1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * (C) 2001 Clemson University and The University of Chicago
4 *
5 * See COPYING in top-level directory.
6 */
7 #include <linux/kernel.h>
8 #include "protocol.h"
9 #include "orangefs-kernel.h"
10 #include "orangefs-dev-proto.h"
11 #include "orangefs-bufmap.h"
12
fsid_of_op(struct orangefs_kernel_op_s * op)13 __s32 fsid_of_op(struct orangefs_kernel_op_s *op)
14 {
15 __s32 fsid = ORANGEFS_FS_ID_NULL;
16
17 if (op) {
18 switch (op->upcall.type) {
19 case ORANGEFS_VFS_OP_FILE_IO:
20 fsid = op->upcall.req.io.refn.fs_id;
21 break;
22 case ORANGEFS_VFS_OP_LOOKUP:
23 fsid = op->upcall.req.lookup.parent_refn.fs_id;
24 break;
25 case ORANGEFS_VFS_OP_CREATE:
26 fsid = op->upcall.req.create.parent_refn.fs_id;
27 break;
28 case ORANGEFS_VFS_OP_GETATTR:
29 fsid = op->upcall.req.getattr.refn.fs_id;
30 break;
31 case ORANGEFS_VFS_OP_REMOVE:
32 fsid = op->upcall.req.remove.parent_refn.fs_id;
33 break;
34 case ORANGEFS_VFS_OP_MKDIR:
35 fsid = op->upcall.req.mkdir.parent_refn.fs_id;
36 break;
37 case ORANGEFS_VFS_OP_READDIR:
38 fsid = op->upcall.req.readdir.refn.fs_id;
39 break;
40 case ORANGEFS_VFS_OP_SETATTR:
41 fsid = op->upcall.req.setattr.refn.fs_id;
42 break;
43 case ORANGEFS_VFS_OP_SYMLINK:
44 fsid = op->upcall.req.sym.parent_refn.fs_id;
45 break;
46 case ORANGEFS_VFS_OP_RENAME:
47 fsid = op->upcall.req.rename.old_parent_refn.fs_id;
48 break;
49 case ORANGEFS_VFS_OP_STATFS:
50 fsid = op->upcall.req.statfs.fs_id;
51 break;
52 case ORANGEFS_VFS_OP_TRUNCATE:
53 fsid = op->upcall.req.truncate.refn.fs_id;
54 break;
55 case ORANGEFS_VFS_OP_RA_FLUSH:
56 fsid = op->upcall.req.ra_cache_flush.refn.fs_id;
57 break;
58 case ORANGEFS_VFS_OP_FS_UMOUNT:
59 fsid = op->upcall.req.fs_umount.fs_id;
60 break;
61 case ORANGEFS_VFS_OP_GETXATTR:
62 fsid = op->upcall.req.getxattr.refn.fs_id;
63 break;
64 case ORANGEFS_VFS_OP_SETXATTR:
65 fsid = op->upcall.req.setxattr.refn.fs_id;
66 break;
67 case ORANGEFS_VFS_OP_LISTXATTR:
68 fsid = op->upcall.req.listxattr.refn.fs_id;
69 break;
70 case ORANGEFS_VFS_OP_REMOVEXATTR:
71 fsid = op->upcall.req.removexattr.refn.fs_id;
72 break;
73 case ORANGEFS_VFS_OP_FSYNC:
74 fsid = op->upcall.req.fsync.refn.fs_id;
75 break;
76 default:
77 break;
78 }
79 }
80 return fsid;
81 }
82
orangefs_inode_flags(struct ORANGEFS_sys_attr_s * attrs)83 static int orangefs_inode_flags(struct ORANGEFS_sys_attr_s *attrs)
84 {
85 int flags = 0;
86 if (attrs->flags & ORANGEFS_IMMUTABLE_FL)
87 flags |= S_IMMUTABLE;
88 else
89 flags &= ~S_IMMUTABLE;
90 if (attrs->flags & ORANGEFS_APPEND_FL)
91 flags |= S_APPEND;
92 else
93 flags &= ~S_APPEND;
94 if (attrs->flags & ORANGEFS_NOATIME_FL)
95 flags |= S_NOATIME;
96 else
97 flags &= ~S_NOATIME;
98 return flags;
99 }
100
orangefs_inode_perms(struct ORANGEFS_sys_attr_s * attrs)101 static int orangefs_inode_perms(struct ORANGEFS_sys_attr_s *attrs)
102 {
103 int perm_mode = 0;
104
105 if (attrs->perms & ORANGEFS_O_EXECUTE)
106 perm_mode |= S_IXOTH;
107 if (attrs->perms & ORANGEFS_O_WRITE)
108 perm_mode |= S_IWOTH;
109 if (attrs->perms & ORANGEFS_O_READ)
110 perm_mode |= S_IROTH;
111
112 if (attrs->perms & ORANGEFS_G_EXECUTE)
113 perm_mode |= S_IXGRP;
114 if (attrs->perms & ORANGEFS_G_WRITE)
115 perm_mode |= S_IWGRP;
116 if (attrs->perms & ORANGEFS_G_READ)
117 perm_mode |= S_IRGRP;
118
119 if (attrs->perms & ORANGEFS_U_EXECUTE)
120 perm_mode |= S_IXUSR;
121 if (attrs->perms & ORANGEFS_U_WRITE)
122 perm_mode |= S_IWUSR;
123 if (attrs->perms & ORANGEFS_U_READ)
124 perm_mode |= S_IRUSR;
125
126 if (attrs->perms & ORANGEFS_G_SGID)
127 perm_mode |= S_ISGID;
128 if (attrs->perms & ORANGEFS_U_SUID)
129 perm_mode |= S_ISUID;
130
131 return perm_mode;
132 }
133
134 /*
135 * NOTE: in kernel land, we never use the sys_attr->link_target for
136 * anything, so don't bother copying it into the sys_attr object here.
137 */
copy_attributes_from_inode(struct inode * inode,struct ORANGEFS_sys_attr_s * attrs,struct iattr * iattr)138 static inline int copy_attributes_from_inode(struct inode *inode,
139 struct ORANGEFS_sys_attr_s *attrs,
140 struct iattr *iattr)
141 {
142 umode_t tmp_mode;
143
144 if (!iattr || !inode || !attrs) {
145 gossip_err("NULL iattr (%p), inode (%p), attrs (%p) "
146 "in copy_attributes_from_inode!\n",
147 iattr,
148 inode,
149 attrs);
150 return -EINVAL;
151 }
152 /*
153 * We need to be careful to only copy the attributes out of the
154 * iattr object that we know are valid.
155 */
156 attrs->mask = 0;
157 if (iattr->ia_valid & ATTR_UID) {
158 attrs->owner = from_kuid(&init_user_ns, iattr->ia_uid);
159 attrs->mask |= ORANGEFS_ATTR_SYS_UID;
160 gossip_debug(GOSSIP_UTILS_DEBUG, "(UID) %d\n", attrs->owner);
161 }
162 if (iattr->ia_valid & ATTR_GID) {
163 attrs->group = from_kgid(&init_user_ns, iattr->ia_gid);
164 attrs->mask |= ORANGEFS_ATTR_SYS_GID;
165 gossip_debug(GOSSIP_UTILS_DEBUG, "(GID) %d\n", attrs->group);
166 }
167
168 if (iattr->ia_valid & ATTR_ATIME) {
169 attrs->mask |= ORANGEFS_ATTR_SYS_ATIME;
170 if (iattr->ia_valid & ATTR_ATIME_SET) {
171 attrs->atime = (time64_t)iattr->ia_atime.tv_sec;
172 attrs->mask |= ORANGEFS_ATTR_SYS_ATIME_SET;
173 }
174 }
175 if (iattr->ia_valid & ATTR_MTIME) {
176 attrs->mask |= ORANGEFS_ATTR_SYS_MTIME;
177 if (iattr->ia_valid & ATTR_MTIME_SET) {
178 attrs->mtime = (time64_t)iattr->ia_mtime.tv_sec;
179 attrs->mask |= ORANGEFS_ATTR_SYS_MTIME_SET;
180 }
181 }
182 if (iattr->ia_valid & ATTR_CTIME)
183 attrs->mask |= ORANGEFS_ATTR_SYS_CTIME;
184
185 /*
186 * ORANGEFS cannot set size with a setattr operation. Probably not
187 * likely to be requested through the VFS, but just in case, don't
188 * worry about ATTR_SIZE
189 */
190
191 if (iattr->ia_valid & ATTR_MODE) {
192 tmp_mode = iattr->ia_mode;
193 if (tmp_mode & (S_ISVTX)) {
194 if (is_root_handle(inode)) {
195 /*
196 * allow sticky bit to be set on root (since
197 * it shows up that way by default anyhow),
198 * but don't show it to the server
199 */
200 tmp_mode -= S_ISVTX;
201 } else {
202 gossip_debug(GOSSIP_UTILS_DEBUG,
203 "%s: setting sticky bit not supported.\n",
204 __func__);
205 return -EINVAL;
206 }
207 }
208
209 if (tmp_mode & (S_ISUID)) {
210 gossip_debug(GOSSIP_UTILS_DEBUG,
211 "%s: setting setuid bit not supported.\n",
212 __func__);
213 return -EINVAL;
214 }
215
216 attrs->perms = ORANGEFS_util_translate_mode(tmp_mode);
217 attrs->mask |= ORANGEFS_ATTR_SYS_PERM;
218 }
219
220 return 0;
221 }
222
orangefs_inode_type(enum orangefs_ds_type objtype)223 static int orangefs_inode_type(enum orangefs_ds_type objtype)
224 {
225 if (objtype == ORANGEFS_TYPE_METAFILE)
226 return S_IFREG;
227 else if (objtype == ORANGEFS_TYPE_DIRECTORY)
228 return S_IFDIR;
229 else if (objtype == ORANGEFS_TYPE_SYMLINK)
230 return S_IFLNK;
231 else
232 return -1;
233 }
234
orangefs_make_bad_inode(struct inode * inode)235 static void orangefs_make_bad_inode(struct inode *inode)
236 {
237 if (is_root_handle(inode)) {
238 /*
239 * if this occurs, the pvfs2-client-core was killed but we
240 * can't afford to lose the inode operations and such
241 * associated with the root handle in any case.
242 */
243 gossip_debug(GOSSIP_UTILS_DEBUG,
244 "*** NOT making bad root inode %pU\n",
245 get_khandle_from_ino(inode));
246 } else {
247 gossip_debug(GOSSIP_UTILS_DEBUG,
248 "*** making bad inode %pU\n",
249 get_khandle_from_ino(inode));
250 make_bad_inode(inode);
251 }
252 }
253
orangefs_inode_is_stale(struct inode * inode,struct ORANGEFS_sys_attr_s * attrs,char * link_target)254 static int orangefs_inode_is_stale(struct inode *inode,
255 struct ORANGEFS_sys_attr_s *attrs, char *link_target)
256 {
257 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
258 int type = orangefs_inode_type(attrs->objtype);
259 /*
260 * If the inode type or symlink target have changed then this
261 * inode is stale.
262 */
263 if (type == -1 || !(inode->i_mode & type)) {
264 orangefs_make_bad_inode(inode);
265 return 1;
266 }
267 if (type == S_IFLNK && strncmp(orangefs_inode->link_target,
268 link_target, ORANGEFS_NAME_MAX)) {
269 orangefs_make_bad_inode(inode);
270 return 1;
271 }
272 return 0;
273 }
274
orangefs_inode_getattr(struct inode * inode,int new,int bypass,u32 request_mask)275 int orangefs_inode_getattr(struct inode *inode, int new, int bypass,
276 u32 request_mask)
277 {
278 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
279 struct orangefs_kernel_op_s *new_op;
280 loff_t inode_size;
281 int ret, type;
282
283 gossip_debug(GOSSIP_UTILS_DEBUG, "%s: called on inode %pU\n", __func__,
284 get_khandle_from_ino(inode));
285
286 if (!new && !bypass) {
287 /*
288 * Must have all the attributes in the mask and be within cache
289 * time.
290 */
291 if ((request_mask & orangefs_inode->getattr_mask) ==
292 request_mask &&
293 time_before(jiffies, orangefs_inode->getattr_time))
294 return 0;
295 }
296
297 new_op = op_alloc(ORANGEFS_VFS_OP_GETATTR);
298 if (!new_op)
299 return -ENOMEM;
300 new_op->upcall.req.getattr.refn = orangefs_inode->refn;
301 /*
302 * Size is the hardest attribute to get. The incremental cost of any
303 * other attribute is essentially zero.
304 */
305 if (request_mask & STATX_SIZE || new)
306 new_op->upcall.req.getattr.mask = ORANGEFS_ATTR_SYS_ALL_NOHINT;
307 else
308 new_op->upcall.req.getattr.mask =
309 ORANGEFS_ATTR_SYS_ALL_NOHINT & ~ORANGEFS_ATTR_SYS_SIZE;
310
311 ret = service_operation(new_op, __func__,
312 get_interruptible_flag(inode));
313 if (ret != 0)
314 goto out;
315
316 if (!new) {
317 ret = orangefs_inode_is_stale(inode,
318 &new_op->downcall.resp.getattr.attributes,
319 new_op->downcall.resp.getattr.link_target);
320 if (ret) {
321 ret = -ESTALE;
322 goto out;
323 }
324 }
325
326 type = orangefs_inode_type(new_op->
327 downcall.resp.getattr.attributes.objtype);
328 switch (type) {
329 case S_IFREG:
330 inode->i_flags = orangefs_inode_flags(&new_op->
331 downcall.resp.getattr.attributes);
332 if (request_mask & STATX_SIZE || new) {
333 inode_size = (loff_t)new_op->
334 downcall.resp.getattr.attributes.size;
335 inode->i_size = inode_size;
336 inode->i_blkbits = ffs(new_op->downcall.resp.getattr.
337 attributes.blksize);
338 spin_lock(&inode->i_lock);
339 inode->i_bytes = inode_size;
340 inode->i_blocks =
341 (inode_size + 512 - inode_size % 512)/512;
342 spin_unlock(&inode->i_lock);
343 }
344 break;
345 case S_IFDIR:
346 if (request_mask & STATX_SIZE || new) {
347 inode->i_size = PAGE_SIZE;
348 spin_lock(&inode->i_lock);
349 inode_set_bytes(inode, inode->i_size);
350 spin_unlock(&inode->i_lock);
351 }
352 set_nlink(inode, 1);
353 break;
354 case S_IFLNK:
355 if (new) {
356 inode->i_size = (loff_t)strlen(new_op->
357 downcall.resp.getattr.link_target);
358 ret = strscpy(orangefs_inode->link_target,
359 new_op->downcall.resp.getattr.link_target,
360 ORANGEFS_NAME_MAX);
361 if (ret == -E2BIG) {
362 ret = -EIO;
363 goto out;
364 }
365 inode->i_link = orangefs_inode->link_target;
366 }
367 break;
368 /* i.e. -1 */
369 default:
370 /* XXX: ESTALE? This is what is done if it is not new. */
371 orangefs_make_bad_inode(inode);
372 ret = -ESTALE;
373 goto out;
374 }
375
376 inode->i_uid = make_kuid(&init_user_ns, new_op->
377 downcall.resp.getattr.attributes.owner);
378 inode->i_gid = make_kgid(&init_user_ns, new_op->
379 downcall.resp.getattr.attributes.group);
380 inode->i_atime.tv_sec = (time64_t)new_op->
381 downcall.resp.getattr.attributes.atime;
382 inode->i_mtime.tv_sec = (time64_t)new_op->
383 downcall.resp.getattr.attributes.mtime;
384 inode->i_ctime.tv_sec = (time64_t)new_op->
385 downcall.resp.getattr.attributes.ctime;
386 inode->i_atime.tv_nsec = 0;
387 inode->i_mtime.tv_nsec = 0;
388 inode->i_ctime.tv_nsec = 0;
389
390 /* special case: mark the root inode as sticky */
391 inode->i_mode = type | (is_root_handle(inode) ? S_ISVTX : 0) |
392 orangefs_inode_perms(&new_op->downcall.resp.getattr.attributes);
393
394 orangefs_inode->getattr_time = jiffies +
395 orangefs_getattr_timeout_msecs*HZ/1000;
396 if (request_mask & STATX_SIZE || new)
397 orangefs_inode->getattr_mask = STATX_BASIC_STATS;
398 else
399 orangefs_inode->getattr_mask = STATX_BASIC_STATS & ~STATX_SIZE;
400 ret = 0;
401 out:
402 op_release(new_op);
403 return ret;
404 }
405
orangefs_inode_check_changed(struct inode * inode)406 int orangefs_inode_check_changed(struct inode *inode)
407 {
408 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
409 struct orangefs_kernel_op_s *new_op;
410 int ret;
411
412 gossip_debug(GOSSIP_UTILS_DEBUG, "%s: called on inode %pU\n", __func__,
413 get_khandle_from_ino(inode));
414
415 new_op = op_alloc(ORANGEFS_VFS_OP_GETATTR);
416 if (!new_op)
417 return -ENOMEM;
418 new_op->upcall.req.getattr.refn = orangefs_inode->refn;
419 new_op->upcall.req.getattr.mask = ORANGEFS_ATTR_SYS_TYPE |
420 ORANGEFS_ATTR_SYS_LNK_TARGET;
421
422 ret = service_operation(new_op, __func__,
423 get_interruptible_flag(inode));
424 if (ret != 0)
425 goto out;
426
427 ret = orangefs_inode_is_stale(inode,
428 &new_op->downcall.resp.getattr.attributes,
429 new_op->downcall.resp.getattr.link_target);
430 out:
431 op_release(new_op);
432 return ret;
433 }
434
435 /*
436 * issues a orangefs setattr request to make sure the new attribute values
437 * take effect if successful. returns 0 on success; -errno otherwise
438 */
orangefs_inode_setattr(struct inode * inode,struct iattr * iattr)439 int orangefs_inode_setattr(struct inode *inode, struct iattr *iattr)
440 {
441 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
442 struct orangefs_kernel_op_s *new_op;
443 int ret;
444
445 new_op = op_alloc(ORANGEFS_VFS_OP_SETATTR);
446 if (!new_op)
447 return -ENOMEM;
448
449 new_op->upcall.req.setattr.refn = orangefs_inode->refn;
450 ret = copy_attributes_from_inode(inode,
451 &new_op->upcall.req.setattr.attributes,
452 iattr);
453 if (ret >= 0) {
454 ret = service_operation(new_op, __func__,
455 get_interruptible_flag(inode));
456
457 gossip_debug(GOSSIP_UTILS_DEBUG,
458 "orangefs_inode_setattr: returning %d\n",
459 ret);
460 }
461
462 op_release(new_op);
463
464 if (ret == 0)
465 orangefs_inode->getattr_time = jiffies - 1;
466
467 return ret;
468 }
469
470 /*
471 * The following is a very dirty hack that is now a permanent part of the
472 * ORANGEFS protocol. See protocol.h for more error definitions.
473 */
474
475 /* The order matches include/orangefs-types.h in the OrangeFS source. */
476 static int PINT_errno_mapping[] = {
477 0, EPERM, ENOENT, EINTR, EIO, ENXIO, EBADF, EAGAIN, ENOMEM,
478 EFAULT, EBUSY, EEXIST, ENODEV, ENOTDIR, EISDIR, EINVAL, EMFILE,
479 EFBIG, ENOSPC, EROFS, EMLINK, EPIPE, EDEADLK, ENAMETOOLONG,
480 ENOLCK, ENOSYS, ENOTEMPTY, ELOOP, EWOULDBLOCK, ENOMSG, EUNATCH,
481 EBADR, EDEADLOCK, ENODATA, ETIME, ENONET, EREMOTE, ECOMM,
482 EPROTO, EBADMSG, EOVERFLOW, ERESTART, EMSGSIZE, EPROTOTYPE,
483 ENOPROTOOPT, EPROTONOSUPPORT, EOPNOTSUPP, EADDRINUSE,
484 EADDRNOTAVAIL, ENETDOWN, ENETUNREACH, ENETRESET, ENOBUFS,
485 ETIMEDOUT, ECONNREFUSED, EHOSTDOWN, EHOSTUNREACH, EALREADY,
486 EACCES, ECONNRESET, ERANGE
487 };
488
orangefs_normalize_to_errno(__s32 error_code)489 int orangefs_normalize_to_errno(__s32 error_code)
490 {
491 __u32 i;
492
493 /* Success */
494 if (error_code == 0) {
495 return 0;
496 /*
497 * This shouldn't ever happen. If it does it should be fixed on the
498 * server.
499 */
500 } else if (error_code > 0) {
501 gossip_err("orangefs: error status received.\n");
502 gossip_err("orangefs: assuming error code is inverted.\n");
503 error_code = -error_code;
504 }
505
506 /*
507 * XXX: This is very bad since error codes from ORANGEFS may not be
508 * suitable for return into userspace.
509 */
510
511 /*
512 * Convert ORANGEFS error values into errno values suitable for return
513 * from the kernel.
514 */
515 if ((-error_code) & ORANGEFS_NON_ERRNO_ERROR_BIT) {
516 if (((-error_code) &
517 (ORANGEFS_ERROR_NUMBER_BITS|ORANGEFS_NON_ERRNO_ERROR_BIT|
518 ORANGEFS_ERROR_BIT)) == ORANGEFS_ECANCEL) {
519 /*
520 * cancellation error codes generally correspond to
521 * a timeout from the client's perspective
522 */
523 error_code = -ETIMEDOUT;
524 } else {
525 /* assume a default error code */
526 gossip_err("%s: bad error code :%d:.\n",
527 __func__,
528 error_code);
529 error_code = -EINVAL;
530 }
531
532 /* Convert ORANGEFS encoded errno values into regular errno values. */
533 } else if ((-error_code) & ORANGEFS_ERROR_BIT) {
534 i = (-error_code) & ~(ORANGEFS_ERROR_BIT|ORANGEFS_ERROR_CLASS_BITS);
535 if (i < ARRAY_SIZE(PINT_errno_mapping))
536 error_code = -PINT_errno_mapping[i];
537 else
538 error_code = -EINVAL;
539
540 /*
541 * Only ORANGEFS protocol error codes should ever come here. Otherwise
542 * there is a bug somewhere.
543 */
544 } else {
545 gossip_err("%s: unknown error code.\n", __func__);
546 error_code = -EINVAL;
547 }
548 return error_code;
549 }
550
551 #define NUM_MODES 11
ORANGEFS_util_translate_mode(int mode)552 __s32 ORANGEFS_util_translate_mode(int mode)
553 {
554 int ret = 0;
555 int i = 0;
556 static int modes[NUM_MODES] = {
557 S_IXOTH, S_IWOTH, S_IROTH,
558 S_IXGRP, S_IWGRP, S_IRGRP,
559 S_IXUSR, S_IWUSR, S_IRUSR,
560 S_ISGID, S_ISUID
561 };
562 static int orangefs_modes[NUM_MODES] = {
563 ORANGEFS_O_EXECUTE, ORANGEFS_O_WRITE, ORANGEFS_O_READ,
564 ORANGEFS_G_EXECUTE, ORANGEFS_G_WRITE, ORANGEFS_G_READ,
565 ORANGEFS_U_EXECUTE, ORANGEFS_U_WRITE, ORANGEFS_U_READ,
566 ORANGEFS_G_SGID, ORANGEFS_U_SUID
567 };
568
569 for (i = 0; i < NUM_MODES; i++)
570 if (mode & modes[i])
571 ret |= orangefs_modes[i];
572
573 return ret;
574 }
575 #undef NUM_MODES
576