projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Orangefs: Don't opencode memcpy.
[linux-block.git] / fs / orangefs / pvfs2-utils.c
CommitLineData
f7be4ee0
MM		 1/*
2 * (C) 2001 Clemson University and The University of Chicago
3 *
4 * See COPYING in top-level directory.
5 */
6#include "protocol.h"
7#include "pvfs2-kernel.h"
8#include "pvfs2-dev-proto.h"
9#include "pvfs2-bufmap.h"
10
11__s32 fsid_of_op(struct pvfs2_kernel_op_s *op)
12{
13 __s32 fsid = PVFS_FS_ID_NULL;
14
15 if (op) {
16 switch (op->upcall.type) {
17 case PVFS2_VFS_OP_FILE_IO:
18 fsid = op->upcall.req.io.refn.fs_id;
19 break;
20 case PVFS2_VFS_OP_LOOKUP:
21 fsid = op->upcall.req.lookup.parent_refn.fs_id;
22 break;
23 case PVFS2_VFS_OP_CREATE:
24 fsid = op->upcall.req.create.parent_refn.fs_id;
25 break;
26 case PVFS2_VFS_OP_GETATTR:
27 fsid = op->upcall.req.getattr.refn.fs_id;
28 break;
29 case PVFS2_VFS_OP_REMOVE:
30 fsid = op->upcall.req.remove.parent_refn.fs_id;
31 break;
32 case PVFS2_VFS_OP_MKDIR:
33 fsid = op->upcall.req.mkdir.parent_refn.fs_id;
34 break;
35 case PVFS2_VFS_OP_READDIR:
36 fsid = op->upcall.req.readdir.refn.fs_id;
37 break;
38 case PVFS2_VFS_OP_SETATTR:
39 fsid = op->upcall.req.setattr.refn.fs_id;
40 break;
41 case PVFS2_VFS_OP_SYMLINK:
42 fsid = op->upcall.req.sym.parent_refn.fs_id;
43 break;
44 case PVFS2_VFS_OP_RENAME:
45 fsid = op->upcall.req.rename.old_parent_refn.fs_id;
46 break;
47 case PVFS2_VFS_OP_STATFS:
48 fsid = op->upcall.req.statfs.fs_id;
49 break;
50 case PVFS2_VFS_OP_TRUNCATE:
51 fsid = op->upcall.req.truncate.refn.fs_id;
52 break;
53 case PVFS2_VFS_OP_MMAP_RA_FLUSH:
54 fsid = op->upcall.req.ra_cache_flush.refn.fs_id;
55 break;
56 case PVFS2_VFS_OP_FS_UMOUNT:
57 fsid = op->upcall.req.fs_umount.fs_id;
58 break;
59 case PVFS2_VFS_OP_GETXATTR:
60 fsid = op->upcall.req.getxattr.refn.fs_id;
61 break;
62 case PVFS2_VFS_OP_SETXATTR:
63 fsid = op->upcall.req.setxattr.refn.fs_id;
64 break;
65 case PVFS2_VFS_OP_LISTXATTR:
66 fsid = op->upcall.req.listxattr.refn.fs_id;
67 break;
68 case PVFS2_VFS_OP_REMOVEXATTR:
69 fsid = op->upcall.req.removexattr.refn.fs_id;
70 break;
71 case PVFS2_VFS_OP_FSYNC:
72 fsid = op->upcall.req.fsync.refn.fs_id;
73 break;
74 default:
75 break;
76 }
77 }
78 return fsid;
79}
80
81static void pvfs2_set_inode_flags(struct inode *inode,
82 struct PVFS_sys_attr_s *attrs)
83{
84 if (attrs->flags & PVFS_IMMUTABLE_FL)
85 inode->i_flags |= S_IMMUTABLE;
86 else
87 inode->i_flags &= ~S_IMMUTABLE;
88
89 if (attrs->flags & PVFS_APPEND_FL)
90 inode->i_flags |= S_APPEND;
91 else
92 inode->i_flags &= ~S_APPEND;
93
94 if (attrs->flags & PVFS_NOATIME_FL)
95 inode->i_flags |= S_NOATIME;
96 else
97 inode->i_flags &= ~S_NOATIME;
98
99}
100
101/* NOTE: symname is ignored unless the inode is a sym link */
102static int copy_attributes_to_inode(struct inode *inode,
103 struct PVFS_sys_attr_s *attrs,
104 char *symname)
105{
106 int ret = -1;
107 int perm_mode = 0;
108 struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode);
109 loff_t inode_size = 0;
110 loff_t rounded_up_size = 0;
111
112
113 /*
114 arbitrarily set the inode block size; FIXME: we need to
115 resolve the difference between the reported inode blocksize
116 and the PAGE_CACHE_SIZE, since our block count will always
117 be wrong.
118
119 For now, we're setting the block count to be the proper
120 number assuming the block size is 512 bytes, and the size is
121 rounded up to the nearest 4K. This is apparently required
122 to get proper size reports from the 'du' shell utility.
123
124 changing the inode->i_blkbits to something other than
125 PAGE_CACHE_SHIFT breaks mmap/execution as we depend on that.
126 */
127 gossip_debug(GOSSIP_UTILS_DEBUG,
128 "attrs->mask = %x (objtype = %s)\n",
129 attrs->mask,
130 attrs->objtype == PVFS_TYPE_METAFILE ? "file" :
131 attrs->objtype == PVFS_TYPE_DIRECTORY ? "directory" :
132 attrs->objtype == PVFS_TYPE_SYMLINK ? "symlink" :
133 "invalid/unknown");
134
135 switch (attrs->objtype) {
136 case PVFS_TYPE_METAFILE:
137 pvfs2_set_inode_flags(inode, attrs);
138 if (attrs->mask & PVFS_ATTR_SYS_SIZE) {
139 inode_size = (loff_t) attrs->size;
140 rounded_up_size =
141 (inode_size + (4096 - (inode_size % 4096)));
142
143 pvfs2_lock_inode(inode);
144 inode->i_bytes = inode_size;
145 inode->i_blocks =
146 (unsigned long)(rounded_up_size / 512);
147 pvfs2_unlock_inode(inode);
148
149 /*
150 * NOTE: make sure all the places we're called
151 * from have the inode->i_sem lock. We're fine
152 * in 99% of the cases since we're mostly
153 * called from a lookup.
154 */
155 inode->i_size = inode_size;
156 }
157 break;
158 case PVFS_TYPE_SYMLINK:
159 if (symname != NULL) {
160 inode->i_size = (loff_t) strlen(symname);
161 break;
162 }
163 /*FALLTHRU*/
164 default:
f0566532
MM		 165 inode->i_size = PAGE_CACHE_SIZE;
166
f7be4ee0 167 pvfs2_lock_inode(inode);
f0566532 168 inode_set_bytes(inode, inode->i_size);
f7be4ee0 169 pvfs2_unlock_inode(inode);
f7be4ee0
MM		 170 break;
171 }
172
173 inode->i_uid = make_kuid(&init_user_ns, attrs->owner);
174 inode->i_gid = make_kgid(&init_user_ns, attrs->group);
175 inode->i_atime.tv_sec = (time_t) attrs->atime;
176 inode->i_mtime.tv_sec = (time_t) attrs->mtime;
177 inode->i_ctime.tv_sec = (time_t) attrs->ctime;
178 inode->i_atime.tv_nsec = 0;
179 inode->i_mtime.tv_nsec = 0;
180 inode->i_ctime.tv_nsec = 0;
181
182 if (attrs->perms & PVFS_O_EXECUTE)
183 perm_mode |= S_IXOTH;
184 if (attrs->perms & PVFS_O_WRITE)
185 perm_mode |= S_IWOTH;
186 if (attrs->perms & PVFS_O_READ)
187 perm_mode |= S_IROTH;
188
189 if (attrs->perms & PVFS_G_EXECUTE)
190 perm_mode |= S_IXGRP;
191 if (attrs->perms & PVFS_G_WRITE)
192 perm_mode |= S_IWGRP;
193 if (attrs->perms & PVFS_G_READ)
194 perm_mode |= S_IRGRP;
195
196 if (attrs->perms & PVFS_U_EXECUTE)
197 perm_mode |= S_IXUSR;
198 if (attrs->perms & PVFS_U_WRITE)
199 perm_mode |= S_IWUSR;
200 if (attrs->perms & PVFS_U_READ)
201 perm_mode |= S_IRUSR;
202
203 if (attrs->perms & PVFS_G_SGID)
204 perm_mode |= S_ISGID;
205 if (attrs->perms & PVFS_U_SUID)
206 perm_mode |= S_ISUID;
207
208 inode->i_mode = perm_mode;
209
210 if (is_root_handle(inode)) {
211 /* special case: mark the root inode as sticky */
212 inode->i_mode |= S_ISVTX;
213 gossip_debug(GOSSIP_UTILS_DEBUG,
214 "Marking inode %pU as sticky\n",
215 get_khandle_from_ino(inode));
216 }
217
218 switch (attrs->objtype) {
219 case PVFS_TYPE_METAFILE:
220 inode->i_mode |= S_IFREG;
221 ret = 0;
222 break;
223 case PVFS_TYPE_DIRECTORY:
224 inode->i_mode |= S_IFDIR;
225 /* NOTE: we have no good way to keep nlink consistent
226 * for directories across clients; keep constant at 1.
227 * Why 1? If we go with 2, then find(1) gets confused
228 * and won't work properly withouth the -noleaf option
229 */
230 set_nlink(inode, 1);
231 ret = 0;
232 break;
233 case PVFS_TYPE_SYMLINK:
234 inode->i_mode |= S_IFLNK;
235
236 /* copy link target to inode private data */
237 if (pvfs2_inode && symname) {
238 strncpy(pvfs2_inode->link_target,
239 symname,
240 PVFS_NAME_MAX);
241 gossip_debug(GOSSIP_UTILS_DEBUG,
242 "Copied attr link target %s\n",
243 pvfs2_inode->link_target);
244 }
245 gossip_debug(GOSSIP_UTILS_DEBUG,
246 "symlink mode %o\n",
247 inode->i_mode);
248 ret = 0;
249 break;
250 default:
251 gossip_err("pvfs2: copy_attributes_to_inode: got invalid attribute type %x\n",
252 attrs->objtype);
253 }
254
255 gossip_debug(GOSSIP_UTILS_DEBUG,
256 "pvfs2: copy_attributes_to_inode: setting i_mode to %o, i_size to %lu\n",
257 inode->i_mode,
258 (unsigned long)i_size_read(inode));
259
260 return ret;
261}
262
263/*
264 * NOTE: in kernel land, we never use the sys_attr->link_target for
265 * anything, so don't bother copying it into the sys_attr object here.
266 */
267static inline int copy_attributes_from_inode(struct inode *inode,
268 struct PVFS_sys_attr_s *attrs,
269 struct iattr *iattr)
270{
271 umode_t tmp_mode;
272
273 if (!iattr || !inode || !attrs) {
274 gossip_err("NULL iattr (%p), inode (%p), attrs (%p) "
275 "in copy_attributes_from_inode!\n",
276 iattr,
277 inode,
278 attrs);
279 return -EINVAL;
280 }
281 /*
282 * We need to be careful to only copy the attributes out of the
283 * iattr object that we know are valid.
284 */
285 attrs->mask = 0;
286 if (iattr->ia_valid & ATTR_UID) {
287 attrs->owner = from_kuid(current_user_ns(), iattr->ia_uid);
288 attrs->mask |= PVFS_ATTR_SYS_UID;
289 gossip_debug(GOSSIP_UTILS_DEBUG, "(UID) %d\n", attrs->owner);
290 }
291 if (iattr->ia_valid & ATTR_GID) {
292 attrs->group = from_kgid(current_user_ns(), iattr->ia_gid);
293 attrs->mask |= PVFS_ATTR_SYS_GID;
294 gossip_debug(GOSSIP_UTILS_DEBUG, "(GID) %d\n", attrs->group);
295 }
296
297 if (iattr->ia_valid & ATTR_ATIME) {
298 attrs->mask |= PVFS_ATTR_SYS_ATIME;
299 if (iattr->ia_valid & ATTR_ATIME_SET) {
300 attrs->atime =
301 pvfs2_convert_time_field((void *)&iattr->ia_atime);
302 attrs->mask |= PVFS_ATTR_SYS_ATIME_SET;
303 }
304 }
305 if (iattr->ia_valid & ATTR_MTIME) {
306 attrs->mask |= PVFS_ATTR_SYS_MTIME;
307 if (iattr->ia_valid & ATTR_MTIME_SET) {
308 attrs->mtime =
309 pvfs2_convert_time_field((void *)&iattr->ia_mtime);
310 attrs->mask |= PVFS_ATTR_SYS_MTIME_SET;
311 }
312 }
313 if (iattr->ia_valid & ATTR_CTIME)
314 attrs->mask |= PVFS_ATTR_SYS_CTIME;
315
316 /*
317 * PVFS2 cannot set size with a setattr operation. Probably not likely
318 * to be requested through the VFS, but just in case, don't worry about
319 * ATTR_SIZE
320 */
321
322 if (iattr->ia_valid & ATTR_MODE) {
323 tmp_mode = iattr->ia_mode;
324 if (tmp_mode & (S_ISVTX)) {
325 if (is_root_handle(inode)) {
326 /*
327 * allow sticky bit to be set on root (since
328 * it shows up that way by default anyhow),
329 * but don't show it to the server
330 */
331 tmp_mode -= S_ISVTX;
332 } else {
333 gossip_debug(GOSSIP_UTILS_DEBUG,
334 "User attempted to set sticky bit on non-root directory; returning EINVAL.\n");
335 return -EINVAL;
336 }
337 }
338
339 if (tmp_mode & (S_ISUID)) {
340 gossip_debug(GOSSIP_UTILS_DEBUG,
341 "Attempting to set setuid bit (not supported); returning EINVAL.\n");
342 return -EINVAL;
343 }
344
345 attrs->perms = PVFS_util_translate_mode(tmp_mode);
346 attrs->mask |= PVFS_ATTR_SYS_PERM;
347 }
348
349 return 0;
350}
351
352/*
353 * issues a pvfs2 getattr request and fills in the appropriate inode
354 * attributes if successful. returns 0 on success; -errno otherwise
355 */
356int pvfs2_inode_getattr(struct inode *inode, __u32 getattr_mask)
357{
358 struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode);
359 struct pvfs2_kernel_op_s *new_op;
360 int ret = -EINVAL;
361
362 gossip_debug(GOSSIP_UTILS_DEBUG,
363 "%s: called on inode %pU\n",
364 __func__,
365 get_khandle_from_ino(inode));
366
367 new_op = op_alloc(PVFS2_VFS_OP_GETATTR);
368 if (!new_op)
369 return -ENOMEM;
370 new_op->upcall.req.getattr.refn = pvfs2_inode->refn;
371 new_op->upcall.req.getattr.mask = getattr_mask;
372
373 ret = service_operation(new_op, __func__,
374 get_interruptible_flag(inode));
375 if (ret != 0)
376 goto out;
377
378 if (copy_attributes_to_inode(inode,
379 &new_op->downcall.resp.getattr.attributes,
380 new_op->downcall.resp.getattr.link_target)) {
381 gossip_err("%s: failed to copy attributes\n", __func__);
382 ret = -ENOENT;
383 goto out;
384 }
385
386 /*
387 * Store blksize in pvfs2 specific part of inode structure; we are
388 * only going to use this to report to stat to make sure it doesn't
389 * perturb any inode related code paths.
390 */
391 if (new_op->downcall.resp.getattr.attributes.objtype ==
392 PVFS_TYPE_METAFILE) {
393 pvfs2_inode->blksize =
394 new_op->downcall.resp.getattr.attributes.blksize;
395 } else {
396 /* mimic behavior of generic_fillattr() for other types. */
397 pvfs2_inode->blksize = (1 << inode->i_blkbits);
398
399 }
400
401out:
402 gossip_debug(GOSSIP_UTILS_DEBUG,
403 "Getattr on handle %pU, "
404 "fsid %d\n (inode ct = %d) returned %d\n",
405 &pvfs2_inode->refn.khandle,
406 pvfs2_inode->refn.fs_id,
407 (int)atomic_read(&inode->i_count),
408 ret);
409
410 op_release(new_op);
411 return ret;
412}
413
414/*
415 * issues a pvfs2 setattr request to make sure the new attribute values
416 * take effect if successful. returns 0 on success; -errno otherwise
417 */
418int pvfs2_inode_setattr(struct inode *inode, struct iattr *iattr)
419{
420 struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode);
421 struct pvfs2_kernel_op_s *new_op;
422 int ret;
423
424 new_op = op_alloc(PVFS2_VFS_OP_SETATTR);
425 if (!new_op)
426 return -ENOMEM;
427
428 new_op->upcall.req.setattr.refn = pvfs2_inode->refn;
429 ret = copy_attributes_from_inode(inode,
430 &new_op->upcall.req.setattr.attributes,
431 iattr);
432 if (ret < 0) {
433 op_release(new_op);
434 return ret;
435 }
436
437 ret = service_operation(new_op, __func__,
438 get_interruptible_flag(inode));
439
440 gossip_debug(GOSSIP_UTILS_DEBUG,
441 "pvfs2_inode_setattr: returning %d\n",
442 ret);
443
444 /* when request is serviced properly, free req op struct */
445 op_release(new_op);
446
447 /*
448 * successful setattr should clear the atime, mtime and
449 * ctime flags.
450 */
451 if (ret == 0) {
452 ClearAtimeFlag(pvfs2_inode);
453 ClearMtimeFlag(pvfs2_inode);
454 ClearCtimeFlag(pvfs2_inode);
455 ClearModeFlag(pvfs2_inode);
456 }
457
458 return ret;
459}
460
461int pvfs2_flush_inode(struct inode *inode)
462{
463 /*
464 * If it is a dirty inode, this function gets called.
465 * Gather all the information that needs to be setattr'ed
466 * Right now, this will only be used for mode, atime, mtime
467 * and/or ctime.
468 */
469 struct iattr wbattr;
470 int ret;
471 int mtime_flag;
472 int ctime_flag;
473 int atime_flag;
474 int mode_flag;
475 struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode);
476
477 memset(&wbattr, 0, sizeof(wbattr));
478
479 /*
480 * check inode flags up front, and clear them if they are set. This
481 * will prevent multiple processes from all trying to flush the same
482 * inode if they call close() simultaneously
483 */
484 mtime_flag = MtimeFlag(pvfs2_inode);
485 ClearMtimeFlag(pvfs2_inode);
486 ctime_flag = CtimeFlag(pvfs2_inode);
487 ClearCtimeFlag(pvfs2_inode);
488 atime_flag = AtimeFlag(pvfs2_inode);
489 ClearAtimeFlag(pvfs2_inode);
490 mode_flag = ModeFlag(pvfs2_inode);
491 ClearModeFlag(pvfs2_inode);
492
493 /* -- Lazy atime,mtime and ctime update --
494 * Note: all times are dictated by server in the new scheme
495 * and not by the clients
496 *
497 * Also mode updates are being handled now..
498 */
499
500 if (mtime_flag)
501 wbattr.ia_valid |= ATTR_MTIME;
502 if (ctime_flag)
503 wbattr.ia_valid |= ATTR_CTIME;
504 if (atime_flag)
505 wbattr.ia_valid |= ATTR_ATIME;
506
507 if (mode_flag) {
508 wbattr.ia_mode = inode->i_mode;
509 wbattr.ia_valid |= ATTR_MODE;
510 }
511
512 gossip_debug(GOSSIP_UTILS_DEBUG,
513 "*********** pvfs2_flush_inode: %pU "
514 "(ia_valid %d)\n",
515 get_khandle_from_ino(inode),
516 wbattr.ia_valid);
517 if (wbattr.ia_valid == 0) {
518 gossip_debug(GOSSIP_UTILS_DEBUG,
519 "pvfs2_flush_inode skipping setattr()\n");
520 return 0;
521 }
522
523 gossip_debug(GOSSIP_UTILS_DEBUG,
524 "pvfs2_flush_inode (%pU) writing mode %o\n",
525 get_khandle_from_ino(inode),
526 inode->i_mode);
527
528 ret = pvfs2_inode_setattr(inode, &wbattr);
529
530 return ret;
531}
532
533int pvfs2_unmount_sb(struct super_block *sb)
534{
535 int ret = -EINVAL;
536 struct pvfs2_kernel_op_s *new_op = NULL;
537
538 gossip_debug(GOSSIP_UTILS_DEBUG,
539 "pvfs2_unmount_sb called on sb %p\n",
540 sb);
541
542 new_op = op_alloc(PVFS2_VFS_OP_FS_UMOUNT);
543 if (!new_op)
544 return -ENOMEM;
545 new_op->upcall.req.fs_umount.id = PVFS2_SB(sb)->id;
546 new_op->upcall.req.fs_umount.fs_id = PVFS2_SB(sb)->fs_id;
547 strncpy(new_op->upcall.req.fs_umount.pvfs2_config_server,
548 PVFS2_SB(sb)->devname,
549 PVFS_MAX_SERVER_ADDR_LEN);
550
551 gossip_debug(GOSSIP_UTILS_DEBUG,
552 "Attempting PVFS2 Unmount via host %s\n",
553 new_op->upcall.req.fs_umount.pvfs2_config_server);
554
555 ret = service_operation(new_op, "pvfs2_fs_umount", 0);
556
557 gossip_debug(GOSSIP_UTILS_DEBUG,
558 "pvfs2_unmount: got return value of %d\n", ret);
559 if (ret)
560 sb = ERR_PTR(ret);
561 else
562 PVFS2_SB(sb)->mount_pending = 1;
563
564 op_release(new_op);
565 return ret;
566}
567
568/*
569 * NOTE: on successful cancellation, be sure to return -EINTR, as
570 * that's the return value the caller expects
571 */
572int pvfs2_cancel_op_in_progress(__u64 tag)
573{
574 int ret = -EINVAL;
575 struct pvfs2_kernel_op_s *new_op = NULL;
576
577 gossip_debug(GOSSIP_UTILS_DEBUG,
578 "pvfs2_cancel_op_in_progress called on tag %llu\n",
579 llu(tag));
580
581 new_op = op_alloc(PVFS2_VFS_OP_CANCEL);
582 if (!new_op)
583 return -ENOMEM;
584 new_op->upcall.req.cancel.op_tag = tag;
585
586 gossip_debug(GOSSIP_UTILS_DEBUG,
587 "Attempting PVFS2 operation cancellation of tag %llu\n",
588 llu(new_op->upcall.req.cancel.op_tag));
589
590 ret = service_operation(new_op, "pvfs2_cancel", PVFS2_OP_CANCELLATION);
591
592 gossip_debug(GOSSIP_UTILS_DEBUG,
593 "pvfs2_cancel_op_in_progress: got return value of %d\n",
594 ret);
595
596 op_release(new_op);
597 return ret;
598}
599
600void pvfs2_op_initialize(struct pvfs2_kernel_op_s *op)
601{
602 if (op) {
603 spin_lock(&op->lock);
604 op->io_completed = 0;
605
606 op->upcall.type = PVFS2_VFS_OP_INVALID;
607 op->downcall.type = PVFS2_VFS_OP_INVALID;
608 op->downcall.status = -1;
609
610 op->op_state = OP_VFS_STATE_UNKNOWN;
611 op->tag = 0;
612 spin_unlock(&op->lock);
613 }
614}
615
616void pvfs2_make_bad_inode(struct inode *inode)
617{
618 if (is_root_handle(inode)) {
619 /*
620 * if this occurs, the pvfs2-client-core was killed but we
621 * can't afford to lose the inode operations and such
622 * associated with the root handle in any case.
623 */
624 gossip_debug(GOSSIP_UTILS_DEBUG,
625 "*** NOT making bad root inode %pU\n",
626 get_khandle_from_ino(inode));
627 } else {
628 gossip_debug(GOSSIP_UTILS_DEBUG,
629 "*** making bad inode %pU\n",
630 get_khandle_from_ino(inode));
631 make_bad_inode(inode);
632 }
633}
634
635/* this code is based on linux/net/sunrpc/clnt.c:rpc_clnt_sigmask */
636void mask_blocked_signals(sigset_t *orig_sigset)
637{
638 unsigned long sigallow = sigmask(SIGKILL);
639 unsigned long irqflags = 0;
640 struct k_sigaction *action = pvfs2_current_sigaction;
641
642 sigallow |= ((action[SIGINT - 1].sa.sa_handler == SIG_DFL) ?
643 sigmask(SIGINT) :
644 0);
645 sigallow |= ((action[SIGQUIT - 1].sa.sa_handler == SIG_DFL) ?
646 sigmask(SIGQUIT) :
647 0);
648
649 spin_lock_irqsave(&pvfs2_current_signal_lock, irqflags);
650 *orig_sigset = current->blocked;
651 siginitsetinv(&current->blocked, sigallow & ~orig_sigset->sig[0]);
652 recalc_sigpending();
653 spin_unlock_irqrestore(&pvfs2_current_signal_lock, irqflags);
654}
655
656/* this code is based on linux/net/sunrpc/clnt.c:rpc_clnt_sigunmask */
657void unmask_blocked_signals(sigset_t *orig_sigset)
658{
659 unsigned long irqflags = 0;
660
661 spin_lock_irqsave(&pvfs2_current_signal_lock, irqflags);
662 current->blocked = *orig_sigset;
663 recalc_sigpending();
664 spin_unlock_irqrestore(&pvfs2_current_signal_lock, irqflags);
665}
666
667__u64 pvfs2_convert_time_field(void *time_ptr)
668{
669 __u64 pvfs2_time;
670 struct timespec *tspec = (struct timespec *)time_ptr;
671
672 pvfs2_time = (__u64) ((time_t) tspec->tv_sec);
673 return pvfs2_time;
674}
675
676/* macro defined in include/pvfs2-types.h */
677DECLARE_ERRNO_MAPPING_AND_FN();
678
679int pvfs2_normalize_to_errno(__s32 error_code)
680{
681 if (error_code > 0) {
682 gossip_err("pvfs2: error status receieved.\n");
683 gossip_err("pvfs2: assuming error code is inverted.\n");
684 error_code = -error_code;
685 }
686
687 /* convert any error codes that are in pvfs2 format */
688 if (IS_PVFS_NON_ERRNO_ERROR(-error_code)) {
689 if (PVFS_NON_ERRNO_ERROR_CODE(-error_code) == PVFS_ECANCEL) {
690 /*
691 * cancellation error codes generally correspond to
692 * a timeout from the client's perspective
693 */
694 error_code = -ETIMEDOUT;
695 } else {
696 /* assume a default error code */
697 gossip_err("pvfs2: warning: got error code without errno equivalent: %d.\n",
698 error_code);
699 error_code = -EINVAL;
700 }
701 } else if (IS_PVFS_ERROR(-error_code)) {
702 error_code = -PVFS_ERROR_TO_ERRNO(-error_code);
703 }
704 return error_code;
705}
706
707#define NUM_MODES 11
708__s32 PVFS_util_translate_mode(int mode)
709{
710 int ret = 0;
711 int i = 0;
712 static int modes[NUM_MODES] = {
713 S_IXOTH, S_IWOTH, S_IROTH,
714 S_IXGRP, S_IWGRP, S_IRGRP,
715 S_IXUSR, S_IWUSR, S_IRUSR,
716 S_ISGID, S_ISUID
717 };
718 static int pvfs2_modes[NUM_MODES] = {
719 PVFS_O_EXECUTE, PVFS_O_WRITE, PVFS_O_READ,
720 PVFS_G_EXECUTE, PVFS_G_WRITE, PVFS_G_READ,
721 PVFS_U_EXECUTE, PVFS_U_WRITE, PVFS_U_READ,
722 PVFS_G_SGID, PVFS_U_SUID
723 };
724
725 for (i = 0; i < NUM_MODES; i++)
726 if (mode & modes[i])
727 ret |= pvfs2_modes[i];
728
729 return ret;
730}
731#undef NUM_MODES
732
733/*
734 * After obtaining a string representation of the client's debug
735 * keywords and their associated masks, this function is called to build an
736 * array of these values.
737 */
738int orangefs_prepare_cdm_array(char *debug_array_string)
739{
740 int i;
741 int rc = -EINVAL;
742 char *cds_head = NULL;
743 char *cds_delimiter = NULL;
744 int keyword_len = 0;
745
746 gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__);
747
748 /*
749 * figure out how many elements the cdm_array needs.
750 */
751 for (i = 0; i < strlen(debug_array_string); i++)
752 if (debug_array_string[i] == '\n')
753 cdm_element_count++;
754
755 if (!cdm_element_count) {
756 pr_info("No elements in client debug array string!\n");
757 goto out;
758 }
759
760 cdm_array =
761 kzalloc(cdm_element_count * sizeof(struct client_debug_mask),
762 GFP_KERNEL);
763 if (!cdm_array) {
764 pr_info("malloc failed for cdm_array!\n");
765 rc = -ENOMEM;
766 goto out;
767 }
768
769 cds_head = debug_array_string;
770
771 for (i = 0; i < cdm_element_count; i++) {
772 cds_delimiter = strchr(cds_head, '\n');
773 *cds_delimiter = '\0';
774
775 keyword_len = strcspn(cds_head, " ");
776
777 cdm_array[i].keyword = kzalloc(keyword_len + 1, GFP_KERNEL);
778 if (!cdm_array[i].keyword) {
779 rc = -ENOMEM;
780 goto out;
781 }
782
783 sscanf(cds_head,
784 "%s %llx %llx",
785 cdm_array[i].keyword,
786 (unsigned long long *)&(cdm_array[i].mask1),
787 (unsigned long long *)&(cdm_array[i].mask2));
788
789 if (!strcmp(cdm_array[i].keyword, PVFS2_VERBOSE))
790 client_verbose_index = i;
791
792 if (!strcmp(cdm_array[i].keyword, PVFS2_ALL))
793 client_all_index = i;
794
795 cds_head = cds_delimiter + 1;
796 }
797
798 rc = cdm_element_count;
799
800 gossip_debug(GOSSIP_UTILS_DEBUG, "%s: rc:%d:\n", __func__, rc);
801
802out:
803
804 return rc;
805
806}
807
808/*
809 * /sys/kernel/debug/orangefs/debug-help can be catted to
810 * see all the available kernel and client debug keywords.
811 *
812 * When the kernel boots, we have no idea what keywords the
813 * client supports, nor their associated masks.
814 *
815 * We pass through this function once at boot and stamp a
816 * boilerplate "we don't know" message for the client in the
817 * debug-help file. We pass through here again when the client
818 * starts and then we can fill out the debug-help file fully.
819 *
820 * The client might be restarted any number of times between
821 * reboots, we only build the debug-help file the first time.
822 */
823int orangefs_prepare_debugfs_help_string(int at_boot)
824{
825 int rc = -EINVAL;
826 int i;
827 int byte_count = 0;
828 char *client_title = "Client Debug Keywords:\n";
829 char *kernel_title = "Kernel Debug Keywords:\n";
830
831 gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__);
832
833 if (at_boot) {
834 byte_count += strlen(HELP_STRING_UNINITIALIZED);
835 client_title = HELP_STRING_UNINITIALIZED;
836 } else {
837 /*
838 * fill the client keyword/mask array and remember
839 * how many elements there were.
840 */
841 cdm_element_count =
842 orangefs_prepare_cdm_array(client_debug_array_string);
843 if (cdm_element_count <= 0)
844 goto out;
845
846 /* Count the bytes destined for debug_help_string. */
847 byte_count += strlen(client_title);
848
849 for (i = 0; i < cdm_element_count; i++) {
850 byte_count += strlen(cdm_array[i].keyword + 2);
851 if (byte_count >= DEBUG_HELP_STRING_SIZE) {
852 pr_info("%s: overflow 1!\n", __func__);
853 goto out;
854 }
855 }
856
857 gossip_debug(GOSSIP_UTILS_DEBUG,
858 "%s: cdm_element_count:%d:\n",
859 __func__,
860 cdm_element_count);
861 }
862
863 byte_count += strlen(kernel_title);
864 for (i = 0; i < num_kmod_keyword_mask_map; i++) {
865 byte_count +=
866 strlen(s_kmod_keyword_mask_map[i].keyword + 2);
867 if (byte_count >= DEBUG_HELP_STRING_SIZE) {
868 pr_info("%s: overflow 2!\n", __func__);
869 goto out;
870 }
871 }
872
873 /* build debug_help_string. */
874 debug_help_string = kzalloc(DEBUG_HELP_STRING_SIZE, GFP_KERNEL);
875 if (!debug_help_string) {
876 rc = -ENOMEM;
877 goto out;
878 }
879
880 strcat(debug_help_string, client_title);
881
882 if (!at_boot) {
883 for (i = 0; i < cdm_element_count; i++) {
884 strcat(debug_help_string, "\t");
885 strcat(debug_help_string, cdm_array[i].keyword);
886 strcat(debug_help_string, "\n");
887 }
888 }
889
890 strcat(debug_help_string, "\n");
891 strcat(debug_help_string, kernel_title);
892
893 for (i = 0; i < num_kmod_keyword_mask_map; i++) {
894 strcat(debug_help_string, "\t");
895 strcat(debug_help_string, s_kmod_keyword_mask_map[i].keyword);
896 strcat(debug_help_string, "\n");
897 }
898
899 rc = 0;
900
901out:
902
903 return rc;
904
905}
906
907/*
908 * kernel = type 0
909 * client = type 1
910 */
911void debug_mask_to_string(void *mask, int type)
912{
913 int i;
914 int len = 0;
915 char *debug_string;
916 int element_count = 0;
917
918 gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__);
919
920 if (type) {
921 debug_string = client_debug_string;
922 element_count = cdm_element_count;
923 } else {
924 debug_string = kernel_debug_string;
925 element_count = num_kmod_keyword_mask_map;
926 }
927
928 memset(debug_string, 0, PVFS2_MAX_DEBUG_STRING_LEN);
929
930 /*
931 * Some keywords, like "all" or "verbose", are amalgams of
932 * numerous other keywords. Make a special check for those
933 * before grinding through the whole mask only to find out
934 * later...
935 */
936 if (check_amalgam_keyword(mask, type))
937 goto out;
938
939 /* Build the debug string. */
940 for (i = 0; i < element_count; i++)
941 if (type)
942 do_c_string(mask, i);
943 else
944 do_k_string(mask, i);
945
946 len = strlen(debug_string);
947
948 if ((len) && (type))
949 client_debug_string[len - 1] = '\0';
950 else if (len)
951 kernel_debug_string[len - 1] = '\0';
952 else if (type)
953 strcpy(client_debug_string, "none");
954 else
955 strcpy(kernel_debug_string, "none");
956
957out:
958gossip_debug(GOSSIP_UTILS_DEBUG, "%s: string:%s:\n", __func__, debug_string);
959
960 return;
961
962}
963
964void do_k_string(void *k_mask, int index)
965{
966 __u64 *mask = (__u64 *) k_mask;
967
968 if (keyword_is_amalgam((char *) s_kmod_keyword_mask_map[index].keyword))
969 goto out;
970
971 if (*mask & s_kmod_keyword_mask_map[index].mask_val) {
972 if ((strlen(kernel_debug_string) +
973 strlen(s_kmod_keyword_mask_map[index].keyword))
974 < PVFS2_MAX_DEBUG_STRING_LEN - 1) {
975 strcat(kernel_debug_string,
976 s_kmod_keyword_mask_map[index].keyword);
977 strcat(kernel_debug_string, ",");
978 } else {
979 gossip_err("%s: overflow!\n", __func__);
980 strcpy(kernel_debug_string, PVFS2_ALL);
981 goto out;
982 }
983 }
984
985out:
986
987 return;
988}
989
990void do_c_string(void *c_mask, int index)
991{
992 struct client_debug_mask *mask = (struct client_debug_mask *) c_mask;
993
994 if (keyword_is_amalgam(cdm_array[index].keyword))
995 goto out;
996
997 if ((mask->mask1 & cdm_array[index].mask1) ||
998 (mask->mask2 & cdm_array[index].mask2)) {
999 if ((strlen(client_debug_string) +
1000 strlen(cdm_array[index].keyword) + 1)
1001 < PVFS2_MAX_DEBUG_STRING_LEN - 2) {
1002 strcat(client_debug_string,
1003 cdm_array[index].keyword);
1004 strcat(client_debug_string, ",");
1005 } else {
1006 gossip_err("%s: overflow!\n", __func__);
1007 strcpy(client_debug_string, PVFS2_ALL);
1008 goto out;
1009 }
1010 }
1011out:
1012 return;
1013}
1014
1015int keyword_is_amalgam(char *keyword)
1016{
1017 int rc = 0;
1018
1019 if ((!strcmp(keyword, PVFS2_ALL)) || (!strcmp(keyword, PVFS2_VERBOSE)))
1020 rc = 1;
1021
1022 return rc;
1023}
1024
1025/*
1026 * kernel = type 0
1027 * client = type 1
1028 *
1029 * return 1 if we found an amalgam.
1030 */
1031int check_amalgam_keyword(void *mask, int type)
1032{
1033 __u64 *k_mask;
1034 struct client_debug_mask *c_mask;
1035 int k_all_index = num_kmod_keyword_mask_map - 1;
1036 int rc = 0;
1037
1038 if (type) {
1039 c_mask = (struct client_debug_mask *) mask;
1040
1041 if ((c_mask->mask1 == cdm_array[client_all_index].mask1) &&
1042 (c_mask->mask2 == cdm_array[client_all_index].mask2)) {
1043 strcpy(client_debug_string, PVFS2_ALL);
1044 rc = 1;
1045 goto out;
1046 }
1047
1048 if ((c_mask->mask1 == cdm_array[client_verbose_index].mask1) &&
1049 (c_mask->mask2 == cdm_array[client_verbose_index].mask2)) {
1050 strcpy(client_debug_string, PVFS2_VERBOSE);
1051 rc = 1;
1052 goto out;
1053 }
1054
1055 } else {
1056 k_mask = (__u64 *) mask;
1057
1058 if (*k_mask >= s_kmod_keyword_mask_map[k_all_index].mask_val) {
1059 strcpy(kernel_debug_string, PVFS2_ALL);
1060 rc = 1;
1061 goto out;
1062 }
1063 }
1064
1065out:
1066
1067 return rc;
1068}
1069
1070/*
1071 * kernel = type 0
1072 * client = type 1
1073 */
1074void debug_string_to_mask(char *debug_string, void *mask, int type)
1075{
1076 char *unchecked_keyword;
1077 int i;
1078 char *strsep_fodder = kstrdup(debug_string, GFP_KERNEL);
eeaa3d44 1079 char *original_pointer;
f7be4ee0
MM		 1080 int element_count = 0;
1081 struct client_debug_mask *c_mask;
1082 __u64 *k_mask;
1083
1084 gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__);
1085
1086 if (type) {
1087 c_mask = (struct client_debug_mask *)mask;
1088 element_count = cdm_element_count;
1089 } else {
1090 k_mask = (__u64 *)mask;
1091 *k_mask = 0;
1092 element_count = num_kmod_keyword_mask_map;
1093 }
1094
eeaa3d44 1095 original_pointer = strsep_fodder;
f7be4ee0
MM		 1096 while ((unchecked_keyword = strsep(&strsep_fodder, ",")))
1097 if (strlen(unchecked_keyword)) {
1098 for (i = 0; i < element_count; i++)
1099 if (type)
1100 do_c_mask(i,
1101 unchecked_keyword,
1102 &c_mask);
1103 else
1104 do_k_mask(i,
1105 unchecked_keyword,
1106 &k_mask);
1107 }
1108
eeaa3d44 1109 kfree(original_pointer);
f7be4ee0
MM		 1110}
1111
1112void do_c_mask(int i,
1113 char *unchecked_keyword,
1114 struct client_debug_mask **sane_mask)
1115{
1116
1117 if (!strcmp(cdm_array[i].keyword, unchecked_keyword)) {
1118 (**sane_mask).mask1 = (**sane_mask).mask1 | cdm_array[i].mask1;
1119 (**sane_mask).mask2 = (**sane_mask).mask2 | cdm_array[i].mask2;
1120 }
1121}
1122
1123void do_k_mask(int i, char *unchecked_keyword, __u64 **sane_mask)
1124{
1125
1126 if (!strcmp(s_kmod_keyword_mask_map[i].keyword, unchecked_keyword))
1127 **sane_mask = (**sane_mask) |
1128 s_kmod_keyword_mask_map[i].mask_val;
1129}
Linux 6.x block layer and io_uring tree(s)