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" | |
11 | #include "pvfs2-kernel.h" | |
12 | #include "pvfs2-dev-proto.h" | |
13 | #include "pvfs2-bufmap.h" | |
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");\ | |
25 | gossip_err("PVFS2 Device Error: You cannot open the device file "); \ | |
26 | gossip_err("\n/dev/%s more than once. Please make sure that\nthere " \ | |
27 | "are no ", PVFS2_REQDEVICE_NAME); \ | |
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", \ | |
32 | PVFS2_REQDEVICE_NAME); \ | |
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 | ||
42 | static void pvfs2_devreq_add_op(struct pvfs2_kernel_op_s *op) | |
43 | { | |
44 | int index = hash_func(op->tag, hash_table_size); | |
45 | ||
46 | spin_lock(&htable_ops_in_progress_lock); | |
47 | list_add_tail(&op->list, &htable_ops_in_progress[index]); | |
48 | spin_unlock(&htable_ops_in_progress_lock); | |
49 | } | |
50 | ||
51 | static struct pvfs2_kernel_op_s *pvfs2_devreq_remove_op(__u64 tag) | |
52 | { | |
53 | struct pvfs2_kernel_op_s *op, *next; | |
54 | int index; | |
55 | ||
56 | index = hash_func(tag, hash_table_size); | |
57 | ||
58 | spin_lock(&htable_ops_in_progress_lock); | |
59 | list_for_each_entry_safe(op, | |
60 | next, | |
61 | &htable_ops_in_progress[index], | |
62 | list) { | |
63 | if (op->tag == tag) { | |
64 | list_del(&op->list); | |
65 | spin_unlock(&htable_ops_in_progress_lock); | |
66 | return op; | |
67 | } | |
68 | } | |
69 | ||
70 | spin_unlock(&htable_ops_in_progress_lock); | |
71 | return NULL; | |
72 | } | |
73 | ||
74 | static int pvfs2_devreq_open(struct inode *inode, struct file *file) | |
75 | { | |
76 | int ret = -EINVAL; | |
77 | ||
78 | if (!(file->f_flags & O_NONBLOCK)) { | |
79 | gossip_err("pvfs2: device cannot be opened in blocking mode\n"); | |
80 | goto out; | |
81 | } | |
82 | ret = -EACCES; | |
83 | gossip_debug(GOSSIP_DEV_DEBUG, "pvfs2-client-core: opening device\n"); | |
84 | mutex_lock(&devreq_mutex); | |
85 | ||
86 | if (open_access_count == 0) { | |
87 | ret = generic_file_open(inode, file); | |
88 | if (ret == 0) | |
89 | open_access_count++; | |
90 | } else { | |
91 | DUMP_DEVICE_ERROR(); | |
92 | } | |
93 | mutex_unlock(&devreq_mutex); | |
94 | ||
95 | out: | |
96 | ||
97 | gossip_debug(GOSSIP_DEV_DEBUG, | |
98 | "pvfs2-client-core: open device complete (ret = %d)\n", | |
99 | ret); | |
100 | return ret; | |
101 | } | |
102 | ||
103 | static ssize_t pvfs2_devreq_read(struct file *file, | |
104 | char __user *buf, | |
105 | size_t count, loff_t *offset) | |
106 | { | |
107 | int ret = 0; | |
108 | ssize_t len = 0; | |
109 | struct pvfs2_kernel_op_s *cur_op = NULL; | |
110 | static __s32 magic = PVFS2_DEVREQ_MAGIC; | |
111 | __s32 proto_ver = PVFS_KERNEL_PROTO_VERSION; | |
112 | ||
113 | if (!(file->f_flags & O_NONBLOCK)) { | |
114 | /* We do not support blocking reads/opens any more */ | |
115 | gossip_err("pvfs2: blocking reads are not supported! (pvfs2-client-core bug)\n"); | |
116 | return -EINVAL; | |
117 | } else { | |
118 | struct pvfs2_kernel_op_s *op = NULL, *temp = NULL; | |
119 | /* get next op (if any) from top of list */ | |
120 | spin_lock(&pvfs2_request_list_lock); | |
121 | list_for_each_entry_safe(op, temp, &pvfs2_request_list, list) { | |
122 | __s32 fsid = fsid_of_op(op); | |
123 | /* | |
124 | * Check if this op's fsid is known and needs | |
125 | * remounting | |
126 | */ | |
127 | if (fsid != PVFS_FS_ID_NULL && | |
128 | fs_mount_pending(fsid) == 1) { | |
129 | gossip_debug(GOSSIP_DEV_DEBUG, | |
130 | "Skipping op tag %llu %s\n", | |
131 | llu(op->tag), | |
132 | get_opname_string(op)); | |
133 | continue; | |
134 | } else { | |
135 | /* | |
136 | * op does not belong to any particular fsid | |
137 | * or already mounted.. let it through | |
138 | */ | |
139 | cur_op = op; | |
140 | spin_lock(&cur_op->lock); | |
141 | list_del(&cur_op->list); | |
142 | cur_op->op_linger_tmp--; | |
143 | /* | |
144 | * if there is a trailer, re-add it to | |
145 | * the request list. | |
146 | */ | |
147 | if (cur_op->op_linger == 2 && | |
148 | cur_op->op_linger_tmp == 1) { | |
149 | if (cur_op->upcall.trailer_size <= 0 || | |
150 | cur_op->upcall.trailer_buf == NULL) | |
151 | gossip_err("BUG:trailer_size is %ld and trailer buf is %p\n", (long)cur_op->upcall.trailer_size, cur_op->upcall.trailer_buf); | |
152 | /* re-add it to the head of the list */ | |
153 | list_add(&cur_op->list, | |
154 | &pvfs2_request_list); | |
155 | } | |
156 | spin_unlock(&cur_op->lock); | |
157 | break; | |
158 | } | |
159 | } | |
160 | spin_unlock(&pvfs2_request_list_lock); | |
161 | } | |
162 | ||
163 | if (cur_op) { | |
164 | spin_lock(&cur_op->lock); | |
165 | ||
166 | gossip_debug(GOSSIP_DEV_DEBUG, | |
167 | "client-core: reading op tag %llu %s\n", | |
168 | llu(cur_op->tag), get_opname_string(cur_op)); | |
169 | if (op_state_in_progress(cur_op) || op_state_serviced(cur_op)) { | |
170 | if (cur_op->op_linger == 1) | |
171 | gossip_err("WARNING: Current op already queued...skipping\n"); | |
172 | } else if (cur_op->op_linger == 1 || | |
173 | (cur_op->op_linger == 2 && | |
174 | cur_op->op_linger_tmp == 0)) { | |
175 | /* | |
176 | * atomically move the operation to the | |
177 | * htable_ops_in_progress | |
178 | */ | |
179 | set_op_state_inprogress(cur_op); | |
180 | pvfs2_devreq_add_op(cur_op); | |
181 | } | |
182 | ||
183 | spin_unlock(&cur_op->lock); | |
184 | ||
185 | /* 2 cases | |
186 | * a) OPs with no trailers | |
187 | * b) OPs with trailers, Stage 1 | |
188 | * Either way push the upcall out | |
189 | */ | |
190 | if (cur_op->op_linger == 1 || | |
191 | (cur_op->op_linger == 2 && cur_op->op_linger_tmp == 1)) { | |
192 | len = MAX_ALIGNED_DEV_REQ_UPSIZE; | |
193 | if ((size_t) len <= count) { | |
194 | ret = copy_to_user(buf, | |
195 | &proto_ver, | |
196 | sizeof(__s32)); | |
197 | if (ret == 0) { | |
198 | ret = copy_to_user(buf + sizeof(__s32), | |
199 | &magic, | |
200 | sizeof(__s32)); | |
201 | if (ret == 0) { | |
202 | ret = copy_to_user(buf+2 * sizeof(__s32), | |
203 | &cur_op->tag, | |
204 | sizeof(__u64)); | |
205 | if (ret == 0) { | |
206 | ret = copy_to_user( | |
207 | buf + | |
208 | 2 * | |
209 | sizeof(__s32) + | |
210 | sizeof(__u64), | |
211 | &cur_op->upcall, | |
212 | sizeof(struct pvfs2_upcall_s)); | |
213 | } | |
214 | } | |
215 | } | |
216 | ||
217 | if (ret) { | |
218 | gossip_err("Failed to copy data to user space\n"); | |
219 | len = -EFAULT; | |
220 | } | |
221 | } else { | |
222 | gossip_err | |
223 | ("Failed to copy data to user space\n"); | |
224 | len = -EIO; | |
225 | } | |
226 | } | |
227 | /* Stage 2: Push the trailer out */ | |
228 | else if (cur_op->op_linger == 2 && cur_op->op_linger_tmp == 0) { | |
229 | len = cur_op->upcall.trailer_size; | |
230 | if ((size_t) len <= count) { | |
231 | ret = copy_to_user(buf, | |
232 | cur_op->upcall.trailer_buf, | |
233 | len); | |
234 | if (ret) { | |
235 | gossip_err("Failed to copy trailer to user space\n"); | |
236 | len = -EFAULT; | |
237 | } | |
238 | } else { | |
239 | gossip_err("Read buffer for trailer is too small (%ld as opposed to %ld)\n", | |
240 | (long)count, | |
241 | (long)len); | |
242 | len = -EIO; | |
243 | } | |
244 | } else { | |
245 | gossip_err("cur_op: %p (op_linger %d), (op_linger_tmp %d), erroneous request list?\n", | |
246 | cur_op, | |
247 | cur_op->op_linger, | |
248 | cur_op->op_linger_tmp); | |
249 | len = 0; | |
250 | } | |
251 | } else if (file->f_flags & O_NONBLOCK) { | |
252 | /* | |
253 | * if in non-blocking mode, return EAGAIN since no requests are | |
254 | * ready yet | |
255 | */ | |
256 | len = -EAGAIN; | |
257 | } | |
258 | return len; | |
259 | } | |
260 | ||
261 | /* Function for writev() callers into the device */ | |
262 | static ssize_t pvfs2_devreq_writev(struct file *file, | |
263 | const struct iovec *iov, | |
264 | size_t count, | |
265 | loff_t *offset) | |
266 | { | |
267 | struct pvfs2_kernel_op_s *op = NULL; | |
268 | void *buffer = NULL; | |
269 | void *ptr = NULL; | |
270 | unsigned long i = 0; | |
271 | static int max_downsize = MAX_ALIGNED_DEV_REQ_DOWNSIZE; | |
272 | int ret = 0, num_remaining = max_downsize; | |
273 | int notrailer_count = 4; /* num elements in iovec without trailer */ | |
274 | int payload_size = 0; | |
275 | __s32 magic = 0; | |
276 | __s32 proto_ver = 0; | |
277 | __u64 tag = 0; | |
278 | ssize_t total_returned_size = 0; | |
279 | ||
280 | /* Either there is a trailer or there isn't */ | |
281 | if (count != notrailer_count && count != (notrailer_count + 1)) { | |
2c590d5f | 282 | gossip_err("Error: Number of iov vectors is (%zu) and notrailer count is %d\n", |
5db11c21 MM |
283 | count, |
284 | notrailer_count); | |
285 | return -EPROTO; | |
286 | } | |
287 | buffer = dev_req_alloc(); | |
288 | if (!buffer) | |
289 | return -ENOMEM; | |
290 | ptr = buffer; | |
291 | ||
292 | for (i = 0; i < notrailer_count; i++) { | |
293 | if (iov[i].iov_len > num_remaining) { | |
294 | gossip_err | |
295 | ("writev error: Freeing buffer and returning\n"); | |
296 | dev_req_release(buffer); | |
297 | return -EMSGSIZE; | |
298 | } | |
299 | ret = copy_from_user(ptr, iov[i].iov_base, iov[i].iov_len); | |
300 | if (ret) { | |
301 | gossip_err("Failed to copy data from user space\n"); | |
302 | dev_req_release(buffer); | |
303 | return -EIO; | |
304 | } | |
305 | num_remaining -= iov[i].iov_len; | |
306 | ptr += iov[i].iov_len; | |
307 | payload_size += iov[i].iov_len; | |
308 | } | |
309 | total_returned_size = payload_size; | |
310 | ||
311 | /* these elements are currently 8 byte aligned (8 bytes for (version + | |
312 | * magic) 8 bytes for tag). If you add another element, either | |
313 | * make it 8 bytes big, or use get_unaligned when asigning. | |
314 | */ | |
315 | ptr = buffer; | |
316 | proto_ver = *((__s32 *) ptr); | |
317 | ptr += sizeof(__s32); | |
318 | ||
319 | magic = *((__s32 *) ptr); | |
320 | ptr += sizeof(__s32); | |
321 | ||
322 | tag = *((__u64 *) ptr); | |
323 | ptr += sizeof(__u64); | |
324 | ||
325 | if (magic != PVFS2_DEVREQ_MAGIC) { | |
326 | gossip_err("Error: Device magic number does not match.\n"); | |
327 | dev_req_release(buffer); | |
328 | return -EPROTO; | |
329 | } | |
330 | ||
331 | /* | |
332 | * proto_ver = 20902 for 2.9.2 | |
333 | */ | |
334 | ||
335 | op = pvfs2_devreq_remove_op(tag); | |
336 | if (op) { | |
337 | /* Increase ref count! */ | |
338 | get_op(op); | |
339 | /* cut off magic and tag from payload size */ | |
340 | payload_size -= (2 * sizeof(__s32) + sizeof(__u64)); | |
341 | if (payload_size <= sizeof(struct pvfs2_downcall_s)) | |
342 | /* copy the passed in downcall into the op */ | |
343 | memcpy(&op->downcall, | |
344 | ptr, | |
345 | sizeof(struct pvfs2_downcall_s)); | |
346 | else | |
347 | gossip_debug(GOSSIP_DEV_DEBUG, | |
348 | "writev: Ignoring %d bytes\n", | |
349 | payload_size); | |
350 | ||
351 | /* Do not allocate needlessly if client-core forgets | |
352 | * to reset trailer size on op errors. | |
353 | */ | |
354 | if (op->downcall.status == 0 && op->downcall.trailer_size > 0) { | |
355 | gossip_debug(GOSSIP_DEV_DEBUG, | |
356 | "writev: trailer size %ld\n", | |
357 | (unsigned long)op->downcall.trailer_size); | |
358 | if (count != (notrailer_count + 1)) { | |
2c590d5f | 359 | gossip_err("Error: trailer size (%ld) is non-zero, no trailer elements though? (%zu)\n", (unsigned long)op->downcall.trailer_size, count); |
5db11c21 MM |
360 | dev_req_release(buffer); |
361 | put_op(op); | |
362 | return -EPROTO; | |
363 | } | |
364 | if (iov[notrailer_count].iov_len > | |
365 | op->downcall.trailer_size) { | |
366 | gossip_err("writev error: trailer size (%ld) != iov_len (%ld)\n", (unsigned long)op->downcall.trailer_size, (unsigned long)iov[notrailer_count].iov_len); | |
367 | dev_req_release(buffer); | |
368 | put_op(op); | |
369 | return -EMSGSIZE; | |
370 | } | |
371 | /* Allocate a buffer large enough to hold the | |
372 | * trailer bytes. | |
373 | */ | |
374 | op->downcall.trailer_buf = | |
375 | vmalloc(op->downcall.trailer_size); | |
376 | if (op->downcall.trailer_buf != NULL) { | |
377 | gossip_debug(GOSSIP_DEV_DEBUG, "vmalloc: %p\n", | |
378 | op->downcall.trailer_buf); | |
379 | ret = copy_from_user(op->downcall.trailer_buf, | |
380 | iov[notrailer_count]. | |
381 | iov_base, | |
382 | iov[notrailer_count]. | |
383 | iov_len); | |
384 | if (ret) { | |
385 | gossip_err("Failed to copy trailer data from user space\n"); | |
386 | dev_req_release(buffer); | |
387 | gossip_debug(GOSSIP_DEV_DEBUG, | |
388 | "vfree: %p\n", | |
389 | op->downcall.trailer_buf); | |
390 | vfree(op->downcall.trailer_buf); | |
391 | op->downcall.trailer_buf = NULL; | |
392 | put_op(op); | |
393 | return -EIO; | |
394 | } | |
395 | } else { | |
396 | /* Change downcall status */ | |
397 | op->downcall.status = -ENOMEM; | |
398 | gossip_err("writev: could not vmalloc for trailer!\n"); | |
399 | } | |
400 | } | |
401 | ||
402 | /* if this operation is an I/O operation and if it was | |
403 | * initiated on behalf of a *synchronous* VFS I/O operation, | |
404 | * only then we need to wait | |
405 | * for all data to be copied before we can return to avoid | |
406 | * buffer corruption and races that can pull the buffers | |
407 | * out from under us. | |
408 | * | |
409 | * Essentially we're synchronizing with other parts of the | |
410 | * vfs implicitly by not allowing the user space | |
411 | * application reading/writing this device to return until | |
412 | * the buffers are done being used. | |
413 | */ | |
414 | if ((op->upcall.type == PVFS2_VFS_OP_FILE_IO && | |
415 | op->upcall.req.io.async_vfs_io == PVFS_VFS_SYNC_IO) || | |
416 | op->upcall.type == PVFS2_VFS_OP_FILE_IOX) { | |
417 | int timed_out = 0; | |
418 | DECLARE_WAITQUEUE(wait_entry, current); | |
419 | ||
420 | /* tell the vfs op waiting on a waitqueue | |
421 | * that this op is done | |
422 | */ | |
423 | spin_lock(&op->lock); | |
424 | set_op_state_serviced(op); | |
425 | spin_unlock(&op->lock); | |
426 | ||
427 | add_wait_queue_exclusive(&op->io_completion_waitq, | |
428 | &wait_entry); | |
429 | wake_up_interruptible(&op->waitq); | |
430 | ||
431 | while (1) { | |
432 | set_current_state(TASK_INTERRUPTIBLE); | |
433 | ||
434 | spin_lock(&op->lock); | |
435 | if (op->io_completed) { | |
436 | spin_unlock(&op->lock); | |
437 | break; | |
438 | } | |
439 | spin_unlock(&op->lock); | |
440 | ||
441 | if (!signal_pending(current)) { | |
442 | int timeout = | |
443 | MSECS_TO_JIFFIES(1000 * | |
444 | op_timeout_secs); | |
445 | if (!schedule_timeout(timeout)) { | |
446 | gossip_debug(GOSSIP_DEV_DEBUG, "*** I/O wait time is up\n"); | |
447 | timed_out = 1; | |
448 | break; | |
449 | } | |
450 | continue; | |
451 | } | |
452 | ||
453 | gossip_debug(GOSSIP_DEV_DEBUG, "*** signal on I/O wait -- aborting\n"); | |
454 | break; | |
455 | } | |
456 | ||
457 | set_current_state(TASK_RUNNING); | |
458 | remove_wait_queue(&op->io_completion_waitq, | |
459 | &wait_entry); | |
460 | ||
461 | /* NOTE: for I/O operations we handle releasing the op | |
462 | * object except in the case of timeout. the reason we | |
463 | * can't free the op in timeout cases is that the op | |
464 | * service logic in the vfs retries operations using | |
465 | * the same op ptr, thus it can't be freed. | |
466 | */ | |
467 | if (!timed_out) | |
468 | op_release(op); | |
469 | } else { | |
470 | ||
471 | /* | |
472 | * tell the vfs op waiting on a waitqueue that | |
473 | * this op is done | |
474 | */ | |
475 | spin_lock(&op->lock); | |
476 | set_op_state_serviced(op); | |
477 | spin_unlock(&op->lock); | |
478 | /* | |
479 | for every other operation (i.e. non-I/O), we need to | |
480 | wake up the callers for downcall completion | |
481 | notification | |
482 | */ | |
483 | wake_up_interruptible(&op->waitq); | |
484 | } | |
485 | } else { | |
486 | /* ignore downcalls that we're not interested in */ | |
487 | gossip_debug(GOSSIP_DEV_DEBUG, | |
488 | "WARNING: No one's waiting for tag %llu\n", | |
489 | llu(tag)); | |
490 | } | |
491 | dev_req_release(buffer); | |
492 | ||
493 | return total_returned_size; | |
494 | } | |
495 | ||
496 | static ssize_t pvfs2_devreq_write_iter(struct kiocb *iocb, | |
497 | struct iov_iter *iter) | |
498 | { | |
499 | return pvfs2_devreq_writev(iocb->ki_filp, | |
500 | iter->iov, | |
501 | iter->nr_segs, | |
502 | &iocb->ki_pos); | |
503 | } | |
504 | ||
505 | /* Returns whether any FS are still pending remounted */ | |
506 | static int mark_all_pending_mounts(void) | |
507 | { | |
508 | int unmounted = 1; | |
509 | struct pvfs2_sb_info_s *pvfs2_sb = NULL; | |
510 | ||
511 | spin_lock(&pvfs2_superblocks_lock); | |
512 | list_for_each_entry(pvfs2_sb, &pvfs2_superblocks, list) { | |
513 | /* All of these file system require a remount */ | |
514 | pvfs2_sb->mount_pending = 1; | |
515 | unmounted = 0; | |
516 | } | |
517 | spin_unlock(&pvfs2_superblocks_lock); | |
518 | return unmounted; | |
519 | } | |
520 | ||
521 | /* | |
522 | * Determine if a given file system needs to be remounted or not | |
523 | * Returns -1 on error | |
524 | * 0 if already mounted | |
525 | * 1 if needs remount | |
526 | */ | |
527 | int fs_mount_pending(__s32 fsid) | |
528 | { | |
529 | int mount_pending = -1; | |
530 | struct pvfs2_sb_info_s *pvfs2_sb = NULL; | |
531 | ||
532 | spin_lock(&pvfs2_superblocks_lock); | |
533 | list_for_each_entry(pvfs2_sb, &pvfs2_superblocks, list) { | |
534 | if (pvfs2_sb->fs_id == fsid) { | |
535 | mount_pending = pvfs2_sb->mount_pending; | |
536 | break; | |
537 | } | |
538 | } | |
539 | spin_unlock(&pvfs2_superblocks_lock); | |
540 | return mount_pending; | |
541 | } | |
542 | ||
543 | /* | |
544 | * NOTE: gets called when the last reference to this device is dropped. | |
545 | * Using the open_access_count variable, we enforce a reference count | |
546 | * on this file so that it can be opened by only one process at a time. | |
547 | * the devreq_mutex is used to make sure all i/o has completed | |
548 | * before we call pvfs_bufmap_finalize, and similar such tricky | |
549 | * situations | |
550 | */ | |
551 | static int pvfs2_devreq_release(struct inode *inode, struct file *file) | |
552 | { | |
553 | int unmounted = 0; | |
554 | ||
555 | gossip_debug(GOSSIP_DEV_DEBUG, | |
556 | "%s:pvfs2-client-core: exiting, closing device\n", | |
557 | __func__); | |
558 | ||
559 | mutex_lock(&devreq_mutex); | |
560 | pvfs_bufmap_finalize(); | |
561 | ||
562 | open_access_count--; | |
563 | ||
564 | unmounted = mark_all_pending_mounts(); | |
565 | gossip_debug(GOSSIP_DEV_DEBUG, "PVFS2 Device Close: Filesystem(s) %s\n", | |
566 | (unmounted ? "UNMOUNTED" : "MOUNTED")); | |
567 | mutex_unlock(&devreq_mutex); | |
568 | ||
569 | /* | |
570 | * Walk through the list of ops in the request list, mark them | |
571 | * as purged and wake them up. | |
572 | */ | |
573 | purge_waiting_ops(); | |
574 | /* | |
575 | * Walk through the hash table of in progress operations; mark | |
576 | * them as purged and wake them up | |
577 | */ | |
578 | purge_inprogress_ops(); | |
579 | gossip_debug(GOSSIP_DEV_DEBUG, | |
580 | "pvfs2-client-core: device close complete\n"); | |
581 | return 0; | |
582 | } | |
583 | ||
584 | int is_daemon_in_service(void) | |
585 | { | |
586 | int in_service; | |
587 | ||
588 | /* | |
589 | * What this function does is checks if client-core is alive | |
590 | * based on the access count we maintain on the device. | |
591 | */ | |
592 | mutex_lock(&devreq_mutex); | |
593 | in_service = open_access_count == 1 ? 0 : -EIO; | |
594 | mutex_unlock(&devreq_mutex); | |
595 | return in_service; | |
596 | } | |
597 | ||
598 | static inline long check_ioctl_command(unsigned int command) | |
599 | { | |
600 | /* Check for valid ioctl codes */ | |
601 | if (_IOC_TYPE(command) != PVFS_DEV_MAGIC) { | |
602 | gossip_err("device ioctl magic numbers don't match! Did you rebuild pvfs2-client-core/libpvfs2? [cmd %x, magic %x != %x]\n", | |
603 | command, | |
604 | _IOC_TYPE(command), | |
605 | PVFS_DEV_MAGIC); | |
606 | return -EINVAL; | |
607 | } | |
608 | /* and valid ioctl commands */ | |
609 | if (_IOC_NR(command) >= PVFS_DEV_MAXNR || _IOC_NR(command) <= 0) { | |
610 | gossip_err("Invalid ioctl command number [%d >= %d]\n", | |
611 | _IOC_NR(command), PVFS_DEV_MAXNR); | |
612 | return -ENOIOCTLCMD; | |
613 | } | |
614 | return 0; | |
615 | } | |
616 | ||
617 | static long dispatch_ioctl_command(unsigned int command, unsigned long arg) | |
618 | { | |
619 | static __s32 magic = PVFS2_DEVREQ_MAGIC; | |
620 | static __s32 max_up_size = MAX_ALIGNED_DEV_REQ_UPSIZE; | |
621 | static __s32 max_down_size = MAX_ALIGNED_DEV_REQ_DOWNSIZE; | |
622 | struct PVFS_dev_map_desc user_desc; | |
623 | int ret = 0; | |
624 | struct dev_mask_info_s mask_info = { 0 }; | |
625 | struct dev_mask2_info_s mask2_info = { 0, 0 }; | |
626 | int upstream_kmod = 1; | |
627 | struct list_head *tmp = NULL; | |
628 | struct pvfs2_sb_info_s *pvfs2_sb = NULL; | |
629 | ||
630 | /* mtmoore: add locking here */ | |
631 | ||
632 | switch (command) { | |
633 | case PVFS_DEV_GET_MAGIC: | |
634 | return ((put_user(magic, (__s32 __user *) arg) == -EFAULT) ? | |
635 | -EIO : | |
636 | 0); | |
637 | case PVFS_DEV_GET_MAX_UPSIZE: | |
638 | return ((put_user(max_up_size, | |
639 | (__s32 __user *) arg) == -EFAULT) ? | |
640 | -EIO : | |
641 | 0); | |
642 | case PVFS_DEV_GET_MAX_DOWNSIZE: | |
643 | return ((put_user(max_down_size, | |
644 | (__s32 __user *) arg) == -EFAULT) ? | |
645 | -EIO : | |
646 | 0); | |
647 | case PVFS_DEV_MAP: | |
648 | ret = copy_from_user(&user_desc, | |
649 | (struct PVFS_dev_map_desc __user *) | |
650 | arg, | |
651 | sizeof(struct PVFS_dev_map_desc)); | |
652 | return ret ? -EIO : pvfs_bufmap_initialize(&user_desc); | |
653 | case PVFS_DEV_REMOUNT_ALL: | |
654 | gossip_debug(GOSSIP_DEV_DEBUG, | |
655 | "pvfs2_devreq_ioctl: got PVFS_DEV_REMOUNT_ALL\n"); | |
656 | ||
657 | /* | |
658 | * remount all mounted pvfs2 volumes to regain the lost | |
659 | * dynamic mount tables (if any) -- NOTE: this is done | |
660 | * without keeping the superblock list locked due to the | |
661 | * upcall/downcall waiting. also, the request semaphore is | |
662 | * used to ensure that no operations will be serviced until | |
663 | * all of the remounts are serviced (to avoid ops between | |
664 | * mounts to fail) | |
665 | */ | |
666 | ret = mutex_lock_interruptible(&request_mutex); | |
667 | if (ret < 0) | |
668 | return ret; | |
669 | gossip_debug(GOSSIP_DEV_DEBUG, | |
670 | "pvfs2_devreq_ioctl: priority remount in progress\n"); | |
671 | list_for_each(tmp, &pvfs2_superblocks) { | |
672 | pvfs2_sb = | |
673 | list_entry(tmp, struct pvfs2_sb_info_s, list); | |
674 | if (pvfs2_sb && (pvfs2_sb->sb)) { | |
675 | gossip_debug(GOSSIP_DEV_DEBUG, | |
676 | "Remounting SB %p\n", | |
677 | pvfs2_sb); | |
678 | ||
679 | ret = pvfs2_remount(pvfs2_sb->sb); | |
680 | if (ret) { | |
681 | gossip_debug(GOSSIP_DEV_DEBUG, | |
682 | "SB %p remount failed\n", | |
683 | pvfs2_sb); | |
684 | break; | |
685 | } | |
686 | } | |
687 | } | |
688 | gossip_debug(GOSSIP_DEV_DEBUG, | |
689 | "pvfs2_devreq_ioctl: priority remount complete\n"); | |
690 | mutex_unlock(&request_mutex); | |
691 | return ret; | |
692 | ||
693 | case PVFS_DEV_UPSTREAM: | |
694 | ret = copy_to_user((void __user *)arg, | |
695 | &upstream_kmod, | |
696 | sizeof(upstream_kmod)); | |
697 | ||
698 | if (ret != 0) | |
699 | return -EIO; | |
700 | else | |
701 | return ret; | |
702 | ||
703 | case PVFS_DEV_CLIENT_MASK: | |
704 | ret = copy_from_user(&mask2_info, | |
705 | (void __user *)arg, | |
706 | sizeof(struct dev_mask2_info_s)); | |
707 | ||
708 | if (ret != 0) | |
709 | return -EIO; | |
710 | ||
711 | client_debug_mask.mask1 = mask2_info.mask1_value; | |
712 | client_debug_mask.mask2 = mask2_info.mask2_value; | |
713 | ||
714 | pr_info("%s: client debug mask has been been received " | |
715 | ":%llx: :%llx:\n", | |
716 | __func__, | |
717 | (unsigned long long)client_debug_mask.mask1, | |
718 | (unsigned long long)client_debug_mask.mask2); | |
719 | ||
720 | return ret; | |
721 | ||
722 | case PVFS_DEV_CLIENT_STRING: | |
723 | ret = copy_from_user(&client_debug_array_string, | |
724 | (void __user *)arg, | |
725 | PVFS2_MAX_DEBUG_STRING_LEN); | |
726 | if (ret != 0) { | |
727 | pr_info("%s: " | |
728 | "PVFS_DEV_CLIENT_STRING: copy_from_user failed" | |
729 | "\n", | |
730 | __func__); | |
731 | return -EIO; | |
732 | } | |
733 | ||
734 | pr_info("%s: client debug array string has been been received." | |
735 | "\n", | |
736 | __func__); | |
737 | ||
738 | if (!help_string_initialized) { | |
739 | ||
740 | /* Free the "we don't know yet" default string... */ | |
741 | kfree(debug_help_string); | |
742 | ||
743 | /* build a proper debug help string */ | |
744 | if (orangefs_prepare_debugfs_help_string(0)) { | |
745 | gossip_err("%s: " | |
746 | "prepare_debugfs_help_string failed" | |
747 | "\n", | |
748 | __func__); | |
749 | return -EIO; | |
750 | } | |
751 | ||
752 | /* Replace the boilerplate boot-time debug-help file. */ | |
753 | debugfs_remove(help_file_dentry); | |
754 | ||
755 | help_file_dentry = | |
756 | debugfs_create_file( | |
757 | ORANGEFS_KMOD_DEBUG_HELP_FILE, | |
758 | 0444, | |
759 | debug_dir, | |
760 | debug_help_string, | |
761 | &debug_help_fops); | |
762 | ||
763 | if (!help_file_dentry) { | |
764 | gossip_err("%s: debugfs_create_file failed for" | |
765 | " :%s:!\n", | |
766 | __func__, | |
767 | ORANGEFS_KMOD_DEBUG_HELP_FILE); | |
768 | return -EIO; | |
769 | } | |
770 | } | |
771 | ||
772 | debug_mask_to_string(&client_debug_mask, 1); | |
773 | ||
774 | debugfs_remove(client_debug_dentry); | |
775 | ||
776 | pvfs2_client_debug_init(); | |
777 | ||
778 | help_string_initialized++; | |
779 | ||
780 | return ret; | |
781 | ||
782 | case PVFS_DEV_DEBUG: | |
783 | ret = copy_from_user(&mask_info, | |
784 | (void __user *)arg, | |
785 | sizeof(mask_info)); | |
786 | ||
787 | if (ret != 0) | |
788 | return -EIO; | |
789 | ||
790 | if (mask_info.mask_type == KERNEL_MASK) { | |
791 | if ((mask_info.mask_value == 0) | |
792 | && (kernel_mask_set_mod_init)) { | |
793 | /* | |
794 | * the kernel debug mask was set when the | |
795 | * kernel module was loaded; don't override | |
796 | * it if the client-core was started without | |
797 | * a value for PVFS2_KMODMASK. | |
798 | */ | |
799 | return 0; | |
800 | } | |
801 | debug_mask_to_string(&mask_info.mask_value, | |
802 | mask_info.mask_type); | |
803 | gossip_debug_mask = mask_info.mask_value; | |
804 | pr_info("PVFS: kernel debug mask has been modified to " | |
805 | ":%s: :%llx:\n", | |
806 | kernel_debug_string, | |
807 | (unsigned long long)gossip_debug_mask); | |
808 | } else if (mask_info.mask_type == CLIENT_MASK) { | |
809 | debug_mask_to_string(&mask_info.mask_value, | |
810 | mask_info.mask_type); | |
811 | pr_info("PVFS: client debug mask has been modified to" | |
812 | ":%s: :%llx:\n", | |
813 | client_debug_string, | |
814 | llu(mask_info.mask_value)); | |
815 | } else { | |
816 | gossip_lerr("Invalid mask type....\n"); | |
817 | return -EINVAL; | |
818 | } | |
819 | ||
820 | return ret; | |
821 | ||
822 | default: | |
823 | return -ENOIOCTLCMD; | |
824 | } | |
825 | return -ENOIOCTLCMD; | |
826 | } | |
827 | ||
828 | static long pvfs2_devreq_ioctl(struct file *file, | |
829 | unsigned int command, unsigned long arg) | |
830 | { | |
831 | long ret; | |
832 | ||
833 | /* Check for properly constructed commands */ | |
834 | ret = check_ioctl_command(command); | |
835 | if (ret < 0) | |
836 | return (int)ret; | |
837 | ||
838 | return (int)dispatch_ioctl_command(command, arg); | |
839 | } | |
840 | ||
841 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
842 | ||
843 | /* Compat structure for the PVFS_DEV_MAP ioctl */ | |
844 | struct PVFS_dev_map_desc32 { | |
845 | compat_uptr_t ptr; | |
846 | __s32 total_size; | |
847 | __s32 size; | |
848 | __s32 count; | |
849 | }; | |
850 | ||
851 | static unsigned long translate_dev_map26(unsigned long args, long *error) | |
852 | { | |
853 | struct PVFS_dev_map_desc32 __user *p32 = (void __user *)args; | |
854 | /* | |
855 | * Depending on the architecture, allocate some space on the | |
856 | * user-call-stack based on our expected layout. | |
857 | */ | |
858 | struct PVFS_dev_map_desc __user *p = | |
859 | compat_alloc_user_space(sizeof(*p)); | |
860 | u32 addr; | |
861 | ||
862 | *error = 0; | |
863 | /* get the ptr from the 32 bit user-space */ | |
864 | if (get_user(addr, &p32->ptr)) | |
865 | goto err; | |
866 | /* try to put that into a 64-bit layout */ | |
867 | if (put_user(compat_ptr(addr), &p->ptr)) | |
868 | goto err; | |
869 | /* copy the remaining fields */ | |
870 | if (copy_in_user(&p->total_size, &p32->total_size, sizeof(__s32))) | |
871 | goto err; | |
872 | if (copy_in_user(&p->size, &p32->size, sizeof(__s32))) | |
873 | goto err; | |
874 | if (copy_in_user(&p->count, &p32->count, sizeof(__s32))) | |
875 | goto err; | |
876 | return (unsigned long)p; | |
877 | err: | |
878 | *error = -EFAULT; | |
879 | return 0; | |
880 | } | |
881 | ||
882 | /* | |
883 | * 32 bit user-space apps' ioctl handlers when kernel modules | |
884 | * is compiled as a 64 bit one | |
885 | */ | |
886 | static long pvfs2_devreq_compat_ioctl(struct file *filp, unsigned int cmd, | |
887 | unsigned long args) | |
888 | { | |
889 | long ret; | |
890 | unsigned long arg = args; | |
891 | ||
892 | /* Check for properly constructed commands */ | |
893 | ret = check_ioctl_command(cmd); | |
894 | if (ret < 0) | |
895 | return ret; | |
896 | if (cmd == PVFS_DEV_MAP) { | |
897 | /* | |
898 | * convert the arguments to what we expect internally | |
899 | * in kernel space | |
900 | */ | |
901 | arg = translate_dev_map26(args, &ret); | |
902 | if (ret < 0) { | |
903 | gossip_err("Could not translate dev map\n"); | |
904 | return ret; | |
905 | } | |
906 | } | |
907 | /* no other ioctl requires translation */ | |
908 | return dispatch_ioctl_command(cmd, arg); | |
909 | } | |
910 | ||
2c590d5f MM |
911 | #endif /* CONFIG_COMPAT is in .config */ |
912 | ||
913 | /* | |
914 | * The following two ioctl32 functions had been refactored into the above | |
915 | * CONFIG_COMPAT ifdef, but that was an over simplification that was | |
916 | * not noticed until we tried to compile on power pc... | |
917 | */ | |
918 | #if (defined(CONFIG_COMPAT) && !defined(HAVE_REGISTER_IOCTL32_CONVERSION)) || !defined(CONFIG_COMPAT) | |
5db11c21 MM |
919 | static int pvfs2_ioctl32_init(void) |
920 | { | |
921 | return 0; | |
922 | } | |
923 | ||
924 | static void pvfs2_ioctl32_cleanup(void) | |
925 | { | |
926 | return; | |
927 | } | |
2c590d5f | 928 | #endif |
5db11c21 MM |
929 | |
930 | /* the assigned character device major number */ | |
931 | static int pvfs2_dev_major; | |
932 | ||
933 | /* | |
934 | * Initialize pvfs2 device specific state: | |
935 | * Must be called at module load time only | |
936 | */ | |
937 | int pvfs2_dev_init(void) | |
938 | { | |
939 | int ret; | |
940 | ||
941 | /* register the ioctl32 sub-system */ | |
942 | ret = pvfs2_ioctl32_init(); | |
943 | if (ret < 0) | |
944 | return ret; | |
945 | ||
946 | /* register pvfs2-req device */ | |
947 | pvfs2_dev_major = register_chrdev(0, | |
948 | PVFS2_REQDEVICE_NAME, | |
949 | &pvfs2_devreq_file_operations); | |
950 | if (pvfs2_dev_major < 0) { | |
951 | gossip_debug(GOSSIP_DEV_DEBUG, | |
952 | "Failed to register /dev/%s (error %d)\n", | |
953 | PVFS2_REQDEVICE_NAME, pvfs2_dev_major); | |
954 | pvfs2_ioctl32_cleanup(); | |
955 | return pvfs2_dev_major; | |
956 | } | |
957 | ||
958 | gossip_debug(GOSSIP_DEV_DEBUG, | |
959 | "*** /dev/%s character device registered ***\n", | |
960 | PVFS2_REQDEVICE_NAME); | |
961 | gossip_debug(GOSSIP_DEV_DEBUG, "'mknod /dev/%s c %d 0'.\n", | |
962 | PVFS2_REQDEVICE_NAME, pvfs2_dev_major); | |
963 | return 0; | |
964 | } | |
965 | ||
966 | void pvfs2_dev_cleanup(void) | |
967 | { | |
968 | unregister_chrdev(pvfs2_dev_major, PVFS2_REQDEVICE_NAME); | |
969 | gossip_debug(GOSSIP_DEV_DEBUG, | |
970 | "*** /dev/%s character device unregistered ***\n", | |
971 | PVFS2_REQDEVICE_NAME); | |
972 | /* unregister the ioctl32 sub-system */ | |
973 | pvfs2_ioctl32_cleanup(); | |
974 | } | |
975 | ||
976 | static unsigned int pvfs2_devreq_poll(struct file *file, | |
977 | struct poll_table_struct *poll_table) | |
978 | { | |
979 | int poll_revent_mask = 0; | |
980 | ||
981 | if (open_access_count == 1) { | |
982 | poll_wait(file, &pvfs2_request_list_waitq, poll_table); | |
983 | ||
984 | spin_lock(&pvfs2_request_list_lock); | |
985 | if (!list_empty(&pvfs2_request_list)) | |
986 | poll_revent_mask |= POLL_IN; | |
987 | spin_unlock(&pvfs2_request_list_lock); | |
988 | } | |
989 | return poll_revent_mask; | |
990 | } | |
991 | ||
992 | const struct file_operations pvfs2_devreq_file_operations = { | |
993 | .owner = THIS_MODULE, | |
994 | .read = pvfs2_devreq_read, | |
995 | .write_iter = pvfs2_devreq_write_iter, | |
996 | .open = pvfs2_devreq_open, | |
997 | .release = pvfs2_devreq_release, | |
998 | .unlocked_ioctl = pvfs2_devreq_ioctl, | |
999 | ||
1000 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
1001 | .compat_ioctl = pvfs2_devreq_compat_ioctl, | |
1002 | #endif | |
1003 | .poll = pvfs2_devreq_poll | |
1004 | }; |