip_err("orangefs: Failed to copy data to user space\n"); | |
8bb8aefd | 256 | spin_lock(&orangefs_request_list_lock); |
24c8d080 | 257 | spin_lock(&cur_op->lock); |
ed42fe05 AV |
258 | if (likely(!op_state_given_up(cur_op))) { |
259 | set_op_state_waiting(cur_op); | |
260 | list_add(&cur_op->list, &orangefs_request_list); | |
261 | } | |
24c8d080 | 262 | spin_unlock(&cur_op->lock); |
8bb8aefd | 263 | spin_unlock(&orangefs_request_list_lock); |
ed42fe05 | 264 | op_release(cur_op); |
24c8d080 | 265 | return -EFAULT; |
5db11c21 MM |
266 | } |
267 | ||
97f10027 | 268 | /* |
b3ae4755 MM |
269 | * Function for writev() callers into the device. |
270 | * | |
271 | * Userspace should have written: | |
272 | * - __u32 version | |
273 | * - __u32 magic | |
274 | * - __u64 tag | |
275 | * - struct orangefs_downcall_s | |
276 | * - trailer buffer (in the case of READDIR operations) | |
97f10027 | 277 | */ |
b3ae4755 MM |
278 | static ssize_t orangefs_devreq_write_iter(struct kiocb *iocb, |
279 | struct iov_iter *iter) | |
5db11c21 | 280 | { |
b3ae4755 | 281 | ssize_t ret; |
8bb8aefd | 282 | struct orangefs_kernel_op_s *op = NULL; |
b3ae4755 MM |
283 | struct { |
284 | __u32 version; | |
285 | __u32 magic; | |
286 | __u64 tag; | |
287 | } head; | |
288 | int total = ret = iov_iter_count(iter); | |
289 | int n; | |
290 | int downcall_size = sizeof(struct orangefs_downcall_s); | |
291 | int head_size = sizeof(head); | |
292 | ||
293 | gossip_debug(GOSSIP_DEV_DEBUG, "%s: total:%d: ret:%zd:\n", | |
294 | __func__, | |
295 | total, | |
296 | ret); | |
5db11c21 | 297 | |
b3ae4755 | 298 | if (total < MAX_DEV_REQ_DOWNSIZE) { |
cf0c2771 | 299 | gossip_err("%s: total:%d: must be at least:%u:\n", |
b3ae4755 MM |
300 | __func__, |
301 | total, | |
cf0c2771 | 302 | (unsigned int) MAX_DEV_REQ_DOWNSIZE); |
ed42fe05 | 303 | return -EFAULT; |
5db11c21 | 304 | } |
b3ae4755 MM |
305 | |
306 | n = copy_from_iter(&head, head_size, iter); | |
307 | if (n < head_size) { | |
308 | gossip_err("%s: failed to copy head.\n", __func__); | |
ed42fe05 | 309 | return -EFAULT; |
97f10027 | 310 | } |
b3ae4755 MM |
311 | |
312 | if (head.version < ORANGEFS_MINIMUM_USERSPACE_VERSION) { | |
313 | gossip_err("%s: userspace claims version" | |
314 | "%d, minimum version required: %d.\n", | |
315 | __func__, | |
316 | head.version, | |
317 | ORANGEFS_MINIMUM_USERSPACE_VERSION); | |
ed42fe05 | 318 | return -EPROTO; |
5db11c21 | 319 | } |
5db11c21 | 320 | |
b3ae4755 MM |
321 | if (head.magic != ORANGEFS_DEVREQ_MAGIC) { |
322 | gossip_err("Error: Device magic number does not match.\n"); | |
ed42fe05 | 323 | return -EPROTO; |
b3ae4755 | 324 | } |
5db11c21 | 325 | |
b3ae4755 MM |
326 | op = orangefs_devreq_remove_op(head.tag); |
327 | if (!op) { | |
328 | gossip_err("WARNING: No one's waiting for tag %llu\n", | |
329 | llu(head.tag)); | |
ed42fe05 | 330 | return ret; |
b3ae4755 | 331 | } |
5db11c21 | 332 | |
b3ae4755 MM |
333 | n = copy_from_iter(&op->downcall, downcall_size, iter); |
334 | if (n != downcall_size) { | |
335 | gossip_err("%s: failed to copy downcall.\n", __func__); | |
b3ae4755 | 336 | ret = -EFAULT; |
ed42fe05 | 337 | goto Broken; |
5db11c21 MM |
338 | } |
339 | ||
b3ae4755 MM |
340 | if (op->downcall.status) |
341 | goto wakeup; | |
97f10027 | 342 | |
b3ae4755 MM |
343 | /* |
344 | * We've successfully peeled off the head and the downcall. | |
345 | * Something has gone awry if total doesn't equal the | |
346 | * sum of head_size, downcall_size and trailer_size. | |
347 | */ | |
348 | if ((head_size + downcall_size + op->downcall.trailer_size) != total) { | |
349 | gossip_err("%s: funky write, head_size:%d" | |
350 | ": downcall_size:%d: trailer_size:%lld" | |
351 | ": total size:%d:\n", | |
352 | __func__, | |
353 | head_size, | |
354 | downcall_size, | |
355 | op->downcall.trailer_size, | |
356 | total); | |
b3ae4755 | 357 | ret = -EFAULT; |
ed42fe05 | 358 | goto Broken; |
b3ae4755 | 359 | } |
97f10027 | 360 | |
b3ae4755 MM |
361 | /* Only READDIR operations should have trailers. */ |
362 | if ((op->downcall.type != ORANGEFS_VFS_OP_READDIR) && | |
363 | (op->downcall.trailer_size != 0)) { | |
364 | gossip_err("%s: %x operation with trailer.", | |
365 | __func__, | |
366 | op->downcall.type); | |
b3ae4755 | 367 | ret = -EFAULT; |
ed42fe05 | 368 | goto Broken; |
b3ae4755 | 369 | } |
97f10027 | 370 | |
b3ae4755 MM |
371 | /* READDIR operations should always have trailers. */ |
372 | if ((op->downcall.type == ORANGEFS_VFS_OP_READDIR) && | |
373 | (op->downcall.trailer_size == 0)) { | |
374 | gossip_err("%s: %x operation with no trailer.", | |
375 | __func__, | |
376 | op->downcall.type); | |
b3ae4755 | 377 | ret = -EFAULT; |
ed42fe05 | 378 | goto Broken; |
b3ae4755 | 379 | } |
97f10027 | 380 | |
b3ae4755 MM |
381 | if (op->downcall.type != ORANGEFS_VFS_OP_READDIR) |
382 | goto wakeup; | |
5db11c21 | 383 | |
b3ae4755 MM |
384 | op->downcall.trailer_buf = |
385 | vmalloc(op->downcall.trailer_size); | |
386 | if (op->downcall.trailer_buf == NULL) { | |
387 | gossip_err("%s: failed trailer vmalloc.\n", | |
388 | __func__); | |
b3ae4755 | 389 | ret = -ENOMEM; |
ed42fe05 | 390 | goto Broken; |
b3ae4755 MM |
391 | } |
392 | memset(op->downcall.trailer_buf, 0, op->downcall.trailer_size); | |
393 | n = copy_from_iter(op->downcall.trailer_buf, | |
394 | op->downcall.trailer_size, | |
395 | iter); | |
396 | if (n != op->downcall.trailer_size) { | |
397 | gossip_err("%s: failed to copy trailer.\n", __func__); | |
398 | vfree(op->downcall.trailer_buf); | |
b3ae4755 | 399 | ret = -EFAULT; |
ed42fe05 | 400 | goto Broken; |
b3ae4755 | 401 | } |
97f10027 | 402 | |
b3ae4755 | 403 | wakeup: |
2a9e5c22 AV |
404 | /* |
405 | * tell the vfs op waiting on a waitqueue | |
406 | * that this op is done | |
407 | */ | |
408 | spin_lock(&op->lock); | |
409 | if (unlikely(op_state_given_up(op))) { | |
410 | spin_unlock(&op->lock); | |
411 | goto out; | |
412 | } | |
413 | set_op_state_serviced(op); | |
414 | spin_unlock(&op->lock); | |
97f10027 | 415 | |
b3ae4755 MM |
416 | /* |
417 | * If this operation is an I/O operation we need to wait | |
418 | * for all data to be copied before we can return to avoid | |
419 | * buffer corruption and races that can pull the buffers | |
420 | * out from under us. | |
421 | * | |
422 | * Essentially we're synchronizing with other parts of the | |
423 | * vfs implicitly by not allowing the user space | |
424 | * application reading/writing this device to return until | |
425 | * the buffers are done being used. | |
426 | */ | |
427 | if (op->downcall.type == ORANGEFS_VFS_OP_FILE_IO) { | |
2a9e5c22 AV |
428 | long n = wait_for_completion_interruptible_timeout(&op->done, |
429 | op_timeout_secs * HZ); | |
430 | if (unlikely(n < 0)) { | |
431 | gossip_debug(GOSSIP_DEV_DEBUG, | |
432 | "%s: signal on I/O wait, aborting\n", | |
433 | __func__); | |
434 | } else if (unlikely(n == 0)) { | |
435 | gossip_debug(GOSSIP_DEV_DEBUG, | |
436 | "%s: timed out.\n", | |
437 | __func__); | |
5db11c21 | 438 | } |
5db11c21 | 439 | } |
b3ae4755 | 440 | out: |
ed42fe05 | 441 | op_release(op); |
b3ae4755 | 442 | return ret; |
ed42fe05 AV |
443 | |
444 | Broken: | |
445 | spin_lock(&op->lock); | |
446 | if (!op_state_given_up(op)) { | |
447 | op->downcall.status = ret; | |
448 | set_op_state_serviced(op); | |
449 | } | |
450 | spin_unlock(&op->lock); | |
451 | goto out; | |
5db11c21 MM |
452 | } |
453 | ||
454 | /* Returns whether any FS are still pending remounted */ | |
455 | static int mark_all_pending_mounts(void) | |
456 | { | |
457 | int unmounted = 1; | |
8bb8aefd | 458 | struct orangefs_sb_info_s *orangefs_sb = NULL; |
5db11c21 | 459 | |
8bb8aefd YL |
460 | spin_lock(&orangefs_superblocks_lock); |
461 | list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) { | |
5db11c21 | 462 | /* All of these file system require a remount */ |
8bb8aefd | 463 | orangefs_sb->mount_pending = 1; |
5db11c21 MM |
464 | unmounted = 0; |
465 | } | |
8bb8aefd | 466 | spin_unlock(&orangefs_superblocks_lock); |
5db11c21 MM |
467 | return unmounted; |
468 | } | |
469 | ||
470 | /* | |
471 | * Determine if a given file system needs to be remounted or not | |
472 | * Returns -1 on error | |
473 | * 0 if already mounted | |
474 | * 1 if needs remount | |
475 | */ | |
476 | int fs_mount_pending(__s32 fsid) | |
477 | { | |
478 | int mount_pending = -1; | |
8bb8aefd | 479 | struct orangefs_sb_info_s *orangefs_sb = NULL; |
5db11c21 | 480 | |
8bb8aefd YL |
481 | spin_lock(&orangefs_superblocks_lock); |
482 | list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) { | |
483 | if (orangefs_sb->fs_id == fsid) { | |
484 | mount_pending = orangefs_sb->mount_pending; | |
5db11c21 MM |
485 | break; |
486 | } | |
487 | } | |
8bb8aefd | 488 | spin_unlock(&orangefs_superblocks_lock); |
5db11c21 MM |
489 | return mount_pending; |
490 | } | |
491 | ||
492 | /* | |
493 | * NOTE: gets called when the last reference to this device is dropped. | |
494 | * Using the open_access_count variable, we enforce a reference count | |
495 | * on this file so that it can be opened by only one process at a time. | |
496 | * the devreq_mutex is used to make sure all i/o has completed | |
8bb8aefd | 497 | * before we call orangefs_bufmap_finalize, and similar such tricky |
5db11c21 MM |
498 | * situations |
499 | */ | |
8bb8aefd | 500 | static int orangefs_devreq_release(struct inode *inode, struct file *file) |
5db11c21 MM |
501 | { |
502 | int unmounted = 0; | |
503 | ||
504 | gossip_debug(GOSSIP_DEV_DEBUG, | |
505 | "%s:pvfs2-client-core: exiting, closing device\n", | |
506 | __func__); | |
507 | ||
508 | mutex_lock(&devreq_mutex); | |
7d221485 | 509 | if (orangefs_get_bufmap_init()) |
90d26aa8 | 510 | orangefs_bufmap_finalize(); |
5db11c21 | 511 | |
fee25ce1 | 512 | open_access_count = -1; |
5db11c21 MM |
513 | |
514 | unmounted = mark_all_pending_mounts(); | |
8bb8aefd | 515 | gossip_debug(GOSSIP_DEV_DEBUG, "ORANGEFS Device Close: Filesystem(s) %s\n", |
5db11c21 | 516 | (unmounted ? "UNMOUNTED" : "MOUNTED")); |
5db11c21 MM |
517 | |
518 | /* | |
519 | * Walk through the list of ops in the request list, mark them | |
520 | * as purged and wake them up. | |
521 | */ | |
522 | purge_waiting_ops(); | |
523 | /* | |
524 | * Walk through the hash table of in progress operations; mark | |
525 | * them as purged and wake them up | |
526 | */ | |
527 | purge_inprogress_ops(); | |
528 | gossip_debug(GOSSIP_DEV_DEBUG, | |
529 | "pvfs2-client-core: device close complete\n"); | |
fee25ce1 AV |
530 | open_access_count = 0; |
531 | mutex_unlock(&devreq_mutex); | |
5db11c21 MM |
532 | return 0; |
533 | } | |
534 | ||
535 | int is_daemon_in_service(void) | |
536 | { | |
537 | int in_service; | |
538 | ||
539 | /* | |
540 | * What this function does is checks if client-core is alive | |
541 | * based on the access count we maintain on the device. | |
542 | */ | |
543 | mutex_lock(&devreq_mutex); | |
544 | in_service = open_access_count == 1 ? 0 : -EIO; | |
545 | mutex_unlock(&devreq_mutex); | |
546 | return in_service; | |
547 | } | |
548 | ||
549 | static inline long check_ioctl_command(unsigned int command) | |
550 | { | |
551 | /* Check for valid ioctl codes */ | |
8bb8aefd | 552 | if (_IOC_TYPE(command) != ORANGEFS_DEV_MAGIC) { |
5db11c21 MM |
553 | gossip_err("device ioctl magic numbers don't match! Did you rebuild pvfs2-client-core/libpvfs2? [cmd %x, magic %x != %x]\n", |
554 | command, | |
555 | _IOC_TYPE(command), | |
8bb8aefd | 556 | ORANGEFS_DEV_MAGIC); |
5db11c21 MM |
557 | return -EINVAL; |
558 | } | |
559 | /* and valid ioctl commands */ | |
8bb8aefd | 560 | if (_IOC_NR(command) >= ORANGEFS_DEV_MAXNR || _IOC_NR(command) <= 0) { |
5db11c21 | 561 | gossip_err("Invalid ioctl command number [%d >= %d]\n", |
8bb8aefd | 562 | _IOC_NR(command), ORANGEFS_DEV_MAXNR); |
5db11c21 MM |
563 | return -ENOIOCTLCMD; |
564 | } | |
565 | return 0; | |
566 | } | |
567 | ||
568 | static long dispatch_ioctl_command(unsigned int command, unsigned long arg) | |
569 | { | |
8bb8aefd | 570 | static __s32 magic = ORANGEFS_DEVREQ_MAGIC; |
a762ae6d MB |
571 | static __s32 max_up_size = MAX_DEV_REQ_UPSIZE; |
572 | static __s32 max_down_size = MAX_DEV_REQ_DOWNSIZE; | |
8bb8aefd | 573 | struct ORANGEFS_dev_map_desc user_desc; |
5db11c21 MM |
574 | int ret = 0; |
575 | struct dev_mask_info_s mask_info = { 0 }; | |
576 | struct dev_mask2_info_s mask2_info = { 0, 0 }; | |
577 | int upstream_kmod = 1; | |
578 | struct list_head *tmp = NULL; | |
8bb8aefd | 579 | struct orangefs_sb_info_s *orangefs_sb = NULL; |
5db11c21 MM |
580 | |
581 | /* mtmoore: add locking here */ | |
582 | ||
583 | switch (command) { | |
8bb8aefd | 584 | case ORANGEFS_DEV_GET_MAGIC: |
5db11c21 MM |
585 | return ((put_user(magic, (__s32 __user *) arg) == -EFAULT) ? |
586 | -EIO : | |
587 | 0); | |
8bb8aefd | 588 | case ORANGEFS_DEV_GET_MAX_UPSIZE: |
5db11c21 MM |
589 | return ((put_user(max_up_size, |
590 | (__s32 __user *) arg) == -EFAULT) ? | |
591 | -EIO : | |
592 | 0); | |
8bb8aefd | 593 | case ORANGEFS_DEV_GET_MAX_DOWNSIZE: |
5db11c21 MM |
594 | return ((put_user(max_down_size, |
595 | (__s32 __user *) arg) == -EFAULT) ? | |
596 | -EIO : | |
597 | 0); | |
8bb8aefd | 598 | case ORANGEFS_DEV_MAP: |
5db11c21 | 599 | ret = copy_from_user(&user_desc, |
8bb8aefd | 600 | (struct ORANGEFS_dev_map_desc __user *) |
5db11c21 | 601 | arg, |
8bb8aefd | 602 | sizeof(struct ORANGEFS_dev_map_desc)); |
7d221485 | 603 | if (orangefs_get_bufmap_init()) { |
90d26aa8 MB |
604 | return -EINVAL; |
605 | } else { | |
606 | return ret ? | |
607 | -EIO : | |
608 | orangefs_bufmap_initialize(&user_desc); | |
609 | } | |
8bb8aefd | 610 | case ORANGEFS_DEV_REMOUNT_ALL: |
5db11c21 | 611 | gossip_debug(GOSSIP_DEV_DEBUG, |
97f10027 MM |
612 | "%s: got ORANGEFS_DEV_REMOUNT_ALL\n", |
613 | __func__); | |
5db11c21 MM |
614 | |
615 | /* | |
8bb8aefd | 616 | * remount all mounted orangefs volumes to regain the lost |
5db11c21 MM |
617 | * dynamic mount tables (if any) -- NOTE: this is done |
618 | * without keeping the superblock list locked due to the | |
619 | * upcall/downcall waiting. also, the request semaphore is | |
620 | * used to ensure that no operations will be serviced until | |
621 | * all of the remounts are serviced (to avoid ops between | |
622 | * mounts to fail) | |
623 | */ | |
624 | ret = mutex_lock_interruptible(&request_mutex); | |
625 | if (ret < 0) | |
626 | return ret; | |
627 | gossip_debug(GOSSIP_DEV_DEBUG, | |
97f10027 MM |
628 | "%s: priority remount in progress\n", |
629 | __func__); | |
8bb8aefd YL |
630 | list_for_each(tmp, &orangefs_superblocks) { |
631 | orangefs_sb = | |
97f10027 MM |
632 | list_entry(tmp, |
633 | struct orangefs_sb_info_s, | |
634 | list); | |
8bb8aefd | 635 | if (orangefs_sb && (orangefs_sb->sb)) { |
5db11c21 | 636 | gossip_debug(GOSSIP_DEV_DEBUG, |
97f10027 MM |
637 | "%s: Remounting SB %p\n", |
638 | __func__, | |
8bb8aefd | 639 | orangefs_sb); |
5db11c21 | 640 | |
8bb8aefd | 641 | ret = orangefs_remount(orangefs_sb->sb); |
5db11c21 MM |
642 | if (ret) { |
643 | gossip_debug(GOSSIP_DEV_DEBUG, | |
644 | "SB %p remount failed\n", | |
8bb8aefd | 645 | orangefs_sb); |
97f10027 | 646 | break; |
5db11c21 MM |
647 | } |
648 | } | |
649 | } | |
650 | gossip_debug(GOSSIP_DEV_DEBUG, | |
97f10027 MM |
651 | "%s: priority remount complete\n", |
652 | __func__); | |
5db11c21 MM |
653 | mutex_unlock(&request_mutex); |
654 | return ret; | |
655 | ||
8bb8aefd | 656 | case ORANGEFS_DEV_UPSTREAM: |
5db11c21 MM |
657 | ret = copy_to_user((void __user *)arg, |
658 | &upstream_kmod, | |
659 | sizeof(upstream_kmod)); | |
660 | ||
661 | if (ret != 0) | |
662 | return -EIO; | |
663 | else | |
664 | return ret; | |
665 | ||
8bb8aefd | 666 | case ORANGEFS_DEV_CLIENT_MASK: |
5db11c21 MM |
667 | ret = copy_from_user(&mask2_info, |
668 | (void __user *)arg, | |
669 | sizeof(struct dev_mask2_info_s)); | |
670 | ||
671 | if (ret != 0) | |
672 | return -EIO; | |
673 | ||
674 | client_debug_mask.mask1 = mask2_info.mask1_value; | |
675 | client_debug_mask.mask2 = mask2_info.mask2_value; | |
676 | ||
677 | pr_info("%s: client debug mask has been been received " | |
678 | ":%llx: :%llx:\n", | |
679 | __func__, | |
680 | (unsigned long long)client_debug_mask.mask1, | |
681 | (unsigned long long)client_debug_mask.mask2); | |
682 | ||
683 | return ret; | |
684 | ||
8bb8aefd | 685 | case ORANGEFS_DEV_CLIENT_STRING: |
5db11c21 MM |
686 | ret = copy_from_user(&client_debug_array_string, |
687 | (void __user *)arg, | |
8bb8aefd | 688 | ORANGEFS_MAX_DEBUG_STRING_LEN); |
5db11c21 | 689 | if (ret != 0) { |
97f10027 | 690 | pr_info("%s: CLIENT_STRING: copy_from_user failed\n", |
5db11c21 MM |
691 | __func__); |
692 | return -EIO; | |
693 | } | |
694 | ||
97f10027 | 695 | pr_info("%s: client debug array string has been received.\n", |
5db11c21 MM |
696 | __func__); |
697 | ||
698 | if (!help_string_initialized) { | |
699 | ||
700 | /* Free the "we don't know yet" default string... */ | |
701 | kfree(debug_help_string); | |
702 | ||
703 | /* build a proper debug help string */ | |
704 | if (orangefs_prepare_debugfs_help_string(0)) { | |
97f10027 | 705 | gossip_err("%s: no debug help string \n", |
5db11c21 MM |
706 | __func__); |
707 | return -EIO; | |
708 | } | |
709 | ||
710 | /* Replace the boilerplate boot-time debug-help file. */ | |
711 | debugfs_remove(help_file_dentry); | |
712 | ||
713 | help_file_dentry = | |
714 | debugfs_create_file( | |
715 | ORANGEFS_KMOD_DEBUG_HELP_FILE, | |
716 | 0444, | |
717 | debug_dir, | |
718 | debug_help_string, | |
719 | &debug_help_fops); | |
720 | ||
721 | if (!help_file_dentry) { | |
722 | gossip_err("%s: debugfs_create_file failed for" | |
723 | " :%s:!\n", | |
724 | __func__, | |
725 | ORANGEFS_KMOD_DEBUG_HELP_FILE); | |
726 | return -EIO; | |
727 | } | |
728 | } | |
729 | ||
730 | debug_mask_to_string(&client_debug_mask, 1); | |
731 | ||
732 | debugfs_remove(client_debug_dentry); | |
733 | ||
8bb8aefd | 734 | orangefs_client_debug_init(); |
5db11c21 MM |
735 | |
736 | help_string_initialized++; | |
737 | ||
738 | return ret; | |
739 | ||
8bb8aefd | 740 | case ORANGEFS_DEV_DEBUG: |
5db11c21 MM |
741 | ret = copy_from_user(&mask_info, |
742 | (void __user *)arg, | |
743 | sizeof(mask_info)); | |
744 | ||
745 | if (ret != 0) | |
746 | return -EIO; | |
747 | ||
748 | if (mask_info.mask_type == KERNEL_MASK) { | |
749 | if ((mask_info.mask_value == 0) | |
750 | && (kernel_mask_set_mod_init)) { | |
751 | /* | |
752 | * the kernel debug mask was set when the | |
753 | * kernel module was loaded; don't override | |
754 | * it if the client-core was started without | |
8bb8aefd | 755 | * a value for ORANGEFS_KMODMASK. |
5db11c21 MM |
756 | */ |
757 | return 0; | |
758 | } | |
759 | debug_mask_to_string(&mask_info.mask_value, | |
760 | mask_info.mask_type); | |
761 | gossip_debug_mask = mask_info.mask_value; | |
97f10027 | 762 | pr_info("%s: kernel debug mask has been modified to " |
5db11c21 | 763 | ":%s: :%llx:\n", |
97f10027 | 764 | __func__, |
5db11c21 MM |
765 | kernel_debug_string, |
766 | (unsigned long long)gossip_debug_mask); | |
767 | } else if (mask_info.mask_type == CLIENT_MASK) { | |
768 | debug_mask_to_string(&mask_info.mask_value, | |
769 | mask_info.mask_type); | |
97f10027 | 770 | pr_info("%s: client debug mask has been modified to" |
5db11c21 | 771 | ":%s: :%llx:\n", |
97f10027 | 772 | __func__, |
5db11c21 MM |
773 | client_debug_string, |
774 | llu(mask_info.mask_value)); | |
775 | } else { | |
776 | gossip_lerr("Invalid mask type....\n"); | |
777 | return -EINVAL; | |
778 | } | |
779 | ||
780 | return ret; | |
781 | ||
782 | default: | |
783 | return -ENOIOCTLCMD; | |
784 | } | |
785 | return -ENOIOCTLCMD; | |
786 | } | |
787 | ||
8bb8aefd | 788 | static long orangefs_devreq_ioctl(struct file *file, |
5db11c21 MM |
789 | unsigned int command, unsigned long arg) |
790 | { | |
791 | long ret; | |
792 | ||
793 | /* Check for properly constructed commands */ | |
794 | ret = check_ioctl_command(command); | |
795 | if (ret < 0) | |
796 | return (int)ret; | |
797 | ||
798 | return (int)dispatch_ioctl_command(command, arg); | |
799 | } | |
800 | ||
801 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
802 | ||
8bb8aefd YL |
803 | /* Compat structure for the ORANGEFS_DEV_MAP ioctl */ |
804 | struct ORANGEFS_dev_map_desc32 { | |
5db11c21 MM |
805 | compat_uptr_t ptr; |
806 | __s32 total_size; | |
807 | __s32 size; | |
808 | __s32 count; | |
809 | }; | |
810 | ||
811 | static unsigned long translate_dev_map26(unsigned long args, long *error) | |
812 | { | |
8bb8aefd | 813 | struct ORANGEFS_dev_map_desc32 __user *p32 = (void __user *)args; |
5db11c21 MM |
814 | /* |
815 | * Depending on the architecture, allocate some space on the | |
816 | * user-call-stack based on our expected layout. | |
817 | */ | |
8bb8aefd | 818 | struct ORANGEFS_dev_map_desc __user *p = |
5db11c21 | 819 | compat_alloc_user_space(sizeof(*p)); |
84d02150 | 820 | compat_uptr_t addr; |
5db11c21 MM |
821 | |
822 | *error = 0; | |
823 | /* get the ptr from the 32 bit user-space */ | |
824 | if (get_user(addr, &p32->ptr)) | |
825 | goto err; | |
826 | /* try to put that into a 64-bit layout */ | |
827 | if (put_user(compat_ptr(addr), &p->ptr)) | |
828 | goto err; | |
829 | /* copy the remaining fields */ | |
830 | if (copy_in_user(&p->total_size, &p32->total_size, sizeof(__s32))) | |
831 | goto err; | |
832 | if (copy_in_user(&p->size, &p32->size, sizeof(__s32))) | |
833 | goto err; | |
834 | if (copy_in_user(&p->count, &p32->count, sizeof(__s32))) | |
835 | goto err; | |
836 | return (unsigned long)p; | |
837 | err: | |
838 | *error = -EFAULT; | |
839 | return 0; | |
840 | } | |
841 | ||
842 | /* | |
843 | * 32 bit user-space apps' ioctl handlers when kernel modules | |
844 | * is compiled as a 64 bit one | |
845 | */ | |
8bb8aefd | 846 | static long orangefs_devreq_compat_ioctl(struct file *filp, unsigned int cmd, |
5db11c21 MM |
847 | unsigned long args) |
848 | { | |
849 | long ret; | |
850 | unsigned long arg = args; | |
851 | ||
852 | /* Check for properly constructed commands */ | |
853 | ret = check_ioctl_command(cmd); | |
854 | if (ret < 0) | |
855 | return ret; | |
8bb8aefd | 856 | if (cmd == ORANGEFS_DEV_MAP) { |
5db11c21 MM |
857 | /* |
858 | * convert the arguments to what we expect internally | |
859 | * in kernel space | |
860 | */ | |
861 | arg = translate_dev_map26(args, &ret); | |
862 | if (ret < 0) { | |
863 | gossip_err("Could not translate dev map\n"); | |
864 | return ret; | |
865 | } | |
866 | } | |
867 | /* no other ioctl requires translation */ | |
868 | return dispatch_ioctl_command(cmd, arg); | |
869 | } | |
870 | ||
2c590d5f MM |
871 | #endif /* CONFIG_COMPAT is in .config */ |
872 | ||
5db11c21 | 873 | /* the assigned character device major number */ |
8bb8aefd | 874 | static int orangefs_dev_major; |
5db11c21 MM |
875 | |
876 | /* | |
8bb8aefd | 877 | * Initialize orangefs device specific state: |
5db11c21 MM |
878 | * Must be called at module load time only |
879 | */ | |
8bb8aefd | 880 | int orangefs_dev_init(void) |
5db11c21 | 881 | { |
8bb8aefd YL |
882 | /* register orangefs-req device */ |
883 | orangefs_dev_major = register_chrdev(0, | |
884 | ORANGEFS_REQDEVICE_NAME, | |
885 | &orangefs_devreq_file_operations); | |
886 | if (orangefs_dev_major < 0) { | |
5db11c21 MM |
887 | gossip_debug(GOSSIP_DEV_DEBUG, |
888 | "Failed to register /dev/%s (error %d)\n", | |
8bb8aefd | 889 | ORANGEFS_REQDEVICE_NAME, orangefs_dev_major); |
8bb8aefd | 890 | return orangefs_dev_major; |
5db11c21 MM |
891 | } |
892 | ||
893 | gossip_debug(GOSSIP_DEV_DEBUG, | |
894 | "*** /dev/%s character device registered ***\n", | |
8bb8aefd | 895 | ORANGEFS_REQDEVICE_NAME); |
5db11c21 | 896 | gossip_debug(GOSSIP_DEV_DEBUG, "'mknod /dev/%s c %d 0'.\n", |
8bb8aefd | 897 | ORANGEFS_REQDEVICE_NAME, orangefs_dev_major); |
5db11c21 MM |
898 | return 0; |
899 | } | |
900 | ||
8bb8aefd | 901 | void orangefs_dev_cleanup(void) |
5db11c21 | 902 | { |
8bb8aefd | 903 | unregister_chrdev(orangefs_dev_major, ORANGEFS_REQDEVICE_NAME); |
5db11c21 MM |
904 | gossip_debug(GOSSIP_DEV_DEBUG, |
905 | "*** /dev/%s character device unregistered ***\n", | |
8bb8aefd | 906 | ORANGEFS_REQDEVICE_NAME); |
5db11c21 MM |
907 | } |
908 | ||
8bb8aefd | 909 | static unsigned int orangefs_devreq_poll(struct file *file, |
5db11c21 MM |
910 | struct poll_table_struct *poll_table) |
911 | { | |
912 | int poll_revent_mask = 0; | |
913 | ||
83595db0 | 914 | poll_wait(file, &orangefs_request_list_waitq, poll_table); |
5db11c21 | 915 | |
83595db0 AV |
916 | if (!list_empty(&orangefs_request_list)) |
917 | poll_revent_mask |= POLL_IN; | |
5db11c21 MM |
918 | return poll_revent_mask; |
919 | } | |
920 | ||
8bb8aefd | 921 | const struct file_operations orangefs_devreq_file_operations = { |
5db11c21 | 922 | .owner = THIS_MODULE, |
8bb8aefd YL |
923 | .read = orangefs_devreq_read, |
924 | .write_iter = orangefs_devreq_write_iter, | |
925 | .open = orangefs_devreq_open, | |
926 | .release = orangefs_devreq_release, | |
927 | .unlocked_ioctl = orangefs_devreq_ioctl, | |
5db11c21 MM |
928 | |
929 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
8bb8aefd | 930 | .compat_ioctl = orangefs_devreq_compat_ioctl, |
5db11c21 | 931 | #endif |
8bb8aefd | 932 | .poll = orangefs_devreq_poll |
5db11c21 | 933 | }; |