]>
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" | |
44f46410 | 14 | #include "orangefs-debugfs.h" |
5db11c21 MM |
15 | |
16 | #include <linux/debugfs.h> | |
17 | #include <linux/slab.h> | |
18 | ||
19 | /* this file implements the /dev/pvfs2-req device node */ | |
20 | ||
482664dd | 21 | uint32_t orangefs_userspace_version; |
f2ee3b75 | 22 | |
5db11c21 MM |
23 | static int open_access_count; |
24 | ||
a0fe0515 MB |
25 | static DEFINE_MUTEX(devreq_mutex); |
26 | ||
5db11c21 MM |
27 | #define DUMP_DEVICE_ERROR() \ |
28 | do { \ | |
29 | gossip_err("*****************************************************\n");\ | |
8bb8aefd | 30 | gossip_err("ORANGEFS Device Error: You cannot open the device file "); \ |
5db11c21 | 31 | gossip_err("\n/dev/%s more than once. Please make sure that\nthere " \ |
8bb8aefd | 32 | "are no ", ORANGEFS_REQDEVICE_NAME); \ |
5db11c21 MM |
33 | gossip_err("instances of a program using this device\ncurrently " \ |
34 | "running. (You must verify this!)\n"); \ | |
35 | gossip_err("For example, you can use the lsof program as follows:\n");\ | |
36 | gossip_err("'lsof | grep %s' (run this as root)\n", \ | |
8bb8aefd | 37 | ORANGEFS_REQDEVICE_NAME); \ |
5db11c21 MM |
38 | gossip_err(" open_access_count = %d\n", open_access_count); \ |
39 | gossip_err("*****************************************************\n");\ | |
40 | } while (0) | |
41 | ||
42 | static int hash_func(__u64 tag, int table_size) | |
43 | { | |
2c590d5f | 44 | return do_div(tag, (unsigned int)table_size); |
5db11c21 MM |
45 | } |
46 | ||
8bb8aefd | 47 | static void orangefs_devreq_add_op(struct orangefs_kernel_op_s *op) |
5db11c21 MM |
48 | { |
49 | int index = hash_func(op->tag, hash_table_size); | |
50 | ||
1d503617 | 51 | list_add_tail(&op->list, &orangefs_htable_ops_in_progress[index]); |
5db11c21 MM |
52 | } |
53 | ||
ca9f518e MM |
54 | /* |
55 | * find the op with this tag and remove it from the in progress | |
56 | * hash table. | |
57 | */ | |
8bb8aefd | 58 | static struct orangefs_kernel_op_s *orangefs_devreq_remove_op(__u64 tag) |
5db11c21 | 59 | { |
8bb8aefd | 60 | struct orangefs_kernel_op_s *op, *next; |
5db11c21 MM |
61 | int index; |
62 | ||
63 | index = hash_func(tag, hash_table_size); | |
64 | ||
1d503617 | 65 | spin_lock(&orangefs_htable_ops_in_progress_lock); |
5db11c21 MM |
66 | list_for_each_entry_safe(op, |
67 | next, | |
1d503617 | 68 | &orangefs_htable_ops_in_progress[index], |
5db11c21 | 69 | list) { |
05a50a5b AV |
70 | if (op->tag == tag && !op_state_purged(op) && |
71 | !op_state_given_up(op)) { | |
ed42fe05 | 72 | list_del_init(&op->list); |
1d503617 | 73 | spin_unlock(&orangefs_htable_ops_in_progress_lock); |
5db11c21 MM |
74 | return op; |
75 | } | |
76 | } | |
77 | ||
1d503617 | 78 | spin_unlock(&orangefs_htable_ops_in_progress_lock); |
5db11c21 MM |
79 | return NULL; |
80 | } | |
81 | ||
acfcbaf1 MB |
82 | /* Returns whether any FS are still pending remounted */ |
83 | static int mark_all_pending_mounts(void) | |
84 | { | |
85 | int unmounted = 1; | |
86 | struct orangefs_sb_info_s *orangefs_sb = NULL; | |
87 | ||
88 | spin_lock(&orangefs_superblocks_lock); | |
89 | list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) { | |
90 | /* All of these file system require a remount */ | |
91 | orangefs_sb->mount_pending = 1; | |
92 | unmounted = 0; | |
93 | } | |
94 | spin_unlock(&orangefs_superblocks_lock); | |
95 | return unmounted; | |
96 | } | |
97 | ||
98 | /* | |
99 | * Determine if a given file system needs to be remounted or not | |
100 | * Returns -1 on error | |
101 | * 0 if already mounted | |
102 | * 1 if needs remount | |
103 | */ | |
104 | static int fs_mount_pending(__s32 fsid) | |
105 | { | |
106 | int mount_pending = -1; | |
107 | struct orangefs_sb_info_s *orangefs_sb = NULL; | |
108 | ||
109 | spin_lock(&orangefs_superblocks_lock); | |
110 | list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) { | |
111 | if (orangefs_sb->fs_id == fsid) { | |
112 | mount_pending = orangefs_sb->mount_pending; | |
113 | break; | |
114 | } | |
115 | } | |
116 | spin_unlock(&orangefs_superblocks_lock); | |
117 | return mount_pending; | |
118 | } | |
119 | ||
8bb8aefd | 120 | static int orangefs_devreq_open(struct inode *inode, struct file *file) |
5db11c21 MM |
121 | { |
122 | int ret = -EINVAL; | |
123 | ||
78fee0b6 JH |
124 | /* in order to ensure that the filesystem driver sees correct UIDs */ |
125 | if (file->f_cred->user_ns != &init_user_ns) { | |
126 | gossip_err("%s: device cannot be opened outside init_user_ns\n", | |
127 | __func__); | |
128 | goto out; | |
129 | } | |
130 | ||
5db11c21 | 131 | if (!(file->f_flags & O_NONBLOCK)) { |
97f10027 MM |
132 | gossip_err("%s: device cannot be opened in blocking mode\n", |
133 | __func__); | |
5db11c21 MM |
134 | goto out; |
135 | } | |
136 | ret = -EACCES; | |
97f10027 | 137 | gossip_debug(GOSSIP_DEV_DEBUG, "client-core: opening device\n"); |
5db11c21 MM |
138 | mutex_lock(&devreq_mutex); |
139 | ||
140 | if (open_access_count == 0) { | |
fee25ce1 | 141 | open_access_count = 1; |
fb6d2526 | 142 | ret = 0; |
5db11c21 MM |
143 | } else { |
144 | DUMP_DEVICE_ERROR(); | |
145 | } | |
146 | mutex_unlock(&devreq_mutex); | |
147 | ||
148 | out: | |
149 | ||
150 | gossip_debug(GOSSIP_DEV_DEBUG, | |
151 | "pvfs2-client-core: open device complete (ret = %d)\n", | |
152 | ret); | |
153 | return ret; | |
154 | } | |
155 | ||
97f10027 | 156 | /* Function for read() callers into the device */ |
8bb8aefd | 157 | static ssize_t orangefs_devreq_read(struct file *file, |
5db11c21 MM |
158 | char __user *buf, |
159 | size_t count, loff_t *offset) | |
160 | { | |
8bb8aefd YL |
161 | struct orangefs_kernel_op_s *op, *temp; |
162 | __s32 proto_ver = ORANGEFS_KERNEL_PROTO_VERSION; | |
163 | static __s32 magic = ORANGEFS_DEVREQ_MAGIC; | |
164 | struct orangefs_kernel_op_s *cur_op = NULL; | |
24c8d080 | 165 | unsigned long ret; |
5db11c21 | 166 | |
24c8d080 | 167 | /* We do not support blocking IO. */ |
5db11c21 | 168 | if (!(file->f_flags & O_NONBLOCK)) { |
97f10027 MM |
169 | gossip_err("%s: blocking read from client-core.\n", |
170 | __func__); | |
5db11c21 | 171 | return -EINVAL; |
24c8d080 MB |
172 | } |
173 | ||
174 | /* | |
a762ae6d | 175 | * The client will do an ioctl to find MAX_DEV_REQ_UPSIZE, then |
24c8d080 MB |
176 | * always read with that size buffer. |
177 | */ | |
a762ae6d | 178 | if (count != MAX_DEV_REQ_UPSIZE) { |
24c8d080 MB |
179 | gossip_err("orangefs: client-core tried to read wrong size\n"); |
180 | return -EINVAL; | |
181 | } | |
182 | ||
ed42fe05 | 183 | restart: |
24c8d080 | 184 | /* Get next op (if any) from top of list. */ |
8bb8aefd YL |
185 | spin_lock(&orangefs_request_list_lock); |
186 | list_for_each_entry_safe(op, temp, &orangefs_request_list, list) { | |
24c8d080 MB |
187 | __s32 fsid; |
188 | /* This lock is held past the end of the loop when we break. */ | |
189 | spin_lock(&op->lock); | |
05a50a5b | 190 | if (unlikely(op_state_purged(op) || op_state_given_up(op))) { |
ed42fe05 AV |
191 | spin_unlock(&op->lock); |
192 | continue; | |
193 | } | |
24c8d080 MB |
194 | |
195 | fsid = fsid_of_op(op); | |
8bb8aefd | 196 | if (fsid != ORANGEFS_FS_ID_NULL) { |
24c8d080 MB |
197 | int ret; |
198 | /* Skip ops whose filesystem needs to be mounted. */ | |
199 | ret = fs_mount_pending(fsid); | |
200 | if (ret == 1) { | |
5db11c21 | 201 | gossip_debug(GOSSIP_DEV_DEBUG, |
5090c967 MM |
202 | "%s: mount pending, skipping op tag " |
203 | "%llu %s\n", | |
204 | __func__, | |
205 | llu(op->tag), | |
206 | get_opname_string(op)); | |
24c8d080 MB |
207 | spin_unlock(&op->lock); |
208 | continue; | |
97f10027 MM |
209 | /* |
210 | * Skip ops whose filesystem we don't know about unless | |
211 | * it is being mounted. | |
212 | */ | |
24c8d080 MB |
213 | /* XXX: is there a better way to detect this? */ |
214 | } else if (ret == -1 && | |
97f10027 MM |
215 | !(op->upcall.type == |
216 | ORANGEFS_VFS_OP_FS_MOUNT || | |
217 | op->upcall.type == | |
218 | ORANGEFS_VFS_OP_GETATTR)) { | |
24c8d080 MB |
219 | gossip_debug(GOSSIP_DEV_DEBUG, |
220 | "orangefs: skipping op tag %llu %s\n", | |
221 | llu(op->tag), get_opname_string(op)); | |
222 | gossip_err( | |
223 | "orangefs: ERROR: fs_mount_pending %d\n", | |
224 | fsid); | |
225 | spin_unlock(&op->lock); | |
5db11c21 | 226 | continue; |
5db11c21 MM |
227 | } |
228 | } | |
24c8d080 MB |
229 | /* |
230 | * Either this op does not pertain to a filesystem, is mounting | |
231 | * a filesystem, or pertains to a mounted filesystem. Let it | |
232 | * through. | |
233 | */ | |
234 | cur_op = op; | |
235 | break; | |
236 | } | |
237 | ||
238 | /* | |
239 | * At this point we either have a valid op and can continue or have not | |
240 | * found an op and must ask the client to try again later. | |
241 | */ | |
242 | if (!cur_op) { | |
8bb8aefd | 243 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 244 | return -EAGAIN; |
5db11c21 MM |
245 | } |
246 | ||
ca9f518e MM |
247 | gossip_debug(GOSSIP_DEV_DEBUG, "%s: reading op tag %llu %s\n", |
248 | __func__, | |
249 | llu(cur_op->tag), | |
250 | get_opname_string(cur_op)); | |
5db11c21 | 251 | |
24c8d080 MB |
252 | /* |
253 | * Such an op should never be on the list in the first place. If so, we | |
254 | * will abort. | |
255 | */ | |
256 | if (op_state_in_progress(cur_op) || op_state_serviced(cur_op)) { | |
257 | gossip_err("orangefs: ERROR: Current op already queued.\n"); | |
05a50a5b | 258 | list_del_init(&cur_op->list); |
5db11c21 | 259 | spin_unlock(&cur_op->lock); |
8bb8aefd | 260 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 261 | return -EAGAIN; |
5db11c21 | 262 | } |
ca9f518e | 263 | |
ed42fe05 | 264 | list_del_init(&cur_op->list); |
8bb8aefd | 265 | spin_unlock(&orangefs_request_list_lock); |
ed42fe05 | 266 | |
24c8d080 MB |
267 | spin_unlock(&cur_op->lock); |
268 | ||
269 | /* Push the upcall out. */ | |
270 | ret = copy_to_user(buf, &proto_ver, sizeof(__s32)); | |
271 | if (ret != 0) | |
272 | goto error; | |
273 | ret = copy_to_user(buf+sizeof(__s32), &magic, sizeof(__s32)); | |
274 | if (ret != 0) | |
275 | goto error; | |
276 | ret = copy_to_user(buf+2 * sizeof(__s32), &cur_op->tag, sizeof(__u64)); | |
277 | if (ret != 0) | |
278 | goto error; | |
279 | ret = copy_to_user(buf+2*sizeof(__s32)+sizeof(__u64), &cur_op->upcall, | |
8bb8aefd | 280 | sizeof(struct orangefs_upcall_s)); |
24c8d080 MB |
281 | if (ret != 0) |
282 | goto error; | |
283 | ||
1d503617 | 284 | spin_lock(&orangefs_htable_ops_in_progress_lock); |
ed42fe05 AV |
285 | spin_lock(&cur_op->lock); |
286 | if (unlikely(op_state_given_up(cur_op))) { | |
287 | spin_unlock(&cur_op->lock); | |
1d503617 | 288 | spin_unlock(&orangefs_htable_ops_in_progress_lock); |
05a50a5b | 289 | complete(&cur_op->waitq); |
ed42fe05 AV |
290 | goto restart; |
291 | } | |
292 | ||
293 | /* | |
294 | * Set the operation to be in progress and move it between lists since | |
295 | * it has been sent to the client. | |
296 | */ | |
297 | set_op_state_inprogress(cur_op); | |
9d9e7ba9 MM |
298 | gossip_debug(GOSSIP_DEV_DEBUG, |
299 | "%s: 1 op:%s: op_state:%d: process:%s:\n", | |
300 | __func__, | |
301 | get_opname_string(cur_op), | |
302 | cur_op->op_state, | |
303 | current->comm); | |
ed42fe05 AV |
304 | orangefs_devreq_add_op(cur_op); |
305 | spin_unlock(&cur_op->lock); | |
1d503617 | 306 | spin_unlock(&orangefs_htable_ops_in_progress_lock); |
ed42fe05 | 307 | |
24c8d080 | 308 | /* The client only asks to read one size buffer. */ |
a762ae6d | 309 | return MAX_DEV_REQ_UPSIZE; |
24c8d080 MB |
310 | error: |
311 | /* | |
312 | * We were unable to copy the op data to the client. Put the op back in | |
313 | * list. If client has crashed, the op will be purged later when the | |
314 | * device is released. | |
315 | */ | |
316 | gossip_err("orangefs: Failed to copy data to user space\n"); | |
8bb8aefd | 317 | spin_lock(&orangefs_request_list_lock); |
24c8d080 | 318 | spin_lock(&cur_op->lock); |
ed42fe05 AV |
319 | if (likely(!op_state_given_up(cur_op))) { |
320 | set_op_state_waiting(cur_op); | |
9d9e7ba9 MM |
321 | gossip_debug(GOSSIP_DEV_DEBUG, |
322 | "%s: 2 op:%s: op_state:%d: process:%s:\n", | |
323 | __func__, | |
324 | get_opname_string(cur_op), | |
325 | cur_op->op_state, | |
326 | current->comm); | |
ed42fe05 | 327 | list_add(&cur_op->list, &orangefs_request_list); |
05a50a5b AV |
328 | spin_unlock(&cur_op->lock); |
329 | } else { | |
330 | spin_unlock(&cur_op->lock); | |
331 | complete(&cur_op->waitq); | |
ed42fe05 | 332 | } |
8bb8aefd | 333 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 334 | return -EFAULT; |
5db11c21 MM |
335 | } |
336 | ||
97f10027 | 337 | /* |
b3ae4755 MM |
338 | * Function for writev() callers into the device. |
339 | * | |
340 | * Userspace should have written: | |
341 | * - __u32 version | |
342 | * - __u32 magic | |
343 | * - __u64 tag | |
344 | * - struct orangefs_downcall_s | |
345 | * - trailer buffer (in the case of READDIR operations) | |
97f10027 | 346 | */ |
b3ae4755 MM |
347 | static ssize_t orangefs_devreq_write_iter(struct kiocb *iocb, |
348 | struct iov_iter *iter) | |
5db11c21 | 349 | { |
b3ae4755 | 350 | ssize_t ret; |
8bb8aefd | 351 | struct orangefs_kernel_op_s *op = NULL; |
b3ae4755 MM |
352 | struct { |
353 | __u32 version; | |
354 | __u32 magic; | |
355 | __u64 tag; | |
356 | } head; | |
357 | int total = ret = iov_iter_count(iter); | |
358 | int n; | |
359 | int downcall_size = sizeof(struct orangefs_downcall_s); | |
360 | int head_size = sizeof(head); | |
361 | ||
362 | gossip_debug(GOSSIP_DEV_DEBUG, "%s: total:%d: ret:%zd:\n", | |
363 | __func__, | |
364 | total, | |
365 | ret); | |
5db11c21 | 366 | |
b3ae4755 | 367 | if (total < MAX_DEV_REQ_DOWNSIZE) { |
cf0c2771 | 368 | gossip_err("%s: total:%d: must be at least:%u:\n", |
b3ae4755 MM |
369 | __func__, |
370 | total, | |
cf0c2771 | 371 | (unsigned int) MAX_DEV_REQ_DOWNSIZE); |
ed42fe05 | 372 | return -EFAULT; |
5db11c21 | 373 | } |
b3ae4755 MM |
374 | |
375 | n = copy_from_iter(&head, head_size, iter); | |
376 | if (n < head_size) { | |
377 | gossip_err("%s: failed to copy head.\n", __func__); | |
ed42fe05 | 378 | return -EFAULT; |
97f10027 | 379 | } |
b3ae4755 MM |
380 | |
381 | if (head.version < ORANGEFS_MINIMUM_USERSPACE_VERSION) { | |
382 | gossip_err("%s: userspace claims version" | |
383 | "%d, minimum version required: %d.\n", | |
384 | __func__, | |
385 | head.version, | |
386 | ORANGEFS_MINIMUM_USERSPACE_VERSION); | |
ed42fe05 | 387 | return -EPROTO; |
5db11c21 | 388 | } |
5db11c21 | 389 | |
b3ae4755 MM |
390 | if (head.magic != ORANGEFS_DEVREQ_MAGIC) { |
391 | gossip_err("Error: Device magic number does not match.\n"); | |
ed42fe05 | 392 | return -EPROTO; |
b3ae4755 | 393 | } |
5db11c21 | 394 | |
482664dd MB |
395 | if (!orangefs_userspace_version) { |
396 | orangefs_userspace_version = head.version; | |
397 | } else if (orangefs_userspace_version != head.version) { | |
f2ee3b75 MB |
398 | gossip_err("Error: userspace version changes\n"); |
399 | return -EPROTO; | |
400 | } | |
401 | ||
ca9f518e | 402 | /* remove the op from the in progress hash table */ |
b3ae4755 MM |
403 | op = orangefs_devreq_remove_op(head.tag); |
404 | if (!op) { | |
405 | gossip_err("WARNING: No one's waiting for tag %llu\n", | |
406 | llu(head.tag)); | |
ed42fe05 | 407 | return ret; |
b3ae4755 | 408 | } |
5db11c21 | 409 | |
b3ae4755 MM |
410 | n = copy_from_iter(&op->downcall, downcall_size, iter); |
411 | if (n != downcall_size) { | |
412 | gossip_err("%s: failed to copy downcall.\n", __func__); | |
5964c1b8 | 413 | goto Efault; |
5db11c21 MM |
414 | } |
415 | ||
b3ae4755 MM |
416 | if (op->downcall.status) |
417 | goto wakeup; | |
97f10027 | 418 | |
b3ae4755 MM |
419 | /* |
420 | * We've successfully peeled off the head and the downcall. | |
421 | * Something has gone awry if total doesn't equal the | |
422 | * sum of head_size, downcall_size and trailer_size. | |
423 | */ | |
424 | if ((head_size + downcall_size + op->downcall.trailer_size) != total) { | |
425 | gossip_err("%s: funky write, head_size:%d" | |
426 | ": downcall_size:%d: trailer_size:%lld" | |
427 | ": total size:%d:\n", | |
428 | __func__, | |
429 | head_size, | |
430 | downcall_size, | |
431 | op->downcall.trailer_size, | |
432 | total); | |
5964c1b8 | 433 | goto Efault; |
b3ae4755 | 434 | } |
97f10027 | 435 | |
b3ae4755 MM |
436 | /* Only READDIR operations should have trailers. */ |
437 | if ((op->downcall.type != ORANGEFS_VFS_OP_READDIR) && | |
438 | (op->downcall.trailer_size != 0)) { | |
439 | gossip_err("%s: %x operation with trailer.", | |
440 | __func__, | |
441 | op->downcall.type); | |
5964c1b8 | 442 | goto Efault; |
b3ae4755 | 443 | } |
97f10027 | 444 | |
b3ae4755 MM |
445 | /* READDIR operations should always have trailers. */ |
446 | if ((op->downcall.type == ORANGEFS_VFS_OP_READDIR) && | |
447 | (op->downcall.trailer_size == 0)) { | |
448 | gossip_err("%s: %x operation with no trailer.", | |
449 | __func__, | |
450 | op->downcall.type); | |
5964c1b8 | 451 | goto Efault; |
b3ae4755 | 452 | } |
97f10027 | 453 | |
b3ae4755 MM |
454 | if (op->downcall.type != ORANGEFS_VFS_OP_READDIR) |
455 | goto wakeup; | |
5db11c21 | 456 | |
b3ae4755 MM |
457 | op->downcall.trailer_buf = |
458 | vmalloc(op->downcall.trailer_size); | |
459 | if (op->downcall.trailer_buf == NULL) { | |
460 | gossip_err("%s: failed trailer vmalloc.\n", | |
461 | __func__); | |
5964c1b8 | 462 | goto Enomem; |
b3ae4755 MM |
463 | } |
464 | memset(op->downcall.trailer_buf, 0, op->downcall.trailer_size); | |
465 | n = copy_from_iter(op->downcall.trailer_buf, | |
466 | op->downcall.trailer_size, | |
467 | iter); | |
468 | if (n != op->downcall.trailer_size) { | |
469 | gossip_err("%s: failed to copy trailer.\n", __func__); | |
470 | vfree(op->downcall.trailer_buf); | |
5964c1b8 | 471 | goto Efault; |
b3ae4755 | 472 | } |
97f10027 | 473 | |
b3ae4755 | 474 | wakeup: |
2a9e5c22 | 475 | /* |
9f08cfe9 MM |
476 | * Return to vfs waitqueue, and back to service_operation |
477 | * through wait_for_matching_downcall. | |
2a9e5c22 AV |
478 | */ |
479 | spin_lock(&op->lock); | |
5964c1b8 | 480 | if (unlikely(op_is_cancel(op))) { |
2a9e5c22 | 481 | spin_unlock(&op->lock); |
78699e29 | 482 | put_cancel(op); |
5964c1b8 AV |
483 | } else if (unlikely(op_state_given_up(op))) { |
484 | spin_unlock(&op->lock); | |
05a50a5b | 485 | complete(&op->waitq); |
5964c1b8 AV |
486 | } else { |
487 | set_op_state_serviced(op); | |
9d9e7ba9 MM |
488 | gossip_debug(GOSSIP_DEV_DEBUG, |
489 | "%s: op:%s: op_state:%d: process:%s:\n", | |
490 | __func__, | |
491 | get_opname_string(op), | |
492 | op->op_state, | |
493 | current->comm); | |
5964c1b8 AV |
494 | spin_unlock(&op->lock); |
495 | } | |
b3ae4755 | 496 | return ret; |
ed42fe05 | 497 | |
5964c1b8 AV |
498 | Efault: |
499 | op->downcall.status = -(ORANGEFS_ERROR_BIT | 9); | |
500 | ret = -EFAULT; | |
501 | goto wakeup; | |
502 | ||
503 | Enomem: | |
504 | op->downcall.status = -(ORANGEFS_ERROR_BIT | 8); | |
505 | ret = -ENOMEM; | |
506 | goto wakeup; | |
5db11c21 MM |
507 | } |
508 | ||
5db11c21 MM |
509 | /* |
510 | * NOTE: gets called when the last reference to this device is dropped. | |
511 | * Using the open_access_count variable, we enforce a reference count | |
512 | * on this file so that it can be opened by only one process at a time. | |
513 | * the devreq_mutex is used to make sure all i/o has completed | |
8bb8aefd | 514 | * before we call orangefs_bufmap_finalize, and similar such tricky |
5db11c21 MM |
515 | * situations |
516 | */ | |
8bb8aefd | 517 | static int orangefs_devreq_release(struct inode *inode, struct file *file) |
5db11c21 MM |
518 | { |
519 | int unmounted = 0; | |
520 | ||
521 | gossip_debug(GOSSIP_DEV_DEBUG, | |
522 | "%s:pvfs2-client-core: exiting, closing device\n", | |
523 | __func__); | |
524 | ||
525 | mutex_lock(&devreq_mutex); | |
ea2c9c9f | 526 | orangefs_bufmap_finalize(); |
5db11c21 | 527 | |
fee25ce1 | 528 | open_access_count = -1; |
5db11c21 MM |
529 | |
530 | unmounted = mark_all_pending_mounts(); | |
8bb8aefd | 531 | gossip_debug(GOSSIP_DEV_DEBUG, "ORANGEFS Device Close: Filesystem(s) %s\n", |
5db11c21 | 532 | (unmounted ? "UNMOUNTED" : "MOUNTED")); |
5db11c21 | 533 | |
5db11c21 | 534 | purge_waiting_ops(); |
5db11c21 | 535 | purge_inprogress_ops(); |
ea2c9c9f AV |
536 | |
537 | orangefs_bufmap_run_down(); | |
538 | ||
5db11c21 MM |
539 | gossip_debug(GOSSIP_DEV_DEBUG, |
540 | "pvfs2-client-core: device close complete\n"); | |
fee25ce1 | 541 | open_access_count = 0; |
482664dd | 542 | orangefs_userspace_version = 0; |
fee25ce1 | 543 | mutex_unlock(&devreq_mutex); |
5db11c21 MM |
544 | return 0; |
545 | } | |
546 | ||
547 | int is_daemon_in_service(void) | |
548 | { | |
549 | int in_service; | |
550 | ||
551 | /* | |
552 | * What this function does is checks if client-core is alive | |
553 | * based on the access count we maintain on the device. | |
554 | */ | |
555 | mutex_lock(&devreq_mutex); | |
556 | in_service = open_access_count == 1 ? 0 : -EIO; | |
557 | mutex_unlock(&devreq_mutex); | |
558 | return in_service; | |
559 | } | |
560 | ||
78699e29 AV |
561 | bool __is_daemon_in_service(void) |
562 | { | |
563 | return open_access_count == 1; | |
564 | } | |
565 | ||
5db11c21 MM |
566 | static inline long check_ioctl_command(unsigned int command) |
567 | { | |
568 | /* Check for valid ioctl codes */ | |
8bb8aefd | 569 | if (_IOC_TYPE(command) != ORANGEFS_DEV_MAGIC) { |
5db11c21 MM |
570 | gossip_err("device ioctl magic numbers don't match! Did you rebuild pvfs2-client-core/libpvfs2? [cmd %x, magic %x != %x]\n", |
571 | command, | |
572 | _IOC_TYPE(command), | |
8bb8aefd | 573 | ORANGEFS_DEV_MAGIC); |
5db11c21 MM |
574 | return -EINVAL; |
575 | } | |
576 | /* and valid ioctl commands */ | |
8bb8aefd | 577 | if (_IOC_NR(command) >= ORANGEFS_DEV_MAXNR || _IOC_NR(command) <= 0) { |
5db11c21 | 578 | gossip_err("Invalid ioctl command number [%d >= %d]\n", |
8bb8aefd | 579 | _IOC_NR(command), ORANGEFS_DEV_MAXNR); |
5db11c21 MM |
580 | return -ENOIOCTLCMD; |
581 | } | |
582 | return 0; | |
583 | } | |
584 | ||
585 | static long dispatch_ioctl_command(unsigned int command, unsigned long arg) | |
586 | { | |
8bb8aefd | 587 | static __s32 magic = ORANGEFS_DEVREQ_MAGIC; |
a762ae6d MB |
588 | static __s32 max_up_size = MAX_DEV_REQ_UPSIZE; |
589 | static __s32 max_down_size = MAX_DEV_REQ_DOWNSIZE; | |
8bb8aefd | 590 | struct ORANGEFS_dev_map_desc user_desc; |
5db11c21 | 591 | int ret = 0; |
5db11c21 | 592 | int upstream_kmod = 1; |
45996492 | 593 | struct orangefs_sb_info_s *orangefs_sb; |
5db11c21 MM |
594 | |
595 | /* mtmoore: add locking here */ | |
596 | ||
597 | switch (command) { | |
8bb8aefd | 598 | case ORANGEFS_DEV_GET_MAGIC: |
5db11c21 MM |
599 | return ((put_user(magic, (__s32 __user *) arg) == -EFAULT) ? |
600 | -EIO : | |
601 | 0); | |
8bb8aefd | 602 | case ORANGEFS_DEV_GET_MAX_UPSIZE: |
5db11c21 MM |
603 | return ((put_user(max_up_size, |
604 | (__s32 __user *) arg) == -EFAULT) ? | |
605 | -EIO : | |
606 | 0); | |
8bb8aefd | 607 | case ORANGEFS_DEV_GET_MAX_DOWNSIZE: |
5db11c21 MM |
608 | return ((put_user(max_down_size, |
609 | (__s32 __user *) arg) == -EFAULT) ? | |
610 | -EIO : | |
611 | 0); | |
8bb8aefd | 612 | case ORANGEFS_DEV_MAP: |
5db11c21 | 613 | ret = copy_from_user(&user_desc, |
8bb8aefd | 614 | (struct ORANGEFS_dev_map_desc __user *) |
5db11c21 | 615 | arg, |
8bb8aefd | 616 | sizeof(struct ORANGEFS_dev_map_desc)); |
ea2c9c9f AV |
617 | /* WTF -EIO and not -EFAULT? */ |
618 | return ret ? -EIO : orangefs_bufmap_initialize(&user_desc); | |
8bb8aefd | 619 | case ORANGEFS_DEV_REMOUNT_ALL: |
5db11c21 | 620 | gossip_debug(GOSSIP_DEV_DEBUG, |
97f10027 MM |
621 | "%s: got ORANGEFS_DEV_REMOUNT_ALL\n", |
622 | __func__); | |
5db11c21 MM |
623 | |
624 | /* | |
8bb8aefd | 625 | * remount all mounted orangefs volumes to regain the lost |
5db11c21 MM |
626 | * dynamic mount tables (if any) -- NOTE: this is done |
627 | * without keeping the superblock list locked due to the | |
adcf34a2 | 628 | * upcall/downcall waiting. also, the request mutex is |
5db11c21 MM |
629 | * used to ensure that no operations will be serviced until |
630 | * all of the remounts are serviced (to avoid ops between | |
631 | * mounts to fail) | |
632 | */ | |
1d503617 | 633 | ret = mutex_lock_interruptible(&orangefs_request_mutex); |
5db11c21 MM |
634 | if (ret < 0) |
635 | return ret; | |
636 | gossip_debug(GOSSIP_DEV_DEBUG, | |
97f10027 MM |
637 | "%s: priority remount in progress\n", |
638 | __func__); | |
45996492 AV |
639 | spin_lock(&orangefs_superblocks_lock); |
640 | list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) { | |
641 | /* | |
642 | * We have to drop the spinlock, so entries can be | |
643 | * removed. They can't be freed, though, so we just | |
644 | * keep the forward pointers and zero the back ones - | |
645 | * that way we can get to the rest of the list. | |
646 | */ | |
647 | if (!orangefs_sb->list.prev) | |
648 | continue; | |
649 | gossip_debug(GOSSIP_DEV_DEBUG, | |
650 | "%s: Remounting SB %p\n", | |
651 | __func__, | |
652 | orangefs_sb); | |
653 | ||
654 | spin_unlock(&orangefs_superblocks_lock); | |
655 | ret = orangefs_remount(orangefs_sb); | |
656 | spin_lock(&orangefs_superblocks_lock); | |
657 | if (ret) { | |
5db11c21 | 658 | gossip_debug(GOSSIP_DEV_DEBUG, |
45996492 | 659 | "SB %p remount failed\n", |
8bb8aefd | 660 | orangefs_sb); |
45996492 | 661 | break; |
5db11c21 MM |
662 | } |
663 | } | |
45996492 | 664 | spin_unlock(&orangefs_superblocks_lock); |
5db11c21 | 665 | gossip_debug(GOSSIP_DEV_DEBUG, |
97f10027 MM |
666 | "%s: priority remount complete\n", |
667 | __func__); | |
1d503617 | 668 | mutex_unlock(&orangefs_request_mutex); |
5db11c21 MM |
669 | return ret; |
670 | ||
8bb8aefd | 671 | case ORANGEFS_DEV_UPSTREAM: |
5db11c21 MM |
672 | ret = copy_to_user((void __user *)arg, |
673 | &upstream_kmod, | |
674 | sizeof(upstream_kmod)); | |
675 | ||
676 | if (ret != 0) | |
677 | return -EIO; | |
678 | else | |
679 | return ret; | |
680 | ||
8bb8aefd | 681 | case ORANGEFS_DEV_CLIENT_MASK: |
44f46410 | 682 | return orangefs_debugfs_new_client_mask((void __user *)arg); |
8bb8aefd | 683 | case ORANGEFS_DEV_CLIENT_STRING: |
44f46410 | 684 | return orangefs_debugfs_new_client_string((void __user *)arg); |
8bb8aefd | 685 | case ORANGEFS_DEV_DEBUG: |
44f46410 | 686 | return orangefs_debugfs_new_debug((void __user *)arg); |
5db11c21 MM |
687 | default: |
688 | return -ENOIOCTLCMD; | |
689 | } | |
690 | return -ENOIOCTLCMD; | |
691 | } | |
692 | ||
8bb8aefd | 693 | static long orangefs_devreq_ioctl(struct file *file, |
5db11c21 MM |
694 | unsigned int command, unsigned long arg) |
695 | { | |
696 | long ret; | |
697 | ||
698 | /* Check for properly constructed commands */ | |
699 | ret = check_ioctl_command(command); | |
700 | if (ret < 0) | |
701 | return (int)ret; | |
702 | ||
703 | return (int)dispatch_ioctl_command(command, arg); | |
704 | } | |
705 | ||
706 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
707 | ||
8bb8aefd YL |
708 | /* Compat structure for the ORANGEFS_DEV_MAP ioctl */ |
709 | struct ORANGEFS_dev_map_desc32 { | |
5db11c21 MM |
710 | compat_uptr_t ptr; |
711 | __s32 total_size; | |
712 | __s32 size; | |
713 | __s32 count; | |
714 | }; | |
715 | ||
716 | static unsigned long translate_dev_map26(unsigned long args, long *error) | |
717 | { | |
8bb8aefd | 718 | struct ORANGEFS_dev_map_desc32 __user *p32 = (void __user *)args; |
5db11c21 MM |
719 | /* |
720 | * Depending on the architecture, allocate some space on the | |
721 | * user-call-stack based on our expected layout. | |
722 | */ | |
8bb8aefd | 723 | struct ORANGEFS_dev_map_desc __user *p = |
5db11c21 | 724 | compat_alloc_user_space(sizeof(*p)); |
84d02150 | 725 | compat_uptr_t addr; |
5db11c21 MM |
726 | |
727 | *error = 0; | |
728 | /* get the ptr from the 32 bit user-space */ | |
729 | if (get_user(addr, &p32->ptr)) | |
730 | goto err; | |
731 | /* try to put that into a 64-bit layout */ | |
732 | if (put_user(compat_ptr(addr), &p->ptr)) | |
733 | goto err; | |
734 | /* copy the remaining fields */ | |
735 | if (copy_in_user(&p->total_size, &p32->total_size, sizeof(__s32))) | |
736 | goto err; | |
737 | if (copy_in_user(&p->size, &p32->size, sizeof(__s32))) | |
738 | goto err; | |
739 | if (copy_in_user(&p->count, &p32->count, sizeof(__s32))) | |
740 | goto err; | |
741 | return (unsigned long)p; | |
742 | err: | |
743 | *error = -EFAULT; | |
744 | return 0; | |
745 | } | |
746 | ||
747 | /* | |
748 | * 32 bit user-space apps' ioctl handlers when kernel modules | |
749 | * is compiled as a 64 bit one | |
750 | */ | |
8bb8aefd | 751 | static long orangefs_devreq_compat_ioctl(struct file *filp, unsigned int cmd, |
5db11c21 MM |
752 | unsigned long args) |
753 | { | |
754 | long ret; | |
755 | unsigned long arg = args; | |
756 | ||
757 | /* Check for properly constructed commands */ | |
758 | ret = check_ioctl_command(cmd); | |
759 | if (ret < 0) | |
760 | return ret; | |
8bb8aefd | 761 | if (cmd == ORANGEFS_DEV_MAP) { |
5db11c21 MM |
762 | /* |
763 | * convert the arguments to what we expect internally | |
764 | * in kernel space | |
765 | */ | |
766 | arg = translate_dev_map26(args, &ret); | |
767 | if (ret < 0) { | |
768 | gossip_err("Could not translate dev map\n"); | |
769 | return ret; | |
770 | } | |
771 | } | |
772 | /* no other ioctl requires translation */ | |
773 | return dispatch_ioctl_command(cmd, arg); | |
774 | } | |
775 | ||
2c590d5f MM |
776 | #endif /* CONFIG_COMPAT is in .config */ |
777 | ||
5db11c21 | 778 | /* the assigned character device major number */ |
8bb8aefd | 779 | static int orangefs_dev_major; |
5db11c21 MM |
780 | |
781 | /* | |
8bb8aefd | 782 | * Initialize orangefs device specific state: |
5db11c21 MM |
783 | * Must be called at module load time only |
784 | */ | |
8bb8aefd | 785 | int orangefs_dev_init(void) |
5db11c21 | 786 | { |
8bb8aefd YL |
787 | /* register orangefs-req device */ |
788 | orangefs_dev_major = register_chrdev(0, | |
789 | ORANGEFS_REQDEVICE_NAME, | |
790 | &orangefs_devreq_file_operations); | |
791 | if (orangefs_dev_major < 0) { | |
5db11c21 MM |
792 | gossip_debug(GOSSIP_DEV_DEBUG, |
793 | "Failed to register /dev/%s (error %d)\n", | |
8bb8aefd | 794 | ORANGEFS_REQDEVICE_NAME, orangefs_dev_major); |
8bb8aefd | 795 | return orangefs_dev_major; |
5db11c21 MM |
796 | } |
797 | ||
798 | gossip_debug(GOSSIP_DEV_DEBUG, | |
799 | "*** /dev/%s character device registered ***\n", | |
8bb8aefd | 800 | ORANGEFS_REQDEVICE_NAME); |
5db11c21 | 801 | gossip_debug(GOSSIP_DEV_DEBUG, "'mknod /dev/%s c %d 0'.\n", |
8bb8aefd | 802 | ORANGEFS_REQDEVICE_NAME, orangefs_dev_major); |
5db11c21 MM |
803 | return 0; |
804 | } | |
805 | ||
8bb8aefd | 806 | void orangefs_dev_cleanup(void) |
5db11c21 | 807 | { |
8bb8aefd | 808 | unregister_chrdev(orangefs_dev_major, ORANGEFS_REQDEVICE_NAME); |
5db11c21 MM |
809 | gossip_debug(GOSSIP_DEV_DEBUG, |
810 | "*** /dev/%s character device unregistered ***\n", | |
8bb8aefd | 811 | ORANGEFS_REQDEVICE_NAME); |
5db11c21 MM |
812 | } |
813 | ||
8bb8aefd | 814 | static unsigned int orangefs_devreq_poll(struct file *file, |
5db11c21 MM |
815 | struct poll_table_struct *poll_table) |
816 | { | |
817 | int poll_revent_mask = 0; | |
818 | ||
83595db0 | 819 | poll_wait(file, &orangefs_request_list_waitq, poll_table); |
5db11c21 | 820 | |
83595db0 AV |
821 | if (!list_empty(&orangefs_request_list)) |
822 | poll_revent_mask |= POLL_IN; | |
5db11c21 MM |
823 | return poll_revent_mask; |
824 | } | |
825 | ||
8bb8aefd | 826 | const struct file_operations orangefs_devreq_file_operations = { |
5db11c21 | 827 | .owner = THIS_MODULE, |
8bb8aefd YL |
828 | .read = orangefs_devreq_read, |
829 | .write_iter = orangefs_devreq_write_iter, | |
830 | .open = orangefs_devreq_open, | |
831 | .release = orangefs_devreq_release, | |
832 | .unlocked_ioctl = orangefs_devreq_ioctl, | |
5db11c21 MM |
833 | |
834 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
8bb8aefd | 835 | .compat_ioctl = orangefs_devreq_compat_ioctl, |
5db11c21 | 836 | #endif |
8bb8aefd | 837 | .poll = orangefs_devreq_poll |
5db11c21 | 838 | }; |