Commit | Line | Data |
---|---|---|
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" | |
575e9461 MM |
7 | #include "orangefs-kernel.h" |
8 | #include "orangefs-dev-proto.h" | |
9 | #include "orangefs-bufmap.h" | |
f7be4ee0 | 10 | |
8bb8aefd | 11 | __s32 fsid_of_op(struct orangefs_kernel_op_s *op) |
f7be4ee0 | 12 | { |
8bb8aefd | 13 | __s32 fsid = ORANGEFS_FS_ID_NULL; |
f7be4ee0 MM |
14 | |
15 | if (op) { | |
16 | switch (op->upcall.type) { | |
8bb8aefd | 17 | case ORANGEFS_VFS_OP_FILE_IO: |
f7be4ee0 MM |
18 | fsid = op->upcall.req.io.refn.fs_id; |
19 | break; | |
8bb8aefd | 20 | case ORANGEFS_VFS_OP_LOOKUP: |
f7be4ee0 MM |
21 | fsid = op->upcall.req.lookup.parent_refn.fs_id; |
22 | break; | |
8bb8aefd | 23 | case ORANGEFS_VFS_OP_CREATE: |
f7be4ee0 MM |
24 | fsid = op->upcall.req.create.parent_refn.fs_id; |
25 | break; | |
8bb8aefd | 26 | case ORANGEFS_VFS_OP_GETATTR: |
f7be4ee0 MM |
27 | fsid = op->upcall.req.getattr.refn.fs_id; |
28 | break; | |
8bb8aefd | 29 | case ORANGEFS_VFS_OP_REMOVE: |
f7be4ee0 MM |
30 | fsid = op->upcall.req.remove.parent_refn.fs_id; |
31 | break; | |
8bb8aefd | 32 | case ORANGEFS_VFS_OP_MKDIR: |
f7be4ee0 MM |
33 | fsid = op->upcall.req.mkdir.parent_refn.fs_id; |
34 | break; | |
8bb8aefd | 35 | case ORANGEFS_VFS_OP_READDIR: |
f7be4ee0 MM |
36 | fsid = op->upcall.req.readdir.refn.fs_id; |
37 | break; | |
8bb8aefd | 38 | case ORANGEFS_VFS_OP_SETATTR: |
f7be4ee0 MM |
39 | fsid = op->upcall.req.setattr.refn.fs_id; |
40 | break; | |
8bb8aefd | 41 | case ORANGEFS_VFS_OP_SYMLINK: |
f7be4ee0 MM |
42 | fsid = op->upcall.req.sym.parent_refn.fs_id; |
43 | break; | |
8bb8aefd | 44 | case ORANGEFS_VFS_OP_RENAME: |
f7be4ee0 MM |
45 | fsid = op->upcall.req.rename.old_parent_refn.fs_id; |
46 | break; | |
8bb8aefd | 47 | case ORANGEFS_VFS_OP_STATFS: |
f7be4ee0 MM |
48 | fsid = op->upcall.req.statfs.fs_id; |
49 | break; | |
8bb8aefd | 50 | case ORANGEFS_VFS_OP_TRUNCATE: |
f7be4ee0 MM |
51 | fsid = op->upcall.req.truncate.refn.fs_id; |
52 | break; | |
8bb8aefd | 53 | case ORANGEFS_VFS_OP_MMAP_RA_FLUSH: |
f7be4ee0 MM |
54 | fsid = op->upcall.req.ra_cache_flush.refn.fs_id; |
55 | break; | |
8bb8aefd | 56 | case ORANGEFS_VFS_OP_FS_UMOUNT: |
f7be4ee0 MM |
57 | fsid = op->upcall.req.fs_umount.fs_id; |
58 | break; | |
8bb8aefd | 59 | case ORANGEFS_VFS_OP_GETXATTR: |
f7be4ee0 MM |
60 | fsid = op->upcall.req.getxattr.refn.fs_id; |
61 | break; | |
8bb8aefd | 62 | case ORANGEFS_VFS_OP_SETXATTR: |
f7be4ee0 MM |
63 | fsid = op->upcall.req.setxattr.refn.fs_id; |
64 | break; | |
8bb8aefd | 65 | case ORANGEFS_VFS_OP_LISTXATTR: |
f7be4ee0 MM |
66 | fsid = op->upcall.req.listxattr.refn.fs_id; |
67 | break; | |
8bb8aefd | 68 | case ORANGEFS_VFS_OP_REMOVEXATTR: |
f7be4ee0 MM |
69 | fsid = op->upcall.req.removexattr.refn.fs_id; |
70 | break; | |
8bb8aefd | 71 | case ORANGEFS_VFS_OP_FSYNC: |
f7be4ee0 MM |
72 | fsid = op->upcall.req.fsync.refn.fs_id; |
73 | break; | |
74 | default: | |
75 | break; | |
76 | } | |
77 | } | |
78 | return fsid; | |
79 | } | |
80 | ||
394f647e | 81 | static int orangefs_inode_flags(struct ORANGEFS_sys_attr_s *attrs) |
f7be4ee0 | 82 | { |
394f647e | 83 | int flags = 0; |
8bb8aefd | 84 | if (attrs->flags & ORANGEFS_IMMUTABLE_FL) |
394f647e | 85 | flags |= S_IMMUTABLE; |
f7be4ee0 | 86 | else |
394f647e | 87 | flags &= ~S_IMMUTABLE; |
8bb8aefd | 88 | if (attrs->flags & ORANGEFS_APPEND_FL) |
394f647e | 89 | flags |= S_APPEND; |
f7be4ee0 | 90 | else |
394f647e | 91 | flags &= ~S_APPEND; |
8bb8aefd | 92 | if (attrs->flags & ORANGEFS_NOATIME_FL) |
394f647e | 93 | flags |= S_NOATIME; |
f7be4ee0 | 94 | else |
394f647e MB |
95 | flags &= ~S_NOATIME; |
96 | return flags; | |
97 | } | |
98 | ||
99 | static int orangefs_inode_perms(struct ORANGEFS_sys_attr_s *attrs) | |
100 | { | |
101 | int perm_mode = 0; | |
102 | ||
103 | if (attrs->perms & ORANGEFS_O_EXECUTE) | |
104 | perm_mode |= S_IXOTH; | |
105 | if (attrs->perms & ORANGEFS_O_WRITE) | |
106 | perm_mode |= S_IWOTH; | |
107 | if (attrs->perms & ORANGEFS_O_READ) | |
108 | perm_mode |= S_IROTH; | |
109 | ||
110 | if (attrs->perms & ORANGEFS_G_EXECUTE) | |
111 | perm_mode |= S_IXGRP; | |
112 | if (attrs->perms & ORANGEFS_G_WRITE) | |
113 | perm_mode |= S_IWGRP; | |
114 | if (attrs->perms & ORANGEFS_G_READ) | |
115 | perm_mode |= S_IRGRP; | |
f7be4ee0 | 116 | |
394f647e MB |
117 | if (attrs->perms & ORANGEFS_U_EXECUTE) |
118 | perm_mode |= S_IXUSR; | |
119 | if (attrs->perms & ORANGEFS_U_WRITE) | |
120 | perm_mode |= S_IWUSR; | |
121 | if (attrs->perms & ORANGEFS_U_READ) | |
122 | perm_mode |= S_IRUSR; | |
123 | ||
124 | if (attrs->perms & ORANGEFS_G_SGID) | |
125 | perm_mode |= S_ISGID; | |
126 | if (attrs->perms & ORANGEFS_U_SUID) | |
127 | perm_mode |= S_ISUID; | |
128 | ||
129 | return perm_mode; | |
f7be4ee0 MM |
130 | } |
131 | ||
132 | /* NOTE: symname is ignored unless the inode is a sym link */ | |
133 | static int copy_attributes_to_inode(struct inode *inode, | |
8bb8aefd | 134 | struct ORANGEFS_sys_attr_s *attrs, |
f7be4ee0 MM |
135 | char *symname) |
136 | { | |
137 | int ret = -1; | |
8bb8aefd | 138 | struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); |
f7be4ee0 MM |
139 | loff_t inode_size = 0; |
140 | loff_t rounded_up_size = 0; | |
141 | ||
142 | ||
143 | /* | |
54804949 MM |
144 | * arbitrarily set the inode block size; FIXME: we need to |
145 | * resolve the difference between the reported inode blocksize | |
146 | * and the PAGE_CACHE_SIZE, since our block count will always | |
147 | * be wrong. | |
148 | * | |
149 | * For now, we're setting the block count to be the proper | |
150 | * number assuming the block size is 512 bytes, and the size is | |
151 | * rounded up to the nearest 4K. This is apparently required | |
152 | * to get proper size reports from the 'du' shell utility. | |
153 | * | |
154 | * changing the inode->i_blkbits to something other than | |
155 | * PAGE_CACHE_SHIFT breaks mmap/execution as we depend on that. | |
f7be4ee0 MM |
156 | */ |
157 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
158 | "attrs->mask = %x (objtype = %s)\n", | |
159 | attrs->mask, | |
8bb8aefd YL |
160 | attrs->objtype == ORANGEFS_TYPE_METAFILE ? "file" : |
161 | attrs->objtype == ORANGEFS_TYPE_DIRECTORY ? "directory" : | |
162 | attrs->objtype == ORANGEFS_TYPE_SYMLINK ? "symlink" : | |
f7be4ee0 MM |
163 | "invalid/unknown"); |
164 | ||
165 | switch (attrs->objtype) { | |
8bb8aefd | 166 | case ORANGEFS_TYPE_METAFILE: |
394f647e | 167 | inode->i_flags = orangefs_inode_flags(attrs); |
8bb8aefd | 168 | if (attrs->mask & ORANGEFS_ATTR_SYS_SIZE) { |
f7be4ee0 MM |
169 | inode_size = (loff_t) attrs->size; |
170 | rounded_up_size = | |
171 | (inode_size + (4096 - (inode_size % 4096))); | |
172 | ||
d57521a6 | 173 | spin_lock(&inode->i_lock); |
f7be4ee0 MM |
174 | inode->i_bytes = inode_size; |
175 | inode->i_blocks = | |
176 | (unsigned long)(rounded_up_size / 512); | |
d57521a6 | 177 | spin_unlock(&inode->i_lock); |
f7be4ee0 MM |
178 | |
179 | /* | |
180 | * NOTE: make sure all the places we're called | |
181 | * from have the inode->i_sem lock. We're fine | |
182 | * in 99% of the cases since we're mostly | |
183 | * called from a lookup. | |
184 | */ | |
185 | inode->i_size = inode_size; | |
186 | } | |
187 | break; | |
8bb8aefd | 188 | case ORANGEFS_TYPE_SYMLINK: |
f7be4ee0 MM |
189 | if (symname != NULL) { |
190 | inode->i_size = (loff_t) strlen(symname); | |
191 | break; | |
192 | } | |
193 | /*FALLTHRU*/ | |
194 | default: | |
f0566532 MM |
195 | inode->i_size = PAGE_CACHE_SIZE; |
196 | ||
d57521a6 | 197 | spin_lock(&inode->i_lock); |
f0566532 | 198 | inode_set_bytes(inode, inode->i_size); |
d57521a6 | 199 | spin_unlock(&inode->i_lock); |
f7be4ee0 MM |
200 | break; |
201 | } | |
202 | ||
203 | inode->i_uid = make_kuid(&init_user_ns, attrs->owner); | |
204 | inode->i_gid = make_kgid(&init_user_ns, attrs->group); | |
be81ce48 AB |
205 | inode->i_atime.tv_sec = (time64_t) attrs->atime; |
206 | inode->i_mtime.tv_sec = (time64_t) attrs->mtime; | |
207 | inode->i_ctime.tv_sec = (time64_t) attrs->ctime; | |
f7be4ee0 MM |
208 | inode->i_atime.tv_nsec = 0; |
209 | inode->i_mtime.tv_nsec = 0; | |
210 | inode->i_ctime.tv_nsec = 0; | |
211 | ||
394f647e | 212 | inode->i_mode = orangefs_inode_perms(attrs); |
f7be4ee0 MM |
213 | |
214 | if (is_root_handle(inode)) { | |
215 | /* special case: mark the root inode as sticky */ | |
216 | inode->i_mode |= S_ISVTX; | |
217 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
218 | "Marking inode %pU as sticky\n", | |
219 | get_khandle_from_ino(inode)); | |
220 | } | |
221 | ||
222 | switch (attrs->objtype) { | |
8bb8aefd | 223 | case ORANGEFS_TYPE_METAFILE: |
f7be4ee0 MM |
224 | inode->i_mode |= S_IFREG; |
225 | ret = 0; | |
226 | break; | |
8bb8aefd | 227 | case ORANGEFS_TYPE_DIRECTORY: |
f7be4ee0 MM |
228 | inode->i_mode |= S_IFDIR; |
229 | /* NOTE: we have no good way to keep nlink consistent | |
230 | * for directories across clients; keep constant at 1. | |
231 | * Why 1? If we go with 2, then find(1) gets confused | |
232 | * and won't work properly withouth the -noleaf option | |
233 | */ | |
234 | set_nlink(inode, 1); | |
235 | ret = 0; | |
236 | break; | |
8bb8aefd | 237 | case ORANGEFS_TYPE_SYMLINK: |
f7be4ee0 MM |
238 | inode->i_mode |= S_IFLNK; |
239 | ||
240 | /* copy link target to inode private data */ | |
8bb8aefd YL |
241 | if (orangefs_inode && symname) { |
242 | strncpy(orangefs_inode->link_target, | |
f7be4ee0 | 243 | symname, |
8bb8aefd | 244 | ORANGEFS_NAME_MAX); |
f7be4ee0 MM |
245 | gossip_debug(GOSSIP_UTILS_DEBUG, |
246 | "Copied attr link target %s\n", | |
8bb8aefd | 247 | orangefs_inode->link_target); |
f7be4ee0 MM |
248 | } |
249 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
250 | "symlink mode %o\n", | |
251 | inode->i_mode); | |
252 | ret = 0; | |
253 | break; | |
254 | default: | |
8bb8aefd | 255 | gossip_err("orangefs: copy_attributes_to_inode: got invalid attribute type %x\n", |
f7be4ee0 MM |
256 | attrs->objtype); |
257 | } | |
258 | ||
259 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
8bb8aefd | 260 | "orangefs: copy_attributes_to_inode: setting i_mode to %o, i_size to %lu\n", |
f7be4ee0 MM |
261 | inode->i_mode, |
262 | (unsigned long)i_size_read(inode)); | |
263 | ||
264 | return ret; | |
265 | } | |
266 | ||
267 | /* | |
268 | * NOTE: in kernel land, we never use the sys_attr->link_target for | |
269 | * anything, so don't bother copying it into the sys_attr object here. | |
270 | */ | |
271 | static inline int copy_attributes_from_inode(struct inode *inode, | |
8bb8aefd | 272 | struct ORANGEFS_sys_attr_s *attrs, |
f7be4ee0 MM |
273 | struct iattr *iattr) |
274 | { | |
275 | umode_t tmp_mode; | |
276 | ||
277 | if (!iattr || !inode || !attrs) { | |
278 | gossip_err("NULL iattr (%p), inode (%p), attrs (%p) " | |
279 | "in copy_attributes_from_inode!\n", | |
280 | iattr, | |
281 | inode, | |
282 | attrs); | |
283 | return -EINVAL; | |
284 | } | |
285 | /* | |
286 | * We need to be careful to only copy the attributes out of the | |
287 | * iattr object that we know are valid. | |
288 | */ | |
289 | attrs->mask = 0; | |
290 | if (iattr->ia_valid & ATTR_UID) { | |
291 | attrs->owner = from_kuid(current_user_ns(), iattr->ia_uid); | |
8bb8aefd | 292 | attrs->mask |= ORANGEFS_ATTR_SYS_UID; |
f7be4ee0 MM |
293 | gossip_debug(GOSSIP_UTILS_DEBUG, "(UID) %d\n", attrs->owner); |
294 | } | |
295 | if (iattr->ia_valid & ATTR_GID) { | |
296 | attrs->group = from_kgid(current_user_ns(), iattr->ia_gid); | |
8bb8aefd | 297 | attrs->mask |= ORANGEFS_ATTR_SYS_GID; |
f7be4ee0 MM |
298 | gossip_debug(GOSSIP_UTILS_DEBUG, "(GID) %d\n", attrs->group); |
299 | } | |
300 | ||
301 | if (iattr->ia_valid & ATTR_ATIME) { | |
8bb8aefd | 302 | attrs->mask |= ORANGEFS_ATTR_SYS_ATIME; |
f7be4ee0 | 303 | if (iattr->ia_valid & ATTR_ATIME_SET) { |
be81ce48 | 304 | attrs->atime = (time64_t)iattr->ia_atime.tv_sec; |
8bb8aefd | 305 | attrs->mask |= ORANGEFS_ATTR_SYS_ATIME_SET; |
f7be4ee0 MM |
306 | } |
307 | } | |
308 | if (iattr->ia_valid & ATTR_MTIME) { | |
8bb8aefd | 309 | attrs->mask |= ORANGEFS_ATTR_SYS_MTIME; |
f7be4ee0 | 310 | if (iattr->ia_valid & ATTR_MTIME_SET) { |
be81ce48 | 311 | attrs->mtime = (time64_t)iattr->ia_mtime.tv_sec; |
8bb8aefd | 312 | attrs->mask |= ORANGEFS_ATTR_SYS_MTIME_SET; |
f7be4ee0 MM |
313 | } |
314 | } | |
315 | if (iattr->ia_valid & ATTR_CTIME) | |
8bb8aefd | 316 | attrs->mask |= ORANGEFS_ATTR_SYS_CTIME; |
f7be4ee0 MM |
317 | |
318 | /* | |
8bb8aefd | 319 | * ORANGEFS cannot set size with a setattr operation. Probably not likely |
f7be4ee0 MM |
320 | * to be requested through the VFS, but just in case, don't worry about |
321 | * ATTR_SIZE | |
322 | */ | |
323 | ||
324 | if (iattr->ia_valid & ATTR_MODE) { | |
325 | tmp_mode = iattr->ia_mode; | |
326 | if (tmp_mode & (S_ISVTX)) { | |
327 | if (is_root_handle(inode)) { | |
328 | /* | |
329 | * allow sticky bit to be set on root (since | |
330 | * it shows up that way by default anyhow), | |
331 | * but don't show it to the server | |
332 | */ | |
333 | tmp_mode -= S_ISVTX; | |
334 | } else { | |
335 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
336 | "User attempted to set sticky bit on non-root directory; returning EINVAL.\n"); | |
337 | return -EINVAL; | |
338 | } | |
339 | } | |
340 | ||
341 | if (tmp_mode & (S_ISUID)) { | |
342 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
343 | "Attempting to set setuid bit (not supported); returning EINVAL.\n"); | |
344 | return -EINVAL; | |
345 | } | |
346 | ||
8bb8aefd YL |
347 | attrs->perms = ORANGEFS_util_translate_mode(tmp_mode); |
348 | attrs->mask |= ORANGEFS_ATTR_SYS_PERM; | |
f7be4ee0 MM |
349 | } |
350 | ||
351 | return 0; | |
352 | } | |
353 | ||
99109822 MB |
354 | static int compare_attributes_to_inode(struct inode *inode, |
355 | struct ORANGEFS_sys_attr_s *attrs, | |
fe88adc3 MB |
356 | char *symname, |
357 | int mask) | |
99109822 MB |
358 | { |
359 | struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); | |
360 | loff_t inode_size, rounded_up_size; | |
361 | ||
fe88adc3 MB |
362 | /* Much of what happens below relies on the type being around. */ |
363 | if (!(mask & ORANGEFS_ATTR_SYS_TYPE)) | |
364 | return 0; | |
365 | ||
366 | if (attrs->objtype == ORANGEFS_TYPE_METAFILE && | |
367 | inode->i_flags != orangefs_inode_flags(attrs)) | |
368 | return 0; | |
369 | ||
99109822 MB |
370 | /* Compare file size. */ |
371 | ||
372 | switch (attrs->objtype) { | |
373 | case ORANGEFS_TYPE_METAFILE: | |
fe88adc3 MB |
374 | if (mask & ORANGEFS_ATTR_SYS_SIZE) { |
375 | inode_size = attrs->size; | |
376 | rounded_up_size = inode_size + | |
377 | (4096 - (inode_size % 4096)); | |
378 | if (inode->i_bytes != inode_size || | |
379 | inode->i_blocks != rounded_up_size/512) | |
380 | return 0; | |
381 | } | |
99109822 MB |
382 | break; |
383 | case ORANGEFS_TYPE_SYMLINK: | |
fe88adc3 MB |
384 | if (mask & ORANGEFS_ATTR_SYS_SIZE) |
385 | if (symname && strlen(symname) != inode->i_size) | |
386 | return 0; | |
99109822 MB |
387 | break; |
388 | default: | |
389 | if (inode->i_size != PAGE_CACHE_SIZE && | |
390 | inode_get_bytes(inode) != PAGE_CACHE_SIZE) | |
391 | return 0; | |
392 | } | |
393 | ||
394 | /* Compare general attributes. */ | |
395 | ||
fe88adc3 MB |
396 | if (mask & ORANGEFS_ATTR_SYS_UID && |
397 | !uid_eq(inode->i_uid, make_kuid(&init_user_ns, attrs->owner))) | |
398 | return 0; | |
399 | if (mask & ORANGEFS_ATTR_SYS_GID && | |
400 | !gid_eq(inode->i_gid, make_kgid(&init_user_ns, attrs->group))) | |
401 | return 0; | |
402 | if (mask & ORANGEFS_ATTR_SYS_ATIME && | |
403 | inode->i_atime.tv_sec != attrs->atime) | |
404 | return 0; | |
405 | if (mask & ORANGEFS_ATTR_SYS_MTIME && | |
406 | inode->i_atime.tv_sec != attrs->mtime) | |
407 | return 0; | |
408 | if (mask & ORANGEFS_ATTR_SYS_CTIME && | |
409 | inode->i_atime.tv_sec != attrs->ctime) | |
410 | return 0; | |
411 | if (inode->i_atime.tv_nsec != 0 || | |
99109822 MB |
412 | inode->i_mtime.tv_nsec != 0 || |
413 | inode->i_ctime.tv_nsec != 0) | |
414 | return 0; | |
415 | ||
fe88adc3 MB |
416 | if (mask & ORANGEFS_ATTR_SYS_PERM && |
417 | (inode->i_mode & ~(S_ISVTX|S_IFREG|S_IFDIR|S_IFLNK)) != | |
99109822 MB |
418 | orangefs_inode_perms(attrs)) |
419 | return 0; | |
420 | ||
421 | if (is_root_handle(inode)) | |
422 | if (!(inode->i_mode & S_ISVTX)) | |
423 | return 0; | |
424 | ||
425 | /* Compare file type. */ | |
426 | ||
427 | switch (attrs->objtype) { | |
428 | case ORANGEFS_TYPE_METAFILE: | |
cf22644a | 429 | if (!S_ISREG(inode->i_mode)) |
99109822 MB |
430 | return 0; |
431 | break; | |
432 | case ORANGEFS_TYPE_DIRECTORY: | |
cf22644a | 433 | if (!S_ISDIR(inode->i_mode)) |
99109822 MB |
434 | return 0; |
435 | if (inode->i_nlink != 1) | |
436 | return 0; | |
437 | break; | |
438 | case ORANGEFS_TYPE_SYMLINK: | |
cf22644a | 439 | if (!S_ISLNK(inode->i_mode)) |
99109822 | 440 | return 0; |
fe88adc3 MB |
441 | if (orangefs_inode && symname && |
442 | mask & ORANGEFS_ATTR_SYS_LNK_TARGET) | |
99109822 MB |
443 | if (strcmp(orangefs_inode->link_target, symname)) |
444 | return 0; | |
445 | break; | |
446 | default: | |
447 | gossip_err("orangefs: compare_attributes_to_inode: got invalid attribute type %x\n", | |
448 | attrs->objtype); | |
449 | ||
450 | } | |
451 | ||
452 | return 1; | |
453 | } | |
454 | ||
f7be4ee0 | 455 | /* |
99109822 MB |
456 | * Issues a orangefs getattr request and fills in the appropriate inode |
457 | * attributes if successful. When check is 0, returns 0 on success and -errno | |
458 | * otherwise. When check is 1, returns 1 on success where the inode is valid | |
459 | * and 0 on success where the inode is stale and -errno otherwise. | |
f7be4ee0 | 460 | */ |
3c9cf98d MB |
461 | int orangefs_inode_old_getattr(struct inode *inode, __u32 getattr_mask, |
462 | int check) | |
f7be4ee0 | 463 | { |
8bb8aefd YL |
464 | struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); |
465 | struct orangefs_kernel_op_s *new_op; | |
f7be4ee0 MM |
466 | int ret = -EINVAL; |
467 | ||
468 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
469 | "%s: called on inode %pU\n", | |
470 | __func__, | |
471 | get_khandle_from_ino(inode)); | |
472 | ||
8bb8aefd | 473 | new_op = op_alloc(ORANGEFS_VFS_OP_GETATTR); |
f7be4ee0 MM |
474 | if (!new_op) |
475 | return -ENOMEM; | |
8bb8aefd | 476 | new_op->upcall.req.getattr.refn = orangefs_inode->refn; |
f7be4ee0 MM |
477 | new_op->upcall.req.getattr.mask = getattr_mask; |
478 | ||
479 | ret = service_operation(new_op, __func__, | |
480 | get_interruptible_flag(inode)); | |
481 | if (ret != 0) | |
482 | goto out; | |
483 | ||
99109822 MB |
484 | if (check) { |
485 | ret = compare_attributes_to_inode(inode, | |
486 | &new_op->downcall.resp.getattr.attributes, | |
fe88adc3 MB |
487 | new_op->downcall.resp.getattr.link_target, |
488 | getattr_mask); | |
99109822 MB |
489 | |
490 | if (new_op->downcall.resp.getattr.attributes.objtype == | |
491 | ORANGEFS_TYPE_METAFILE) { | |
492 | if (orangefs_inode->blksize != | |
493 | new_op->downcall.resp.getattr.attributes.blksize) | |
494 | ret = 0; | |
495 | } else { | |
496 | if (orangefs_inode->blksize != 1 << inode->i_blkbits) | |
497 | ret = 0; | |
498 | } | |
f7be4ee0 | 499 | } else { |
99109822 MB |
500 | if (copy_attributes_to_inode(inode, |
501 | &new_op->downcall.resp.getattr.attributes, | |
502 | new_op->downcall.resp.getattr.link_target)) { | |
503 | gossip_err("%s: failed to copy attributes\n", __func__); | |
504 | ret = -ENOENT; | |
505 | goto out; | |
506 | } | |
f7be4ee0 | 507 | |
99109822 MB |
508 | /* |
509 | * Store blksize in orangefs specific part of inode structure; | |
510 | * we are only going to use this to report to stat to make sure | |
511 | * it doesn't perturb any inode related code paths. | |
512 | */ | |
513 | if (new_op->downcall.resp.getattr.attributes.objtype == | |
514 | ORANGEFS_TYPE_METAFILE) { | |
515 | orangefs_inode->blksize = new_op->downcall.resp. | |
516 | getattr.attributes.blksize; | |
517 | } else { | |
518 | /* | |
519 | * mimic behavior of generic_fillattr() for other file | |
520 | * types. | |
521 | */ | |
522 | orangefs_inode->blksize = (1 << inode->i_blkbits); | |
523 | ||
524 | } | |
f7be4ee0 MM |
525 | } |
526 | ||
527 | out: | |
528 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
529 | "Getattr on handle %pU, " | |
530 | "fsid %d\n (inode ct = %d) returned %d\n", | |
8bb8aefd YL |
531 | &orangefs_inode->refn.khandle, |
532 | orangefs_inode->refn.fs_id, | |
f7be4ee0 MM |
533 | (int)atomic_read(&inode->i_count), |
534 | ret); | |
535 | ||
536 | op_release(new_op); | |
537 | return ret; | |
538 | } | |
539 | ||
3c9cf98d MB |
540 | static int orangefs_inode_type(enum orangefs_ds_type objtype) |
541 | { | |
542 | if (objtype == ORANGEFS_TYPE_METAFILE) | |
543 | return S_IFREG; | |
544 | else if (objtype == ORANGEFS_TYPE_DIRECTORY) | |
545 | return S_IFDIR; | |
546 | else if (objtype == ORANGEFS_TYPE_SYMLINK) | |
547 | return S_IFLNK; | |
548 | else | |
549 | return -1; | |
550 | } | |
551 | ||
552 | int orangefs_inode_getattr(struct inode *inode, int new, int size) | |
553 | { | |
554 | struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); | |
555 | struct orangefs_kernel_op_s *new_op; | |
556 | loff_t inode_size, rounded_up_size; | |
557 | int ret; | |
558 | ||
559 | gossip_debug(GOSSIP_UTILS_DEBUG, "%s: called on inode %pU\n", __func__, | |
560 | get_khandle_from_ino(inode)); | |
561 | ||
562 | new_op = op_alloc(ORANGEFS_VFS_OP_GETATTR); | |
563 | if (!new_op) | |
564 | return -ENOMEM; | |
565 | new_op->upcall.req.getattr.refn = orangefs_inode->refn; | |
566 | new_op->upcall.req.getattr.mask = size ? | |
567 | ORANGEFS_ATTR_SYS_ALL_NOHINT : ORANGEFS_ATTR_SYS_ALL_NOHINT_NOSIZE; | |
568 | ||
569 | ret = service_operation(new_op, __func__, | |
570 | get_interruptible_flag(inode)); | |
571 | if (ret != 0) | |
572 | goto out; | |
573 | ||
574 | ret = orangefs_inode_type(new_op-> | |
575 | downcall.resp.getattr.attributes.objtype); | |
576 | if (!new) { | |
577 | /* | |
578 | * If the inode type or symlink target have changed then this | |
579 | * inode is stale. | |
580 | */ | |
581 | if (ret == -1 || !(inode->i_mode & ret)) { | |
582 | orangefs_make_bad_inode(inode); | |
583 | ret = -ESTALE; | |
584 | goto out; | |
585 | } | |
586 | if (ret == S_IFLNK && strncmp(orangefs_inode->link_target, | |
587 | new_op->downcall.resp.getattr.link_target, | |
588 | ORANGEFS_NAME_MAX)) { | |
589 | orangefs_make_bad_inode(inode); | |
590 | ret = -ESTALE; | |
591 | goto out; | |
592 | } | |
593 | } | |
594 | ||
595 | switch (ret) { | |
596 | case S_IFREG: | |
597 | inode->i_flags = orangefs_inode_flags(&new_op-> | |
598 | downcall.resp.getattr.attributes); | |
599 | if (size) { | |
600 | inode_size = (loff_t)new_op-> | |
601 | downcall.resp.getattr.attributes.size; | |
602 | rounded_up_size = | |
603 | (inode_size + (4096 - (inode_size % 4096))); | |
604 | inode->i_size = inode_size; | |
605 | orangefs_inode->blksize = | |
606 | new_op->downcall.resp.getattr.attributes.blksize; | |
607 | spin_lock(&inode->i_lock); | |
608 | inode->i_bytes = inode_size; | |
609 | inode->i_blocks = | |
610 | (unsigned long)(rounded_up_size / 512); | |
611 | spin_unlock(&inode->i_lock); | |
612 | } | |
613 | break; | |
614 | case S_IFDIR: | |
615 | inode->i_size = PAGE_CACHE_SIZE; | |
616 | orangefs_inode->blksize = (1 << inode->i_blkbits); | |
617 | spin_lock(&inode->i_lock); | |
618 | inode_set_bytes(inode, inode->i_size); | |
619 | spin_unlock(&inode->i_lock); | |
620 | set_nlink(inode, 1); | |
621 | break; | |
622 | case S_IFLNK: | |
623 | if (new) { | |
624 | inode->i_size = (loff_t)strlen(new_op-> | |
625 | downcall.resp.getattr.link_target); | |
626 | orangefs_inode->blksize = (1 << inode->i_blkbits); | |
627 | strlcpy(orangefs_inode->link_target, | |
628 | new_op->downcall.resp.getattr.link_target, | |
629 | ORANGEFS_NAME_MAX); | |
630 | } | |
631 | break; | |
632 | } | |
633 | ||
634 | inode->i_uid = make_kuid(&init_user_ns, new_op-> | |
635 | downcall.resp.getattr.attributes.owner); | |
636 | inode->i_gid = make_kgid(&init_user_ns, new_op-> | |
637 | downcall.resp.getattr.attributes.group); | |
638 | inode->i_atime.tv_sec = (time64_t)new_op-> | |
639 | downcall.resp.getattr.attributes.atime; | |
640 | inode->i_mtime.tv_sec = (time64_t)new_op-> | |
641 | downcall.resp.getattr.attributes.mtime; | |
642 | inode->i_ctime.tv_sec = (time64_t)new_op-> | |
643 | downcall.resp.getattr.attributes.ctime; | |
644 | inode->i_atime.tv_nsec = 0; | |
645 | inode->i_mtime.tv_nsec = 0; | |
646 | inode->i_ctime.tv_nsec = 0; | |
647 | ||
648 | /* special case: mark the root inode as sticky */ | |
649 | inode->i_mode = ret | (is_root_handle(inode) ? S_ISVTX : 0) | | |
650 | orangefs_inode_perms(&new_op->downcall.resp.getattr.attributes); | |
651 | ||
652 | ret = 0; | |
653 | out: | |
654 | op_release(new_op); | |
655 | return ret; | |
656 | } | |
657 | ||
f7be4ee0 | 658 | /* |
8bb8aefd | 659 | * issues a orangefs setattr request to make sure the new attribute values |
f7be4ee0 MM |
660 | * take effect if successful. returns 0 on success; -errno otherwise |
661 | */ | |
8bb8aefd | 662 | int orangefs_inode_setattr(struct inode *inode, struct iattr *iattr) |
f7be4ee0 | 663 | { |
8bb8aefd YL |
664 | struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); |
665 | struct orangefs_kernel_op_s *new_op; | |
f7be4ee0 MM |
666 | int ret; |
667 | ||
8bb8aefd | 668 | new_op = op_alloc(ORANGEFS_VFS_OP_SETATTR); |
f7be4ee0 MM |
669 | if (!new_op) |
670 | return -ENOMEM; | |
671 | ||
8bb8aefd | 672 | new_op->upcall.req.setattr.refn = orangefs_inode->refn; |
f7be4ee0 MM |
673 | ret = copy_attributes_from_inode(inode, |
674 | &new_op->upcall.req.setattr.attributes, | |
675 | iattr); | |
ed42fe05 AV |
676 | if (ret >= 0) { |
677 | ret = service_operation(new_op, __func__, | |
f7be4ee0 MM |
678 | get_interruptible_flag(inode)); |
679 | ||
ed42fe05 AV |
680 | gossip_debug(GOSSIP_UTILS_DEBUG, |
681 | "orangefs_inode_setattr: returning %d\n", | |
682 | ret); | |
683 | } | |
f7be4ee0 | 684 | |
f7be4ee0 MM |
685 | op_release(new_op); |
686 | ||
687 | /* | |
688 | * successful setattr should clear the atime, mtime and | |
689 | * ctime flags. | |
690 | */ | |
691 | if (ret == 0) { | |
8bb8aefd YL |
692 | ClearAtimeFlag(orangefs_inode); |
693 | ClearMtimeFlag(orangefs_inode); | |
694 | ClearCtimeFlag(orangefs_inode); | |
695 | ClearModeFlag(orangefs_inode); | |
f7be4ee0 MM |
696 | } |
697 | ||
698 | return ret; | |
699 | } | |
700 | ||
8bb8aefd | 701 | int orangefs_flush_inode(struct inode *inode) |
f7be4ee0 MM |
702 | { |
703 | /* | |
704 | * If it is a dirty inode, this function gets called. | |
705 | * Gather all the information that needs to be setattr'ed | |
706 | * Right now, this will only be used for mode, atime, mtime | |
707 | * and/or ctime. | |
708 | */ | |
709 | struct iattr wbattr; | |
710 | int ret; | |
711 | int mtime_flag; | |
712 | int ctime_flag; | |
713 | int atime_flag; | |
714 | int mode_flag; | |
8bb8aefd | 715 | struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); |
f7be4ee0 MM |
716 | |
717 | memset(&wbattr, 0, sizeof(wbattr)); | |
718 | ||
719 | /* | |
720 | * check inode flags up front, and clear them if they are set. This | |
721 | * will prevent multiple processes from all trying to flush the same | |
722 | * inode if they call close() simultaneously | |
723 | */ | |
8bb8aefd YL |
724 | mtime_flag = MtimeFlag(orangefs_inode); |
725 | ClearMtimeFlag(orangefs_inode); | |
726 | ctime_flag = CtimeFlag(orangefs_inode); | |
727 | ClearCtimeFlag(orangefs_inode); | |
728 | atime_flag = AtimeFlag(orangefs_inode); | |
729 | ClearAtimeFlag(orangefs_inode); | |
730 | mode_flag = ModeFlag(orangefs_inode); | |
731 | ClearModeFlag(orangefs_inode); | |
f7be4ee0 MM |
732 | |
733 | /* -- Lazy atime,mtime and ctime update -- | |
734 | * Note: all times are dictated by server in the new scheme | |
735 | * and not by the clients | |
736 | * | |
737 | * Also mode updates are being handled now.. | |
738 | */ | |
739 | ||
740 | if (mtime_flag) | |
741 | wbattr.ia_valid |= ATTR_MTIME; | |
742 | if (ctime_flag) | |
743 | wbattr.ia_valid |= ATTR_CTIME; | |
744 | if (atime_flag) | |
745 | wbattr.ia_valid |= ATTR_ATIME; | |
746 | ||
747 | if (mode_flag) { | |
748 | wbattr.ia_mode = inode->i_mode; | |
749 | wbattr.ia_valid |= ATTR_MODE; | |
750 | } | |
751 | ||
752 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
8bb8aefd | 753 | "*********** orangefs_flush_inode: %pU " |
f7be4ee0 MM |
754 | "(ia_valid %d)\n", |
755 | get_khandle_from_ino(inode), | |
756 | wbattr.ia_valid); | |
757 | if (wbattr.ia_valid == 0) { | |
758 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
8bb8aefd | 759 | "orangefs_flush_inode skipping setattr()\n"); |
f7be4ee0 MM |
760 | return 0; |
761 | } | |
762 | ||
763 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
8bb8aefd | 764 | "orangefs_flush_inode (%pU) writing mode %o\n", |
f7be4ee0 MM |
765 | get_khandle_from_ino(inode), |
766 | inode->i_mode); | |
767 | ||
8bb8aefd | 768 | ret = orangefs_inode_setattr(inode, &wbattr); |
f7be4ee0 MM |
769 | |
770 | return ret; | |
771 | } | |
772 | ||
8bb8aefd | 773 | int orangefs_unmount_sb(struct super_block *sb) |
f7be4ee0 MM |
774 | { |
775 | int ret = -EINVAL; | |
8bb8aefd | 776 | struct orangefs_kernel_op_s *new_op = NULL; |
f7be4ee0 MM |
777 | |
778 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
8bb8aefd | 779 | "orangefs_unmount_sb called on sb %p\n", |
f7be4ee0 MM |
780 | sb); |
781 | ||
8bb8aefd | 782 | new_op = op_alloc(ORANGEFS_VFS_OP_FS_UMOUNT); |
f7be4ee0 MM |
783 | if (!new_op) |
784 | return -ENOMEM; | |
8bb8aefd YL |
785 | new_op->upcall.req.fs_umount.id = ORANGEFS_SB(sb)->id; |
786 | new_op->upcall.req.fs_umount.fs_id = ORANGEFS_SB(sb)->fs_id; | |
787 | strncpy(new_op->upcall.req.fs_umount.orangefs_config_server, | |
788 | ORANGEFS_SB(sb)->devname, | |
789 | ORANGEFS_MAX_SERVER_ADDR_LEN); | |
f7be4ee0 MM |
790 | |
791 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
8bb8aefd YL |
792 | "Attempting ORANGEFS Unmount via host %s\n", |
793 | new_op->upcall.req.fs_umount.orangefs_config_server); | |
f7be4ee0 | 794 | |
8bb8aefd | 795 | ret = service_operation(new_op, "orangefs_fs_umount", 0); |
f7be4ee0 MM |
796 | |
797 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
8bb8aefd | 798 | "orangefs_unmount: got return value of %d\n", ret); |
f7be4ee0 MM |
799 | if (ret) |
800 | sb = ERR_PTR(ret); | |
801 | else | |
8bb8aefd | 802 | ORANGEFS_SB(sb)->mount_pending = 1; |
f7be4ee0 MM |
803 | |
804 | op_release(new_op); | |
805 | return ret; | |
806 | } | |
807 | ||
8bb8aefd | 808 | void orangefs_make_bad_inode(struct inode *inode) |
f7be4ee0 MM |
809 | { |
810 | if (is_root_handle(inode)) { | |
811 | /* | |
812 | * if this occurs, the pvfs2-client-core was killed but we | |
813 | * can't afford to lose the inode operations and such | |
814 | * associated with the root handle in any case. | |
815 | */ | |
816 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
817 | "*** NOT making bad root inode %pU\n", | |
818 | get_khandle_from_ino(inode)); | |
819 | } else { | |
820 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
821 | "*** making bad inode %pU\n", | |
822 | get_khandle_from_ino(inode)); | |
823 | make_bad_inode(inode); | |
824 | } | |
825 | } | |
826 | ||
54804949 MM |
827 | /* |
828 | * The following is a very dirty hack that is now a permanent part of the | |
8bb8aefd | 829 | * ORANGEFS protocol. See protocol.h for more error definitions. |
54804949 | 830 | */ |
894ac432 | 831 | |
8bb8aefd | 832 | /* The order matches include/orangefs-types.h in the OrangeFS source. */ |
894ac432 | 833 | static int PINT_errno_mapping[] = { |
54804949 MM |
834 | 0, EPERM, ENOENT, EINTR, EIO, ENXIO, EBADF, EAGAIN, ENOMEM, |
835 | EFAULT, EBUSY, EEXIST, ENODEV, ENOTDIR, EISDIR, EINVAL, EMFILE, | |
836 | EFBIG, ENOSPC, EROFS, EMLINK, EPIPE, EDEADLK, ENAMETOOLONG, | |
837 | ENOLCK, ENOSYS, ENOTEMPTY, ELOOP, EWOULDBLOCK, ENOMSG, EUNATCH, | |
838 | EBADR, EDEADLOCK, ENODATA, ETIME, ENONET, EREMOTE, ECOMM, | |
839 | EPROTO, EBADMSG, EOVERFLOW, ERESTART, EMSGSIZE, EPROTOTYPE, | |
840 | ENOPROTOOPT, EPROTONOSUPPORT, EOPNOTSUPP, EADDRINUSE, | |
841 | EADDRNOTAVAIL, ENETDOWN, ENETUNREACH, ENETRESET, ENOBUFS, | |
842 | ETIMEDOUT, ECONNREFUSED, EHOSTDOWN, EHOSTUNREACH, EALREADY, | |
843 | EACCES, ECONNRESET, ERANGE | |
894ac432 | 844 | }; |
f7be4ee0 | 845 | |
8bb8aefd | 846 | int orangefs_normalize_to_errno(__s32 error_code) |
f7be4ee0 | 847 | { |
54804949 MM |
848 | __u32 i; |
849 | ||
894ac432 MB |
850 | /* Success */ |
851 | if (error_code == 0) { | |
852 | return 0; | |
54804949 MM |
853 | /* |
854 | * This shouldn't ever happen. If it does it should be fixed on the | |
855 | * server. | |
856 | */ | |
894ac432 | 857 | } else if (error_code > 0) { |
8bb8aefd YL |
858 | gossip_err("orangefs: error status receieved.\n"); |
859 | gossip_err("orangefs: assuming error code is inverted.\n"); | |
f7be4ee0 MM |
860 | error_code = -error_code; |
861 | } | |
862 | ||
54804949 | 863 | /* |
8bb8aefd | 864 | * XXX: This is very bad since error codes from ORANGEFS may not be |
54804949 MM |
865 | * suitable for return into userspace. |
866 | */ | |
894ac432 | 867 | |
54804949 | 868 | /* |
8bb8aefd | 869 | * Convert ORANGEFS error values into errno values suitable for return |
54804949 MM |
870 | * from the kernel. |
871 | */ | |
8bb8aefd | 872 | if ((-error_code) & ORANGEFS_NON_ERRNO_ERROR_BIT) { |
894ac432 | 873 | if (((-error_code) & |
8bb8aefd YL |
874 | (ORANGEFS_ERROR_NUMBER_BITS|ORANGEFS_NON_ERRNO_ERROR_BIT| |
875 | ORANGEFS_ERROR_BIT)) == ORANGEFS_ECANCEL) { | |
f7be4ee0 MM |
876 | /* |
877 | * cancellation error codes generally correspond to | |
878 | * a timeout from the client's perspective | |
879 | */ | |
880 | error_code = -ETIMEDOUT; | |
881 | } else { | |
882 | /* assume a default error code */ | |
8bb8aefd | 883 | gossip_err("orangefs: warning: got error code without errno equivalent: %d.\n", error_code); |
f7be4ee0 MM |
884 | error_code = -EINVAL; |
885 | } | |
894ac432 | 886 | |
8bb8aefd YL |
887 | /* Convert ORANGEFS encoded errno values into regular errno values. */ |
888 | } else if ((-error_code) & ORANGEFS_ERROR_BIT) { | |
889 | i = (-error_code) & ~(ORANGEFS_ERROR_BIT|ORANGEFS_ERROR_CLASS_BITS); | |
54804949 | 890 | if (i < sizeof(PINT_errno_mapping)/sizeof(*PINT_errno_mapping)) |
894ac432 MB |
891 | error_code = -PINT_errno_mapping[i]; |
892 | else | |
893 | error_code = -EINVAL; | |
894 | ||
54804949 | 895 | /* |
8bb8aefd | 896 | * Only ORANGEFS protocol error codes should ever come here. Otherwise |
54804949 MM |
897 | * there is a bug somewhere. |
898 | */ | |
894ac432 | 899 | } else { |
8bb8aefd | 900 | gossip_err("orangefs: orangefs_normalize_to_errno: got error code which is not from ORANGEFS.\n"); |
f7be4ee0 MM |
901 | } |
902 | return error_code; | |
903 | } | |
904 | ||
905 | #define NUM_MODES 11 | |
8bb8aefd | 906 | __s32 ORANGEFS_util_translate_mode(int mode) |
f7be4ee0 MM |
907 | { |
908 | int ret = 0; | |
909 | int i = 0; | |
910 | static int modes[NUM_MODES] = { | |
911 | S_IXOTH, S_IWOTH, S_IROTH, | |
912 | S_IXGRP, S_IWGRP, S_IRGRP, | |
913 | S_IXUSR, S_IWUSR, S_IRUSR, | |
914 | S_ISGID, S_ISUID | |
915 | }; | |
8bb8aefd YL |
916 | static int orangefs_modes[NUM_MODES] = { |
917 | ORANGEFS_O_EXECUTE, ORANGEFS_O_WRITE, ORANGEFS_O_READ, | |
918 | ORANGEFS_G_EXECUTE, ORANGEFS_G_WRITE, ORANGEFS_G_READ, | |
919 | ORANGEFS_U_EXECUTE, ORANGEFS_U_WRITE, ORANGEFS_U_READ, | |
920 | ORANGEFS_G_SGID, ORANGEFS_U_SUID | |
f7be4ee0 MM |
921 | }; |
922 | ||
923 | for (i = 0; i < NUM_MODES; i++) | |
924 | if (mode & modes[i]) | |
8bb8aefd | 925 | ret |= orangefs_modes[i]; |
f7be4ee0 MM |
926 | |
927 | return ret; | |
928 | } | |
929 | #undef NUM_MODES | |
930 | ||
931 | /* | |
932 | * After obtaining a string representation of the client's debug | |
933 | * keywords and their associated masks, this function is called to build an | |
934 | * array of these values. | |
935 | */ | |
936 | int orangefs_prepare_cdm_array(char *debug_array_string) | |
937 | { | |
938 | int i; | |
939 | int rc = -EINVAL; | |
940 | char *cds_head = NULL; | |
941 | char *cds_delimiter = NULL; | |
942 | int keyword_len = 0; | |
943 | ||
944 | gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__); | |
945 | ||
946 | /* | |
947 | * figure out how many elements the cdm_array needs. | |
948 | */ | |
949 | for (i = 0; i < strlen(debug_array_string); i++) | |
950 | if (debug_array_string[i] == '\n') | |
951 | cdm_element_count++; | |
952 | ||
953 | if (!cdm_element_count) { | |
954 | pr_info("No elements in client debug array string!\n"); | |
955 | goto out; | |
956 | } | |
957 | ||
958 | cdm_array = | |
959 | kzalloc(cdm_element_count * sizeof(struct client_debug_mask), | |
960 | GFP_KERNEL); | |
961 | if (!cdm_array) { | |
962 | pr_info("malloc failed for cdm_array!\n"); | |
963 | rc = -ENOMEM; | |
964 | goto out; | |
965 | } | |
966 | ||
967 | cds_head = debug_array_string; | |
968 | ||
969 | for (i = 0; i < cdm_element_count; i++) { | |
970 | cds_delimiter = strchr(cds_head, '\n'); | |
971 | *cds_delimiter = '\0'; | |
972 | ||
973 | keyword_len = strcspn(cds_head, " "); | |
974 | ||
975 | cdm_array[i].keyword = kzalloc(keyword_len + 1, GFP_KERNEL); | |
976 | if (!cdm_array[i].keyword) { | |
977 | rc = -ENOMEM; | |
978 | goto out; | |
979 | } | |
980 | ||
981 | sscanf(cds_head, | |
982 | "%s %llx %llx", | |
983 | cdm_array[i].keyword, | |
984 | (unsigned long long *)&(cdm_array[i].mask1), | |
985 | (unsigned long long *)&(cdm_array[i].mask2)); | |
986 | ||
8bb8aefd | 987 | if (!strcmp(cdm_array[i].keyword, ORANGEFS_VERBOSE)) |
f7be4ee0 MM |
988 | client_verbose_index = i; |
989 | ||
8bb8aefd | 990 | if (!strcmp(cdm_array[i].keyword, ORANGEFS_ALL)) |
f7be4ee0 MM |
991 | client_all_index = i; |
992 | ||
993 | cds_head = cds_delimiter + 1; | |
994 | } | |
995 | ||
996 | rc = cdm_element_count; | |
997 | ||
998 | gossip_debug(GOSSIP_UTILS_DEBUG, "%s: rc:%d:\n", __func__, rc); | |
999 | ||
1000 | out: | |
1001 | ||
1002 | return rc; | |
1003 | ||
1004 | } | |
1005 | ||
1006 | /* | |
1007 | * /sys/kernel/debug/orangefs/debug-help can be catted to | |
1008 | * see all the available kernel and client debug keywords. | |
1009 | * | |
1010 | * When the kernel boots, we have no idea what keywords the | |
1011 | * client supports, nor their associated masks. | |
1012 | * | |
1013 | * We pass through this function once at boot and stamp a | |
1014 | * boilerplate "we don't know" message for the client in the | |
1015 | * debug-help file. We pass through here again when the client | |
1016 | * starts and then we can fill out the debug-help file fully. | |
1017 | * | |
1018 | * The client might be restarted any number of times between | |
1019 | * reboots, we only build the debug-help file the first time. | |
1020 | */ | |
1021 | int orangefs_prepare_debugfs_help_string(int at_boot) | |
1022 | { | |
1023 | int rc = -EINVAL; | |
1024 | int i; | |
1025 | int byte_count = 0; | |
1026 | char *client_title = "Client Debug Keywords:\n"; | |
1027 | char *kernel_title = "Kernel Debug Keywords:\n"; | |
1028 | ||
1029 | gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__); | |
1030 | ||
1031 | if (at_boot) { | |
1032 | byte_count += strlen(HELP_STRING_UNINITIALIZED); | |
1033 | client_title = HELP_STRING_UNINITIALIZED; | |
1034 | } else { | |
1035 | /* | |
1036 | * fill the client keyword/mask array and remember | |
1037 | * how many elements there were. | |
1038 | */ | |
1039 | cdm_element_count = | |
1040 | orangefs_prepare_cdm_array(client_debug_array_string); | |
1041 | if (cdm_element_count <= 0) | |
1042 | goto out; | |
1043 | ||
1044 | /* Count the bytes destined for debug_help_string. */ | |
1045 | byte_count += strlen(client_title); | |
1046 | ||
1047 | for (i = 0; i < cdm_element_count; i++) { | |
1048 | byte_count += strlen(cdm_array[i].keyword + 2); | |
1049 | if (byte_count >= DEBUG_HELP_STRING_SIZE) { | |
1050 | pr_info("%s: overflow 1!\n", __func__); | |
1051 | goto out; | |
1052 | } | |
1053 | } | |
1054 | ||
1055 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
1056 | "%s: cdm_element_count:%d:\n", | |
1057 | __func__, | |
1058 | cdm_element_count); | |
1059 | } | |
1060 | ||
1061 | byte_count += strlen(kernel_title); | |
1062 | for (i = 0; i < num_kmod_keyword_mask_map; i++) { | |
1063 | byte_count += | |
1064 | strlen(s_kmod_keyword_mask_map[i].keyword + 2); | |
1065 | if (byte_count >= DEBUG_HELP_STRING_SIZE) { | |
1066 | pr_info("%s: overflow 2!\n", __func__); | |
1067 | goto out; | |
1068 | } | |
1069 | } | |
1070 | ||
1071 | /* build debug_help_string. */ | |
1072 | debug_help_string = kzalloc(DEBUG_HELP_STRING_SIZE, GFP_KERNEL); | |
1073 | if (!debug_help_string) { | |
1074 | rc = -ENOMEM; | |
1075 | goto out; | |
1076 | } | |
1077 | ||
1078 | strcat(debug_help_string, client_title); | |
1079 | ||
1080 | if (!at_boot) { | |
1081 | for (i = 0; i < cdm_element_count; i++) { | |
1082 | strcat(debug_help_string, "\t"); | |
1083 | strcat(debug_help_string, cdm_array[i].keyword); | |
1084 | strcat(debug_help_string, "\n"); | |
1085 | } | |
1086 | } | |
1087 | ||
1088 | strcat(debug_help_string, "\n"); | |
1089 | strcat(debug_help_string, kernel_title); | |
1090 | ||
1091 | for (i = 0; i < num_kmod_keyword_mask_map; i++) { | |
1092 | strcat(debug_help_string, "\t"); | |
1093 | strcat(debug_help_string, s_kmod_keyword_mask_map[i].keyword); | |
1094 | strcat(debug_help_string, "\n"); | |
1095 | } | |
1096 | ||
1097 | rc = 0; | |
1098 | ||
1099 | out: | |
1100 | ||
1101 | return rc; | |
1102 | ||
1103 | } | |
1104 | ||
1105 | /* | |
1106 | * kernel = type 0 | |
1107 | * client = type 1 | |
1108 | */ | |
1109 | void debug_mask_to_string(void *mask, int type) | |
1110 | { | |
1111 | int i; | |
1112 | int len = 0; | |
1113 | char *debug_string; | |
1114 | int element_count = 0; | |
1115 | ||
1116 | gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__); | |
1117 | ||
1118 | if (type) { | |
1119 | debug_string = client_debug_string; | |
1120 | element_count = cdm_element_count; | |
1121 | } else { | |
1122 | debug_string = kernel_debug_string; | |
1123 | element_count = num_kmod_keyword_mask_map; | |
1124 | } | |
1125 | ||
8bb8aefd | 1126 | memset(debug_string, 0, ORANGEFS_MAX_DEBUG_STRING_LEN); |
f7be4ee0 MM |
1127 | |
1128 | /* | |
1129 | * Some keywords, like "all" or "verbose", are amalgams of | |
1130 | * numerous other keywords. Make a special check for those | |
1131 | * before grinding through the whole mask only to find out | |
1132 | * later... | |
1133 | */ | |
1134 | if (check_amalgam_keyword(mask, type)) | |
1135 | goto out; | |
1136 | ||
1137 | /* Build the debug string. */ | |
1138 | for (i = 0; i < element_count; i++) | |
1139 | if (type) | |
1140 | do_c_string(mask, i); | |
1141 | else | |
1142 | do_k_string(mask, i); | |
1143 | ||
1144 | len = strlen(debug_string); | |
1145 | ||
1146 | if ((len) && (type)) | |
1147 | client_debug_string[len - 1] = '\0'; | |
1148 | else if (len) | |
1149 | kernel_debug_string[len - 1] = '\0'; | |
1150 | else if (type) | |
1151 | strcpy(client_debug_string, "none"); | |
1152 | else | |
1153 | strcpy(kernel_debug_string, "none"); | |
1154 | ||
1155 | out: | |
1156 | gossip_debug(GOSSIP_UTILS_DEBUG, "%s: string:%s:\n", __func__, debug_string); | |
1157 | ||
1158 | return; | |
1159 | ||
1160 | } | |
1161 | ||
1162 | void do_k_string(void *k_mask, int index) | |
1163 | { | |
1164 | __u64 *mask = (__u64 *) k_mask; | |
1165 | ||
1166 | if (keyword_is_amalgam((char *) s_kmod_keyword_mask_map[index].keyword)) | |
54804949 | 1167 | goto out; |
f7be4ee0 MM |
1168 | |
1169 | if (*mask & s_kmod_keyword_mask_map[index].mask_val) { | |
1170 | if ((strlen(kernel_debug_string) + | |
1171 | strlen(s_kmod_keyword_mask_map[index].keyword)) | |
8bb8aefd | 1172 | < ORANGEFS_MAX_DEBUG_STRING_LEN - 1) { |
f7be4ee0 MM |
1173 | strcat(kernel_debug_string, |
1174 | s_kmod_keyword_mask_map[index].keyword); | |
1175 | strcat(kernel_debug_string, ","); | |
1176 | } else { | |
1177 | gossip_err("%s: overflow!\n", __func__); | |
8bb8aefd | 1178 | strcpy(kernel_debug_string, ORANGEFS_ALL); |
f7be4ee0 MM |
1179 | goto out; |
1180 | } | |
1181 | } | |
1182 | ||
1183 | out: | |
1184 | ||
1185 | return; | |
1186 | } | |
1187 | ||
1188 | void do_c_string(void *c_mask, int index) | |
1189 | { | |
1190 | struct client_debug_mask *mask = (struct client_debug_mask *) c_mask; | |
1191 | ||
1192 | if (keyword_is_amalgam(cdm_array[index].keyword)) | |
1193 | goto out; | |
1194 | ||
1195 | if ((mask->mask1 & cdm_array[index].mask1) || | |
1196 | (mask->mask2 & cdm_array[index].mask2)) { | |
1197 | if ((strlen(client_debug_string) + | |
1198 | strlen(cdm_array[index].keyword) + 1) | |
8bb8aefd | 1199 | < ORANGEFS_MAX_DEBUG_STRING_LEN - 2) { |
f7be4ee0 MM |
1200 | strcat(client_debug_string, |
1201 | cdm_array[index].keyword); | |
1202 | strcat(client_debug_string, ","); | |
1203 | } else { | |
1204 | gossip_err("%s: overflow!\n", __func__); | |
8bb8aefd | 1205 | strcpy(client_debug_string, ORANGEFS_ALL); |
f7be4ee0 MM |
1206 | goto out; |
1207 | } | |
1208 | } | |
1209 | out: | |
1210 | return; | |
1211 | } | |
1212 | ||
1213 | int keyword_is_amalgam(char *keyword) | |
1214 | { | |
1215 | int rc = 0; | |
1216 | ||
8bb8aefd | 1217 | if ((!strcmp(keyword, ORANGEFS_ALL)) || (!strcmp(keyword, ORANGEFS_VERBOSE))) |
f7be4ee0 MM |
1218 | rc = 1; |
1219 | ||
1220 | return rc; | |
1221 | } | |
1222 | ||
1223 | /* | |
1224 | * kernel = type 0 | |
1225 | * client = type 1 | |
1226 | * | |
1227 | * return 1 if we found an amalgam. | |
1228 | */ | |
1229 | int check_amalgam_keyword(void *mask, int type) | |
1230 | { | |
1231 | __u64 *k_mask; | |
1232 | struct client_debug_mask *c_mask; | |
1233 | int k_all_index = num_kmod_keyword_mask_map - 1; | |
1234 | int rc = 0; | |
1235 | ||
1236 | if (type) { | |
1237 | c_mask = (struct client_debug_mask *) mask; | |
1238 | ||
1239 | if ((c_mask->mask1 == cdm_array[client_all_index].mask1) && | |
1240 | (c_mask->mask2 == cdm_array[client_all_index].mask2)) { | |
8bb8aefd | 1241 | strcpy(client_debug_string, ORANGEFS_ALL); |
f7be4ee0 MM |
1242 | rc = 1; |
1243 | goto out; | |
1244 | } | |
1245 | ||
1246 | if ((c_mask->mask1 == cdm_array[client_verbose_index].mask1) && | |
1247 | (c_mask->mask2 == cdm_array[client_verbose_index].mask2)) { | |
8bb8aefd | 1248 | strcpy(client_debug_string, ORANGEFS_VERBOSE); |
f7be4ee0 MM |
1249 | rc = 1; |
1250 | goto out; | |
1251 | } | |
1252 | ||
1253 | } else { | |
1254 | k_mask = (__u64 *) mask; | |
1255 | ||
1256 | if (*k_mask >= s_kmod_keyword_mask_map[k_all_index].mask_val) { | |
8bb8aefd | 1257 | strcpy(kernel_debug_string, ORANGEFS_ALL); |
f7be4ee0 MM |
1258 | rc = 1; |
1259 | goto out; | |
1260 | } | |
1261 | } | |
1262 | ||
1263 | out: | |
1264 | ||
1265 | return rc; | |
1266 | } | |
1267 | ||
1268 | /* | |
1269 | * kernel = type 0 | |
1270 | * client = type 1 | |
1271 | */ | |
1272 | void debug_string_to_mask(char *debug_string, void *mask, int type) | |
1273 | { | |
1274 | char *unchecked_keyword; | |
1275 | int i; | |
1276 | char *strsep_fodder = kstrdup(debug_string, GFP_KERNEL); | |
eeaa3d44 | 1277 | char *original_pointer; |
f7be4ee0 MM |
1278 | int element_count = 0; |
1279 | struct client_debug_mask *c_mask; | |
1280 | __u64 *k_mask; | |
1281 | ||
1282 | gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__); | |
1283 | ||
1284 | if (type) { | |
1285 | c_mask = (struct client_debug_mask *)mask; | |
1286 | element_count = cdm_element_count; | |
1287 | } else { | |
1288 | k_mask = (__u64 *)mask; | |
1289 | *k_mask = 0; | |
1290 | element_count = num_kmod_keyword_mask_map; | |
1291 | } | |
1292 | ||
eeaa3d44 | 1293 | original_pointer = strsep_fodder; |
f7be4ee0 MM |
1294 | while ((unchecked_keyword = strsep(&strsep_fodder, ","))) |
1295 | if (strlen(unchecked_keyword)) { | |
1296 | for (i = 0; i < element_count; i++) | |
1297 | if (type) | |
1298 | do_c_mask(i, | |
1299 | unchecked_keyword, | |
1300 | &c_mask); | |
1301 | else | |
1302 | do_k_mask(i, | |
1303 | unchecked_keyword, | |
1304 | &k_mask); | |
1305 | } | |
1306 | ||
eeaa3d44 | 1307 | kfree(original_pointer); |
f7be4ee0 MM |
1308 | } |
1309 | ||
1310 | void do_c_mask(int i, | |
1311 | char *unchecked_keyword, | |
1312 | struct client_debug_mask **sane_mask) | |
1313 | { | |
1314 | ||
1315 | if (!strcmp(cdm_array[i].keyword, unchecked_keyword)) { | |
1316 | (**sane_mask).mask1 = (**sane_mask).mask1 | cdm_array[i].mask1; | |
1317 | (**sane_mask).mask2 = (**sane_mask).mask2 | cdm_array[i].mask2; | |
1318 | } | |
1319 | } | |
1320 | ||
1321 | void do_k_mask(int i, char *unchecked_keyword, __u64 **sane_mask) | |
1322 | { | |
1323 | ||
1324 | if (!strcmp(s_kmod_keyword_mask_map[i].keyword, unchecked_keyword)) | |
1325 | **sane_mask = (**sane_mask) | | |
1326 | s_kmod_keyword_mask_map[i].mask_val; | |
1327 | } |