]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
5db11c21 MM |
2 | /* |
3 | * (C) 2001 Clemson University and The University of Chicago | |
4 | * | |
5 | * Changes by Acxiom Corporation to add protocol version to kernel | |
6 | * communication, Copyright Acxiom Corporation, 2005. | |
7 | * | |
8 | * See COPYING in top-level directory. | |
9 | */ | |
10 | ||
11 | #include "protocol.h" | |
575e9461 MM |
12 | #include "orangefs-kernel.h" |
13 | #include "orangefs-dev-proto.h" | |
14 | #include "orangefs-bufmap.h" | |
44f46410 | 15 | #include "orangefs-debugfs.h" |
5db11c21 MM |
16 | |
17 | #include <linux/debugfs.h> | |
18 | #include <linux/slab.h> | |
19 | ||
20 | /* this file implements the /dev/pvfs2-req device node */ | |
21 | ||
482664dd | 22 | uint32_t orangefs_userspace_version; |
f2ee3b75 | 23 | |
5db11c21 MM |
24 | static int open_access_count; |
25 | ||
a0fe0515 MB |
26 | static DEFINE_MUTEX(devreq_mutex); |
27 | ||
5db11c21 MM |
28 | #define DUMP_DEVICE_ERROR() \ |
29 | do { \ | |
30 | gossip_err("*****************************************************\n");\ | |
8bb8aefd | 31 | gossip_err("ORANGEFS Device Error: You cannot open the device file "); \ |
5db11c21 | 32 | gossip_err("\n/dev/%s more than once. Please make sure that\nthere " \ |
8bb8aefd | 33 | "are no ", ORANGEFS_REQDEVICE_NAME); \ |
5db11c21 MM |
34 | gossip_err("instances of a program using this device\ncurrently " \ |
35 | "running. (You must verify this!)\n"); \ | |
36 | gossip_err("For example, you can use the lsof program as follows:\n");\ | |
37 | gossip_err("'lsof | grep %s' (run this as root)\n", \ | |
8bb8aefd | 38 | ORANGEFS_REQDEVICE_NAME); \ |
5db11c21 MM |
39 | gossip_err(" open_access_count = %d\n", open_access_count); \ |
40 | gossip_err("*****************************************************\n");\ | |
41 | } while (0) | |
42 | ||
43 | static int hash_func(__u64 tag, int table_size) | |
44 | { | |
2c590d5f | 45 | return do_div(tag, (unsigned int)table_size); |
5db11c21 MM |
46 | } |
47 | ||
8bb8aefd | 48 | static void orangefs_devreq_add_op(struct orangefs_kernel_op_s *op) |
5db11c21 MM |
49 | { |
50 | int index = hash_func(op->tag, hash_table_size); | |
51 | ||
1d503617 | 52 | list_add_tail(&op->list, &orangefs_htable_ops_in_progress[index]); |
5db11c21 MM |
53 | } |
54 | ||
ca9f518e MM |
55 | /* |
56 | * find the op with this tag and remove it from the in progress | |
57 | * hash table. | |
58 | */ | |
8bb8aefd | 59 | static struct orangefs_kernel_op_s *orangefs_devreq_remove_op(__u64 tag) |
5db11c21 | 60 | { |
8bb8aefd | 61 | struct orangefs_kernel_op_s *op, *next; |
5db11c21 MM |
62 | int index; |
63 | ||
64 | index = hash_func(tag, hash_table_size); | |
65 | ||
1d503617 | 66 | spin_lock(&orangefs_htable_ops_in_progress_lock); |
5db11c21 MM |
67 | list_for_each_entry_safe(op, |
68 | next, | |
1d503617 | 69 | &orangefs_htable_ops_in_progress[index], |
5db11c21 | 70 | list) { |
05a50a5b AV |
71 | if (op->tag == tag && !op_state_purged(op) && |
72 | !op_state_given_up(op)) { | |
ed42fe05 | 73 | list_del_init(&op->list); |
1d503617 | 74 | spin_unlock(&orangefs_htable_ops_in_progress_lock); |
5db11c21 MM |
75 | return op; |
76 | } | |
77 | } | |
78 | ||
1d503617 | 79 | spin_unlock(&orangefs_htable_ops_in_progress_lock); |
5db11c21 MM |
80 | return NULL; |
81 | } | |
82 | ||
acfcbaf1 MB |
83 | /* Returns whether any FS are still pending remounted */ |
84 | static int mark_all_pending_mounts(void) | |
85 | { | |
86 | int unmounted = 1; | |
87 | struct orangefs_sb_info_s *orangefs_sb = NULL; | |
88 | ||
89 | spin_lock(&orangefs_superblocks_lock); | |
90 | list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) { | |
91 | /* All of these file system require a remount */ | |
92 | orangefs_sb->mount_pending = 1; | |
93 | unmounted = 0; | |
94 | } | |
95 | spin_unlock(&orangefs_superblocks_lock); | |
96 | return unmounted; | |
97 | } | |
98 | ||
99 | /* | |
100 | * Determine if a given file system needs to be remounted or not | |
101 | * Returns -1 on error | |
102 | * 0 if already mounted | |
103 | * 1 if needs remount | |
104 | */ | |
105 | static int fs_mount_pending(__s32 fsid) | |
106 | { | |
107 | int mount_pending = -1; | |
108 | struct orangefs_sb_info_s *orangefs_sb = NULL; | |
109 | ||
110 | spin_lock(&orangefs_superblocks_lock); | |
111 | list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) { | |
112 | if (orangefs_sb->fs_id == fsid) { | |
113 | mount_pending = orangefs_sb->mount_pending; | |
114 | break; | |
115 | } | |
116 | } | |
117 | spin_unlock(&orangefs_superblocks_lock); | |
118 | return mount_pending; | |
119 | } | |
120 | ||
8bb8aefd | 121 | static int orangefs_devreq_open(struct inode *inode, struct file *file) |
5db11c21 MM |
122 | { |
123 | int ret = -EINVAL; | |
124 | ||
78fee0b6 JH |
125 | /* in order to ensure that the filesystem driver sees correct UIDs */ |
126 | if (file->f_cred->user_ns != &init_user_ns) { | |
127 | gossip_err("%s: device cannot be opened outside init_user_ns\n", | |
128 | __func__); | |
129 | goto out; | |
130 | } | |
131 | ||
5db11c21 | 132 | if (!(file->f_flags & O_NONBLOCK)) { |
97f10027 MM |
133 | gossip_err("%s: device cannot be opened in blocking mode\n", |
134 | __func__); | |
5db11c21 MM |
135 | goto out; |
136 | } | |
137 | ret = -EACCES; | |
97f10027 | 138 | gossip_debug(GOSSIP_DEV_DEBUG, "client-core: opening device\n"); |
5db11c21 MM |
139 | mutex_lock(&devreq_mutex); |
140 | ||
141 | if (open_access_count == 0) { | |
fee25ce1 | 142 | open_access_count = 1; |
fb6d2526 | 143 | ret = 0; |
5db11c21 MM |
144 | } else { |
145 | DUMP_DEVICE_ERROR(); | |
146 | } | |
147 | mutex_unlock(&devreq_mutex); | |
148 | ||
149 | out: | |
150 | ||
151 | gossip_debug(GOSSIP_DEV_DEBUG, | |
152 | "pvfs2-client-core: open device complete (ret = %d)\n", | |
153 | ret); | |
154 | return ret; | |
155 | } | |
156 | ||
97f10027 | 157 | /* Function for read() callers into the device */ |
8bb8aefd | 158 | static ssize_t orangefs_devreq_read(struct file *file, |
5db11c21 MM |
159 | char __user *buf, |
160 | size_t count, loff_t *offset) | |
161 | { | |
8bb8aefd YL |
162 | struct orangefs_kernel_op_s *op, *temp; |
163 | __s32 proto_ver = ORANGEFS_KERNEL_PROTO_VERSION; | |
164 | static __s32 magic = ORANGEFS_DEVREQ_MAGIC; | |
a0ec1ded | 165 | struct orangefs_kernel_op_s *cur_op; |
24c8d080 | 166 | unsigned long ret; |
5db11c21 | 167 | |
24c8d080 | 168 | /* We do not support blocking IO. */ |
5db11c21 | 169 | if (!(file->f_flags & O_NONBLOCK)) { |
97f10027 MM |
170 | gossip_err("%s: blocking read from client-core.\n", |
171 | __func__); | |
5db11c21 | 172 | return -EINVAL; |
24c8d080 MB |
173 | } |
174 | ||
175 | /* | |
a762ae6d | 176 | * The client will do an ioctl to find MAX_DEV_REQ_UPSIZE, then |
24c8d080 MB |
177 | * always read with that size buffer. |
178 | */ | |
a762ae6d | 179 | if (count != MAX_DEV_REQ_UPSIZE) { |
24c8d080 MB |
180 | gossip_err("orangefs: client-core tried to read wrong size\n"); |
181 | return -EINVAL; | |
182 | } | |
183 | ||
b7a57cca MB |
184 | /* Check for an empty list before locking. */ |
185 | if (list_empty(&orangefs_request_list)) | |
186 | return -EAGAIN; | |
187 | ||
ed42fe05 | 188 | restart: |
a0ec1ded | 189 | cur_op = NULL; |
24c8d080 | 190 | /* Get next op (if any) from top of list. */ |
8bb8aefd YL |
191 | spin_lock(&orangefs_request_list_lock); |
192 | list_for_each_entry_safe(op, temp, &orangefs_request_list, list) { | |
24c8d080 MB |
193 | __s32 fsid; |
194 | /* This lock is held past the end of the loop when we break. */ | |
195 | spin_lock(&op->lock); | |
05a50a5b | 196 | if (unlikely(op_state_purged(op) || op_state_given_up(op))) { |
ed42fe05 AV |
197 | spin_unlock(&op->lock); |
198 | continue; | |
199 | } | |
24c8d080 MB |
200 | |
201 | fsid = fsid_of_op(op); | |
8bb8aefd | 202 | if (fsid != ORANGEFS_FS_ID_NULL) { |
24c8d080 MB |
203 | int ret; |
204 | /* Skip ops whose filesystem needs to be mounted. */ | |
205 | ret = fs_mount_pending(fsid); | |
206 | if (ret == 1) { | |
5db11c21 | 207 | gossip_debug(GOSSIP_DEV_DEBUG, |
5090c967 MM |
208 | "%s: mount pending, skipping op tag " |
209 | "%llu %s\n", | |
210 | __func__, | |
211 | llu(op->tag), | |
212 | get_opname_string(op)); | |
24c8d080 MB |
213 | spin_unlock(&op->lock); |
214 | continue; | |
97f10027 MM |
215 | /* |
216 | * Skip ops whose filesystem we don't know about unless | |
1ec1688c MB |
217 | * it is being mounted or unmounted. It is possible for |
218 | * a filesystem we don't know about to be unmounted if | |
219 | * it fails to mount in the kernel after userspace has | |
220 | * been sent the mount request. | |
97f10027 | 221 | */ |
24c8d080 MB |
222 | /* XXX: is there a better way to detect this? */ |
223 | } else if (ret == -1 && | |
97f10027 MM |
224 | !(op->upcall.type == |
225 | ORANGEFS_VFS_OP_FS_MOUNT || | |
226 | op->upcall.type == | |
1ec1688c MB |
227 | ORANGEFS_VFS_OP_GETATTR || |
228 | op->upcall.type == | |
229 | ORANGEFS_VFS_OP_FS_UMOUNT)) { | |
24c8d080 MB |
230 | gossip_debug(GOSSIP_DEV_DEBUG, |
231 | "orangefs: skipping op tag %llu %s\n", | |
232 | llu(op->tag), get_opname_string(op)); | |
233 | gossip_err( | |
234 | "orangefs: ERROR: fs_mount_pending %d\n", | |
235 | fsid); | |
236 | spin_unlock(&op->lock); | |
5db11c21 | 237 | continue; |
5db11c21 MM |
238 | } |
239 | } | |
24c8d080 MB |
240 | /* |
241 | * Either this op does not pertain to a filesystem, is mounting | |
242 | * a filesystem, or pertains to a mounted filesystem. Let it | |
243 | * through. | |
244 | */ | |
245 | cur_op = op; | |
246 | break; | |
247 | } | |
248 | ||
249 | /* | |
250 | * At this point we either have a valid op and can continue or have not | |
251 | * found an op and must ask the client to try again later. | |
252 | */ | |
253 | if (!cur_op) { | |
8bb8aefd | 254 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 255 | return -EAGAIN; |
5db11c21 MM |
256 | } |
257 | ||
ca9f518e MM |
258 | gossip_debug(GOSSIP_DEV_DEBUG, "%s: reading op tag %llu %s\n", |
259 | __func__, | |
260 | llu(cur_op->tag), | |
261 | get_opname_string(cur_op)); | |
5db11c21 | 262 | |
24c8d080 MB |
263 | /* |
264 | * Such an op should never be on the list in the first place. If so, we | |
265 | * will abort. | |
266 | */ | |
267 | if (op_state_in_progress(cur_op) || op_state_serviced(cur_op)) { | |
268 | gossip_err("orangefs: ERROR: Current op already queued.\n"); | |
05a50a5b | 269 | list_del_init(&cur_op->list); |
5db11c21 | 270 | spin_unlock(&cur_op->lock); |
8bb8aefd | 271 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 272 | return -EAGAIN; |
5db11c21 | 273 | } |
ca9f518e | 274 | |
ed42fe05 | 275 | list_del_init(&cur_op->list); |
8bb8aefd | 276 | spin_unlock(&orangefs_request_list_lock); |
ed42fe05 | 277 | |
24c8d080 MB |
278 | spin_unlock(&cur_op->lock); |
279 | ||
280 | /* Push the upcall out. */ | |
281 | ret = copy_to_user(buf, &proto_ver, sizeof(__s32)); | |
282 | if (ret != 0) | |
283 | goto error; | |
284 | ret = copy_to_user(buf+sizeof(__s32), &magic, sizeof(__s32)); | |
285 | if (ret != 0) | |
286 | goto error; | |
287 | ret = copy_to_user(buf+2 * sizeof(__s32), &cur_op->tag, sizeof(__u64)); | |
288 | if (ret != 0) | |
289 | goto error; | |
290 | ret = copy_to_user(buf+2*sizeof(__s32)+sizeof(__u64), &cur_op->upcall, | |
8bb8aefd | 291 | sizeof(struct orangefs_upcall_s)); |
24c8d080 MB |
292 | if (ret != 0) |
293 | goto error; | |
294 | ||
1d503617 | 295 | spin_lock(&orangefs_htable_ops_in_progress_lock); |
ed42fe05 AV |
296 | spin_lock(&cur_op->lock); |
297 | if (unlikely(op_state_given_up(cur_op))) { | |
298 | spin_unlock(&cur_op->lock); | |
1d503617 | 299 | spin_unlock(&orangefs_htable_ops_in_progress_lock); |
05a50a5b | 300 | complete(&cur_op->waitq); |
ed42fe05 AV |
301 | goto restart; |
302 | } | |
303 | ||
304 | /* | |
305 | * Set the operation to be in progress and move it between lists since | |
306 | * it has been sent to the client. | |
307 | */ | |
308 | set_op_state_inprogress(cur_op); | |
9d9e7ba9 MM |
309 | gossip_debug(GOSSIP_DEV_DEBUG, |
310 | "%s: 1 op:%s: op_state:%d: process:%s:\n", | |
311 | __func__, | |
312 | get_opname_string(cur_op), | |
313 | cur_op->op_state, | |
314 | current->comm); | |
ed42fe05 AV |
315 | orangefs_devreq_add_op(cur_op); |
316 | spin_unlock(&cur_op->lock); | |
1d503617 | 317 | spin_unlock(&orangefs_htable_ops_in_progress_lock); |
ed42fe05 | 318 | |
24c8d080 | 319 | /* The client only asks to read one size buffer. */ |
a762ae6d | 320 | return MAX_DEV_REQ_UPSIZE; |
24c8d080 MB |
321 | error: |
322 | /* | |
323 | * We were unable to copy the op data to the client. Put the op back in | |
324 | * list. If client has crashed, the op will be purged later when the | |
325 | * device is released. | |
326 | */ | |
327 | gossip_err("orangefs: Failed to copy data to user space\n"); | |
8bb8aefd | 328 | spin_lock(&orangefs_request_list_lock); |
24c8d080 | 329 | spin_lock(&cur_op->lock); |
ed42fe05 AV |
330 | if (likely(!op_state_given_up(cur_op))) { |
331 | set_op_state_waiting(cur_op); | |
9d9e7ba9 MM |
332 | gossip_debug(GOSSIP_DEV_DEBUG, |
333 | "%s: 2 op:%s: op_state:%d: process:%s:\n", | |
334 | __func__, | |
335 | get_opname_string(cur_op), | |
336 | cur_op->op_state, | |
337 | current->comm); | |
ed42fe05 | 338 | list_add(&cur_op->list, &orangefs_request_list); |
05a50a5b AV |
339 | spin_unlock(&cur_op->lock); |
340 | } else { | |
341 | spin_unlock(&cur_op->lock); | |
342 | complete(&cur_op->waitq); | |
ed42fe05 | 343 | } |
8bb8aefd | 344 | spin_unlock(&orangefs_request_list_lock); |
24c8d080 | 345 | return -EFAULT; |
5db11c21 MM |
346 | } |
347 | ||
97f10027 | 348 | /* |
b3ae4755 MM |
349 | * Function for writev() callers into the device. |
350 | * | |
351 | * Userspace should have written: | |
352 | * - __u32 version | |
353 | * - __u32 magic | |
354 | * - __u64 tag | |
355 | * - struct orangefs_downcall_s | |
356 | * - trailer buffer (in the case of READDIR operations) | |
97f10027 | 357 | */ |
b3ae4755 MM |
358 | static ssize_t orangefs_devreq_write_iter(struct kiocb *iocb, |
359 | struct iov_iter *iter) | |
5db11c21 | 360 | { |
b3ae4755 | 361 | ssize_t ret; |
8bb8aefd | 362 | struct orangefs_kernel_op_s *op = NULL; |
b3ae4755 MM |
363 | struct { |
364 | __u32 version; | |
365 | __u32 magic; | |
366 | __u64 tag; | |
367 | } head; | |
368 | int total = ret = iov_iter_count(iter); | |
b3ae4755 MM |
369 | int downcall_size = sizeof(struct orangefs_downcall_s); |
370 | int head_size = sizeof(head); | |
371 | ||
372 | gossip_debug(GOSSIP_DEV_DEBUG, "%s: total:%d: ret:%zd:\n", | |
373 | __func__, | |
374 | total, | |
375 | ret); | |
5db11c21 | 376 | |
b3ae4755 | 377 | if (total < MAX_DEV_REQ_DOWNSIZE) { |
cf0c2771 | 378 | gossip_err("%s: total:%d: must be at least:%u:\n", |
b3ae4755 MM |
379 | __func__, |
380 | total, | |
cf0c2771 | 381 | (unsigned int) MAX_DEV_REQ_DOWNSIZE); |
ed42fe05 | 382 | return -EFAULT; |
5db11c21 | 383 | } |
b3ae4755 | 384 | |
cbbd26b8 | 385 | if (!copy_from_iter_full(&head, head_size, iter)) { |
b3ae4755 | 386 | gossip_err("%s: failed to copy head.\n", __func__); |
ed42fe05 | 387 | return -EFAULT; |
97f10027 | 388 | } |
b3ae4755 MM |
389 | |
390 | if (head.version < ORANGEFS_MINIMUM_USERSPACE_VERSION) { | |
391 | gossip_err("%s: userspace claims version" | |
392 | "%d, minimum version required: %d.\n", | |
393 | __func__, | |
394 | head.version, | |
395 | ORANGEFS_MINIMUM_USERSPACE_VERSION); | |
ed42fe05 | 396 | return -EPROTO; |
5db11c21 | 397 | } |
5db11c21 | 398 | |
b3ae4755 MM |
399 | if (head.magic != ORANGEFS_DEVREQ_MAGIC) { |
400 | gossip_err("Error: Device magic number does not match.\n"); | |
ed42fe05 | 401 | return -EPROTO; |
b3ae4755 | 402 | } |
5db11c21 | 403 | |
482664dd MB |
404 | if (!orangefs_userspace_version) { |
405 | orangefs_userspace_version = head.version; | |
406 | } else if (orangefs_userspace_version != head.version) { | |
f2ee3b75 MB |
407 | gossip_err("Error: userspace version changes\n"); |
408 | return -EPROTO; | |
409 | } | |
410 | ||
ca9f518e | 411 | /* remove the op from the in progress hash table */ |
b3ae4755 MM |
412 | op = orangefs_devreq_remove_op(head.tag); |
413 | if (!op) { | |
05973c2e MM |
414 | gossip_debug(GOSSIP_DEV_DEBUG, |
415 | "%s: No one's waiting for tag %llu\n", | |
416 | __func__, llu(head.tag)); | |
ed42fe05 | 417 | return ret; |
b3ae4755 | 418 | } |
5db11c21 | 419 | |
cbbd26b8 | 420 | if (!copy_from_iter_full(&op->downcall, downcall_size, iter)) { |
b3ae4755 | 421 | gossip_err("%s: failed to copy downcall.\n", __func__); |
5964c1b8 | 422 | goto Efault; |
5db11c21 MM |
423 | } |
424 | ||
b3ae4755 MM |
425 | if (op->downcall.status) |
426 | goto wakeup; | |
97f10027 | 427 | |
b3ae4755 MM |
428 | /* |
429 | * We've successfully peeled off the head and the downcall. | |
430 | * Something has gone awry if total doesn't equal the | |
431 | * sum of head_size, downcall_size and trailer_size. | |
432 | */ | |
433 | if ((head_size + downcall_size + op->downcall.trailer_size) != total) { | |
434 | gossip_err("%s: funky write, head_size:%d" | |
435 | ": downcall_size:%d: trailer_size:%lld" | |
436 | ": total size:%d:\n", | |
437 | __func__, | |
438 | head_size, | |
439 | downcall_size, | |
440 | op->downcall.trailer_size, | |
441 | total); | |
5964c1b8 | 442 | goto Efault; |
b3ae4755 | 443 | } |
97f10027 | 444 | |
b3ae4755 MM |
445 | /* Only READDIR operations should have trailers. */ |
446 | if ((op->downcall.type != ORANGEFS_VFS_OP_READDIR) && | |
447 | (op->downcall.trailer_size != 0)) { | |
448 | gossip_err("%s: %x operation with trailer.", | |
449 | __func__, | |
450 | op->downcall.type); | |
5964c1b8 | 451 | goto Efault; |
b3ae4755 | 452 | } |
97f10027 | 453 | |
b3ae4755 MM |
454 | /* READDIR operations should always have trailers. */ |
455 | if ((op->downcall.type == ORANGEFS_VFS_OP_READDIR) && | |
456 | (op->downcall.trailer_size == 0)) { | |
457 | gossip_err("%s: %x operation with no trailer.", | |
458 | __func__, | |
459 | op->downcall.type); | |
5964c1b8 | 460 | goto Efault; |
b3ae4755 | 461 | } |
97f10027 | 462 | |
b3ae4755 MM |
463 | if (op->downcall.type != ORANGEFS_VFS_OP_READDIR) |
464 | goto wakeup; | |
5db11c21 | 465 | |
07a25853 ME |
466 | op->downcall.trailer_buf = vmalloc(op->downcall.trailer_size); |
467 | if (!op->downcall.trailer_buf) | |
5964c1b8 | 468 | goto Enomem; |
07a25853 | 469 | |
b3ae4755 | 470 | memset(op->downcall.trailer_buf, 0, op->downcall.trailer_size); |
cbbd26b8 AV |
471 | if (!copy_from_iter_full(op->downcall.trailer_buf, |
472 | op->downcall.trailer_size, iter)) { | |
b3ae4755 MM |
473 | gossip_err("%s: failed to copy trailer.\n", __func__); |
474 | vfree(op->downcall.trailer_buf); | |
5964c1b8 | 475 | goto Efault; |
b3ae4755 | 476 | } |
97f10027 | 477 | |
b3ae4755 | 478 | wakeup: |
2a9e5c22 | 479 | /* |
9f08cfe9 MM |
480 | * Return to vfs waitqueue, and back to service_operation |
481 | * through wait_for_matching_downcall. | |
2a9e5c22 AV |
482 | */ |
483 | spin_lock(&op->lock); | |
5964c1b8 | 484 | if (unlikely(op_is_cancel(op))) { |
2a9e5c22 | 485 | spin_unlock(&op->lock); |
78699e29 | 486 | put_cancel(op); |
5964c1b8 AV |
487 | } else if (unlikely(op_state_given_up(op))) { |
488 | spin_unlock(&op->lock); | |
05a50a5b | 489 | complete(&op->waitq); |
5964c1b8 AV |
490 | } else { |
491 | set_op_state_serviced(op); | |
9d9e7ba9 MM |
492 | gossip_debug(GOSSIP_DEV_DEBUG, |
493 | "%s: op:%s: op_state:%d: process:%s:\n", | |
494 | __func__, | |
495 | get_opname_string(op), | |
496 | op->op_state, | |
497 | current->comm); | |
5964c1b8 AV |
498 | spin_unlock(&op->lock); |
499 | } | |
b3ae4755 | 500 | return ret; |
ed42fe05 | 501 | |
5964c1b8 AV |
502 | Efault: |
503 | op->downcall.status = -(ORANGEFS_ERROR_BIT | 9); | |
504 | ret = -EFAULT; | |
505 | goto wakeup; | |
506 | ||
507 | Enomem: | |
508 | op->downcall.status = -(ORANGEFS_ERROR_BIT | 8); | |
509 | ret = -ENOMEM; | |
510 | goto wakeup; | |
5db11c21 MM |
511 | } |
512 | ||
5db11c21 MM |
513 | /* |
514 | * NOTE: gets called when the last reference to this device is dropped. | |
515 | * Using the open_access_count variable, we enforce a reference count | |
516 | * on this file so that it can be opened by only one process at a time. | |
517 | * the devreq_mutex is used to make sure all i/o has completed | |
8bb8aefd | 518 | * before we call orangefs_bufmap_finalize, and similar such tricky |
5db11c21 MM |
519 | * situations |
520 | */ | |
8bb8aefd | 521 | static int orangefs_devreq_release(struct inode *inode, struct file *file) |
5db11c21 MM |
522 | { |
523 | int unmounted = 0; | |
524 | ||
525 | gossip_debug(GOSSIP_DEV_DEBUG, | |
526 | "%s:pvfs2-client-core: exiting, closing device\n", | |
527 | __func__); | |
528 | ||
529 | mutex_lock(&devreq_mutex); | |
ea2c9c9f | 530 | orangefs_bufmap_finalize(); |
5db11c21 | 531 | |
fee25ce1 | 532 | open_access_count = -1; |
5db11c21 MM |
533 | |
534 | unmounted = mark_all_pending_mounts(); | |
8bb8aefd | 535 | gossip_debug(GOSSIP_DEV_DEBUG, "ORANGEFS Device Close: Filesystem(s) %s\n", |
5db11c21 | 536 | (unmounted ? "UNMOUNTED" : "MOUNTED")); |
5db11c21 | 537 | |
5db11c21 | 538 | purge_waiting_ops(); |
5db11c21 | 539 | purge_inprogress_ops(); |
ea2c9c9f AV |
540 | |
541 | orangefs_bufmap_run_down(); | |
542 | ||
5db11c21 MM |
543 | gossip_debug(GOSSIP_DEV_DEBUG, |
544 | "pvfs2-client-core: device close complete\n"); | |
fee25ce1 | 545 | open_access_count = 0; |
482664dd | 546 | orangefs_userspace_version = 0; |
fee25ce1 | 547 | mutex_unlock(&devreq_mutex); |
5db11c21 MM |
548 | return 0; |
549 | } | |
550 | ||
551 | int is_daemon_in_service(void) | |
552 | { | |
553 | int in_service; | |
554 | ||
555 | /* | |
556 | * What this function does is checks if client-core is alive | |
557 | * based on the access count we maintain on the device. | |
558 | */ | |
559 | mutex_lock(&devreq_mutex); | |
560 | in_service = open_access_count == 1 ? 0 : -EIO; | |
561 | mutex_unlock(&devreq_mutex); | |
562 | return in_service; | |
563 | } | |
564 | ||
78699e29 AV |
565 | bool __is_daemon_in_service(void) |
566 | { | |
567 | return open_access_count == 1; | |
568 | } | |
569 | ||
5db11c21 MM |
570 | static inline long check_ioctl_command(unsigned int command) |
571 | { | |
572 | /* Check for valid ioctl codes */ | |
8bb8aefd | 573 | if (_IOC_TYPE(command) != ORANGEFS_DEV_MAGIC) { |
5db11c21 MM |
574 | gossip_err("device ioctl magic numbers don't match! Did you rebuild pvfs2-client-core/libpvfs2? [cmd %x, magic %x != %x]\n", |
575 | command, | |
576 | _IOC_TYPE(command), | |
8bb8aefd | 577 | ORANGEFS_DEV_MAGIC); |
5db11c21 MM |
578 | return -EINVAL; |
579 | } | |
580 | /* and valid ioctl commands */ | |
8bb8aefd | 581 | if (_IOC_NR(command) >= ORANGEFS_DEV_MAXNR || _IOC_NR(command) <= 0) { |
5db11c21 | 582 | gossip_err("Invalid ioctl command number [%d >= %d]\n", |
8bb8aefd | 583 | _IOC_NR(command), ORANGEFS_DEV_MAXNR); |
5db11c21 MM |
584 | return -ENOIOCTLCMD; |
585 | } | |
586 | return 0; | |
587 | } | |
588 | ||
589 | static long dispatch_ioctl_command(unsigned int command, unsigned long arg) | |
590 | { | |
8bb8aefd | 591 | static __s32 magic = ORANGEFS_DEVREQ_MAGIC; |
a762ae6d MB |
592 | static __s32 max_up_size = MAX_DEV_REQ_UPSIZE; |
593 | static __s32 max_down_size = MAX_DEV_REQ_DOWNSIZE; | |
8bb8aefd | 594 | struct ORANGEFS_dev_map_desc user_desc; |
5db11c21 | 595 | int ret = 0; |
5db11c21 | 596 | int upstream_kmod = 1; |
45996492 | 597 | struct orangefs_sb_info_s *orangefs_sb; |
5db11c21 MM |
598 | |
599 | /* mtmoore: add locking here */ | |
600 | ||
601 | switch (command) { | |
8bb8aefd | 602 | case ORANGEFS_DEV_GET_MAGIC: |
5db11c21 MM |
603 | return ((put_user(magic, (__s32 __user *) arg) == -EFAULT) ? |
604 | -EIO : | |
605 | 0); | |
8bb8aefd | 606 | case ORANGEFS_DEV_GET_MAX_UPSIZE: |
5db11c21 MM |
607 | return ((put_user(max_up_size, |
608 | (__s32 __user *) arg) == -EFAULT) ? | |
609 | -EIO : | |
610 | 0); | |
8bb8aefd | 611 | case ORANGEFS_DEV_GET_MAX_DOWNSIZE: |
5db11c21 MM |
612 | return ((put_user(max_down_size, |
613 | (__s32 __user *) arg) == -EFAULT) ? | |
614 | -EIO : | |
615 | 0); | |
8bb8aefd | 616 | case ORANGEFS_DEV_MAP: |
5db11c21 | 617 | ret = copy_from_user(&user_desc, |
8bb8aefd | 618 | (struct ORANGEFS_dev_map_desc __user *) |
5db11c21 | 619 | arg, |
8bb8aefd | 620 | sizeof(struct ORANGEFS_dev_map_desc)); |
ea2c9c9f AV |
621 | /* WTF -EIO and not -EFAULT? */ |
622 | return ret ? -EIO : orangefs_bufmap_initialize(&user_desc); | |
8bb8aefd | 623 | case ORANGEFS_DEV_REMOUNT_ALL: |
5db11c21 | 624 | gossip_debug(GOSSIP_DEV_DEBUG, |
97f10027 MM |
625 | "%s: got ORANGEFS_DEV_REMOUNT_ALL\n", |
626 | __func__); | |
5db11c21 MM |
627 | |
628 | /* | |
8bb8aefd | 629 | * remount all mounted orangefs volumes to regain the lost |
5db11c21 MM |
630 | * dynamic mount tables (if any) -- NOTE: this is done |
631 | * without keeping the superblock list locked due to the | |
adcf34a2 | 632 | * upcall/downcall waiting. also, the request mutex is |
5db11c21 MM |
633 | * used to ensure that no operations will be serviced until |
634 | * all of the remounts are serviced (to avoid ops between | |
635 | * mounts to fail) | |
636 | */ | |
1d503617 | 637 | ret = mutex_lock_interruptible(&orangefs_request_mutex); |
5db11c21 MM |
638 | if (ret < 0) |
639 | return ret; | |
640 | gossip_debug(GOSSIP_DEV_DEBUG, | |
97f10027 MM |
641 | "%s: priority remount in progress\n", |
642 | __func__); | |
45996492 AV |
643 | spin_lock(&orangefs_superblocks_lock); |
644 | list_for_each_entry(orangefs_sb, &orangefs_superblocks, list) { | |
645 | /* | |
646 | * We have to drop the spinlock, so entries can be | |
647 | * removed. They can't be freed, though, so we just | |
648 | * keep the forward pointers and zero the back ones - | |
649 | * that way we can get to the rest of the list. | |
650 | */ | |
651 | if (!orangefs_sb->list.prev) | |
652 | continue; | |
653 | gossip_debug(GOSSIP_DEV_DEBUG, | |
654 | "%s: Remounting SB %p\n", | |
655 | __func__, | |
656 | orangefs_sb); | |
657 | ||
658 | spin_unlock(&orangefs_superblocks_lock); | |
659 | ret = orangefs_remount(orangefs_sb); | |
660 | spin_lock(&orangefs_superblocks_lock); | |
661 | if (ret) { | |
5db11c21 | 662 | gossip_debug(GOSSIP_DEV_DEBUG, |
45996492 | 663 | "SB %p remount failed\n", |
8bb8aefd | 664 | orangefs_sb); |
45996492 | 665 | break; |
5db11c21 MM |
666 | } |
667 | } | |
45996492 | 668 | spin_unlock(&orangefs_superblocks_lock); |
5db11c21 | 669 | gossip_debug(GOSSIP_DEV_DEBUG, |
97f10027 MM |
670 | "%s: priority remount complete\n", |
671 | __func__); | |
1d503617 | 672 | mutex_unlock(&orangefs_request_mutex); |
5db11c21 MM |
673 | return ret; |
674 | ||
8bb8aefd | 675 | case ORANGEFS_DEV_UPSTREAM: |
5db11c21 MM |
676 | ret = copy_to_user((void __user *)arg, |
677 | &upstream_kmod, | |
678 | sizeof(upstream_kmod)); | |
679 | ||
680 | if (ret != 0) | |
681 | return -EIO; | |
682 | else | |
683 | return ret; | |
684 | ||
8bb8aefd | 685 | case ORANGEFS_DEV_CLIENT_MASK: |
44f46410 | 686 | return orangefs_debugfs_new_client_mask((void __user *)arg); |
8bb8aefd | 687 | case ORANGEFS_DEV_CLIENT_STRING: |
44f46410 | 688 | return orangefs_debugfs_new_client_string((void __user *)arg); |
8bb8aefd | 689 | case ORANGEFS_DEV_DEBUG: |
44f46410 | 690 | return orangefs_debugfs_new_debug((void __user *)arg); |
5db11c21 MM |
691 | default: |
692 | return -ENOIOCTLCMD; | |
693 | } | |
694 | return -ENOIOCTLCMD; | |
695 | } | |
696 | ||
8bb8aefd | 697 | static long orangefs_devreq_ioctl(struct file *file, |
5db11c21 MM |
698 | unsigned int command, unsigned long arg) |
699 | { | |
700 | long ret; | |
701 | ||
702 | /* Check for properly constructed commands */ | |
703 | ret = check_ioctl_command(command); | |
704 | if (ret < 0) | |
705 | return (int)ret; | |
706 | ||
707 | return (int)dispatch_ioctl_command(command, arg); | |
708 | } | |
709 | ||
710 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
711 | ||
8bb8aefd YL |
712 | /* Compat structure for the ORANGEFS_DEV_MAP ioctl */ |
713 | struct ORANGEFS_dev_map_desc32 { | |
5db11c21 MM |
714 | compat_uptr_t ptr; |
715 | __s32 total_size; | |
716 | __s32 size; | |
717 | __s32 count; | |
718 | }; | |
719 | ||
720 | static unsigned long translate_dev_map26(unsigned long args, long *error) | |
721 | { | |
8bb8aefd | 722 | struct ORANGEFS_dev_map_desc32 __user *p32 = (void __user *)args; |
5db11c21 MM |
723 | /* |
724 | * Depending on the architecture, allocate some space on the | |
725 | * user-call-stack based on our expected layout. | |
726 | */ | |
8bb8aefd | 727 | struct ORANGEFS_dev_map_desc __user *p = |
5db11c21 | 728 | compat_alloc_user_space(sizeof(*p)); |
84d02150 | 729 | compat_uptr_t addr; |
5db11c21 MM |
730 | |
731 | *error = 0; | |
732 | /* get the ptr from the 32 bit user-space */ | |
733 | if (get_user(addr, &p32->ptr)) | |
734 | goto err; | |
735 | /* try to put that into a 64-bit layout */ | |
736 | if (put_user(compat_ptr(addr), &p->ptr)) | |
737 | goto err; | |
738 | /* copy the remaining fields */ | |
739 | if (copy_in_user(&p->total_size, &p32->total_size, sizeof(__s32))) | |
740 | goto err; | |
741 | if (copy_in_user(&p->size, &p32->size, sizeof(__s32))) | |
742 | goto err; | |
743 | if (copy_in_user(&p->count, &p32->count, sizeof(__s32))) | |
744 | goto err; | |
745 | return (unsigned long)p; | |
746 | err: | |
747 | *error = -EFAULT; | |
748 | return 0; | |
749 | } | |
750 | ||
751 | /* | |
752 | * 32 bit user-space apps' ioctl handlers when kernel modules | |
753 | * is compiled as a 64 bit one | |
754 | */ | |
8bb8aefd | 755 | static long orangefs_devreq_compat_ioctl(struct file *filp, unsigned int cmd, |
5db11c21 MM |
756 | unsigned long args) |
757 | { | |
758 | long ret; | |
759 | unsigned long arg = args; | |
760 | ||
761 | /* Check for properly constructed commands */ | |
762 | ret = check_ioctl_command(cmd); | |
763 | if (ret < 0) | |
764 | return ret; | |
8bb8aefd | 765 | if (cmd == ORANGEFS_DEV_MAP) { |
5db11c21 MM |
766 | /* |
767 | * convert the arguments to what we expect internally | |
768 | * in kernel space | |
769 | */ | |
770 | arg = translate_dev_map26(args, &ret); | |
771 | if (ret < 0) { | |
772 | gossip_err("Could not translate dev map\n"); | |
773 | return ret; | |
774 | } | |
775 | } | |
776 | /* no other ioctl requires translation */ | |
777 | return dispatch_ioctl_command(cmd, arg); | |
778 | } | |
779 | ||
2c590d5f MM |
780 | #endif /* CONFIG_COMPAT is in .config */ |
781 | ||
5db11c21 | 782 | /* the assigned character device major number */ |
8bb8aefd | 783 | static int orangefs_dev_major; |
5db11c21 MM |
784 | |
785 | /* | |
8bb8aefd | 786 | * Initialize orangefs device specific state: |
5db11c21 MM |
787 | * Must be called at module load time only |
788 | */ | |
8bb8aefd | 789 | int orangefs_dev_init(void) |
5db11c21 | 790 | { |
8bb8aefd YL |
791 | /* register orangefs-req device */ |
792 | orangefs_dev_major = register_chrdev(0, | |
793 | ORANGEFS_REQDEVICE_NAME, | |
794 | &orangefs_devreq_file_operations); | |
795 | if (orangefs_dev_major < 0) { | |
5db11c21 MM |
796 | gossip_debug(GOSSIP_DEV_DEBUG, |
797 | "Failed to register /dev/%s (error %d)\n", | |
8bb8aefd | 798 | ORANGEFS_REQDEVICE_NAME, orangefs_dev_major); |
8bb8aefd | 799 | return orangefs_dev_major; |
5db11c21 MM |
800 | } |
801 | ||
802 | gossip_debug(GOSSIP_DEV_DEBUG, | |
803 | "*** /dev/%s character device registered ***\n", | |
8bb8aefd | 804 | ORANGEFS_REQDEVICE_NAME); |
5db11c21 | 805 | gossip_debug(GOSSIP_DEV_DEBUG, "'mknod /dev/%s c %d 0'.\n", |
8bb8aefd | 806 | ORANGEFS_REQDEVICE_NAME, orangefs_dev_major); |
5db11c21 MM |
807 | return 0; |
808 | } | |
809 | ||
8bb8aefd | 810 | void orangefs_dev_cleanup(void) |
5db11c21 | 811 | { |
8bb8aefd | 812 | unregister_chrdev(orangefs_dev_major, ORANGEFS_REQDEVICE_NAME); |
5db11c21 MM |
813 | gossip_debug(GOSSIP_DEV_DEBUG, |
814 | "*** /dev/%s character device unregistered ***\n", | |
8bb8aefd | 815 | ORANGEFS_REQDEVICE_NAME); |
5db11c21 MM |
816 | } |
817 | ||
076ccb76 | 818 | static __poll_t orangefs_devreq_poll(struct file *file, |
5db11c21 MM |
819 | struct poll_table_struct *poll_table) |
820 | { | |
076ccb76 | 821 | __poll_t poll_revent_mask = 0; |
5db11c21 | 822 | |
83595db0 | 823 | poll_wait(file, &orangefs_request_list_waitq, poll_table); |
5db11c21 | 824 | |
83595db0 | 825 | if (!list_empty(&orangefs_request_list)) |
e410c603 | 826 | poll_revent_mask |= POLLIN; |
5db11c21 MM |
827 | return poll_revent_mask; |
828 | } | |
829 | ||
8bb8aefd | 830 | const struct file_operations orangefs_devreq_file_operations = { |
5db11c21 | 831 | .owner = THIS_MODULE, |
8bb8aefd YL |
832 | .read = orangefs_devreq_read, |
833 | .write_iter = orangefs_devreq_write_iter, | |
834 | .open = orangefs_devreq_open, | |
835 | .release = orangefs_devreq_release, | |
836 | .unlocked_ioctl = orangefs_devreq_ioctl, | |
5db11c21 MM |
837 | |
838 | #ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ | |
8bb8aefd | 839 | .compat_ioctl = orangefs_devreq_compat_ioctl, |
5db11c21 | 840 | #endif |
8bb8aefd | 841 | .poll = orangefs_devreq_poll |
5db11c21 | 842 | }; |