2 * (C) 2001 Clemson University and The University of Chicago
4 * See COPYING in top-level directory.
7 #include "orangefs-kernel.h"
8 #include "orangefs-dev-proto.h"
9 #include "orangefs-bufmap.h"
11 __s32 fsid_of_op(struct orangefs_kernel_op_s *op)
13 __s32 fsid = ORANGEFS_FS_ID_NULL;
16 switch (op->upcall.type) {
17 case ORANGEFS_VFS_OP_FILE_IO:
18 fsid = op->upcall.req.io.refn.fs_id;
20 case ORANGEFS_VFS_OP_LOOKUP:
21 fsid = op->upcall.req.lookup.parent_refn.fs_id;
23 case ORANGEFS_VFS_OP_CREATE:
24 fsid = op->upcall.req.create.parent_refn.fs_id;
26 case ORANGEFS_VFS_OP_GETATTR:
27 fsid = op->upcall.req.getattr.refn.fs_id;
29 case ORANGEFS_VFS_OP_REMOVE:
30 fsid = op->upcall.req.remove.parent_refn.fs_id;
32 case ORANGEFS_VFS_OP_MKDIR:
33 fsid = op->upcall.req.mkdir.parent_refn.fs_id;
35 case ORANGEFS_VFS_OP_READDIR:
36 fsid = op->upcall.req.readdir.refn.fs_id;
38 case ORANGEFS_VFS_OP_SETATTR:
39 fsid = op->upcall.req.setattr.refn.fs_id;
41 case ORANGEFS_VFS_OP_SYMLINK:
42 fsid = op->upcall.req.sym.parent_refn.fs_id;
44 case ORANGEFS_VFS_OP_RENAME:
45 fsid = op->upcall.req.rename.old_parent_refn.fs_id;
47 case ORANGEFS_VFS_OP_STATFS:
48 fsid = op->upcall.req.statfs.fs_id;
50 case ORANGEFS_VFS_OP_TRUNCATE:
51 fsid = op->upcall.req.truncate.refn.fs_id;
53 case ORANGEFS_VFS_OP_MMAP_RA_FLUSH:
54 fsid = op->upcall.req.ra_cache_flush.refn.fs_id;
56 case ORANGEFS_VFS_OP_FS_UMOUNT:
57 fsid = op->upcall.req.fs_umount.fs_id;
59 case ORANGEFS_VFS_OP_GETXATTR:
60 fsid = op->upcall.req.getxattr.refn.fs_id;
62 case ORANGEFS_VFS_OP_SETXATTR:
63 fsid = op->upcall.req.setxattr.refn.fs_id;
65 case ORANGEFS_VFS_OP_LISTXATTR:
66 fsid = op->upcall.req.listxattr.refn.fs_id;
68 case ORANGEFS_VFS_OP_REMOVEXATTR:
69 fsid = op->upcall.req.removexattr.refn.fs_id;
71 case ORANGEFS_VFS_OP_FSYNC:
72 fsid = op->upcall.req.fsync.refn.fs_id;
81 static int orangefs_inode_flags(struct ORANGEFS_sys_attr_s *attrs)
84 if (attrs->flags & ORANGEFS_IMMUTABLE_FL)
87 flags &= ~S_IMMUTABLE;
88 if (attrs->flags & ORANGEFS_APPEND_FL)
92 if (attrs->flags & ORANGEFS_NOATIME_FL)
99 static int orangefs_inode_perms(struct ORANGEFS_sys_attr_s *attrs)
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;
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;
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;
124 if (attrs->perms & ORANGEFS_G_SGID)
125 perm_mode |= S_ISGID;
126 if (attrs->perms & ORANGEFS_U_SUID)
127 perm_mode |= S_ISUID;
132 /* NOTE: symname is ignored unless the inode is a sym link */
133 static int copy_attributes_to_inode(struct inode *inode,
134 struct ORANGEFS_sys_attr_s *attrs,
138 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
139 loff_t inode_size = 0;
140 loff_t rounded_up_size = 0;
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
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.
154 * changing the inode->i_blkbits to something other than
155 * PAGE_CACHE_SHIFT breaks mmap/execution as we depend on that.
157 gossip_debug(GOSSIP_UTILS_DEBUG,
158 "attrs->mask = %x (objtype = %s)\n",
160 attrs->objtype == ORANGEFS_TYPE_METAFILE ? "file" :
161 attrs->objtype == ORANGEFS_TYPE_DIRECTORY ? "directory" :
162 attrs->objtype == ORANGEFS_TYPE_SYMLINK ? "symlink" :
165 switch (attrs->objtype) {
166 case ORANGEFS_TYPE_METAFILE:
167 inode->i_flags = orangefs_inode_flags(attrs);
168 if (attrs->mask & ORANGEFS_ATTR_SYS_SIZE) {
169 inode_size = (loff_t) attrs->size;
171 (inode_size + (4096 - (inode_size % 4096)));
173 spin_lock(&inode->i_lock);
174 inode->i_bytes = inode_size;
176 (unsigned long)(rounded_up_size / 512);
177 spin_unlock(&inode->i_lock);
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.
185 inode->i_size = inode_size;
188 case ORANGEFS_TYPE_SYMLINK:
189 if (symname != NULL) {
190 inode->i_size = (loff_t) strlen(symname);
195 inode->i_size = PAGE_CACHE_SIZE;
197 spin_lock(&inode->i_lock);
198 inode_set_bytes(inode, inode->i_size);
199 spin_unlock(&inode->i_lock);
203 inode->i_uid = make_kuid(&init_user_ns, attrs->owner);
204 inode->i_gid = make_kgid(&init_user_ns, attrs->group);
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;
208 inode->i_atime.tv_nsec = 0;
209 inode->i_mtime.tv_nsec = 0;
210 inode->i_ctime.tv_nsec = 0;
212 inode->i_mode = orangefs_inode_perms(attrs);
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));
222 switch (attrs->objtype) {
223 case ORANGEFS_TYPE_METAFILE:
224 inode->i_mode |= S_IFREG;
227 case ORANGEFS_TYPE_DIRECTORY:
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
237 case ORANGEFS_TYPE_SYMLINK:
238 inode->i_mode |= S_IFLNK;
240 /* copy link target to inode private data */
241 if (orangefs_inode && symname) {
242 strncpy(orangefs_inode->link_target,
245 gossip_debug(GOSSIP_UTILS_DEBUG,
246 "Copied attr link target %s\n",
247 orangefs_inode->link_target);
249 gossip_debug(GOSSIP_UTILS_DEBUG,
255 gossip_err("orangefs: copy_attributes_to_inode: got invalid attribute type %x\n",
259 gossip_debug(GOSSIP_UTILS_DEBUG,
260 "orangefs: copy_attributes_to_inode: setting i_mode to %o, i_size to %lu\n",
262 (unsigned long)i_size_read(inode));
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.
271 static inline int copy_attributes_from_inode(struct inode *inode,
272 struct ORANGEFS_sys_attr_s *attrs,
277 if (!iattr || !inode || !attrs) {
278 gossip_err("NULL iattr (%p), inode (%p), attrs (%p) "
279 "in copy_attributes_from_inode!\n",
286 * We need to be careful to only copy the attributes out of the
287 * iattr object that we know are valid.
290 if (iattr->ia_valid & ATTR_UID) {
291 attrs->owner = from_kuid(current_user_ns(), iattr->ia_uid);
292 attrs->mask |= ORANGEFS_ATTR_SYS_UID;
293 gossip_debug(GOSSIP_UTILS_DEBUG, "(UID) %d\n", attrs->owner);
295 if (iattr->ia_valid & ATTR_GID) {
296 attrs->group = from_kgid(current_user_ns(), iattr->ia_gid);
297 attrs->mask |= ORANGEFS_ATTR_SYS_GID;
298 gossip_debug(GOSSIP_UTILS_DEBUG, "(GID) %d\n", attrs->group);
301 if (iattr->ia_valid & ATTR_ATIME) {
302 attrs->mask |= ORANGEFS_ATTR_SYS_ATIME;
303 if (iattr->ia_valid & ATTR_ATIME_SET) {
304 attrs->atime = (time64_t)iattr->ia_atime.tv_sec;
305 attrs->mask |= ORANGEFS_ATTR_SYS_ATIME_SET;
308 if (iattr->ia_valid & ATTR_MTIME) {
309 attrs->mask |= ORANGEFS_ATTR_SYS_MTIME;
310 if (iattr->ia_valid & ATTR_MTIME_SET) {
311 attrs->mtime = (time64_t)iattr->ia_mtime.tv_sec;
312 attrs->mask |= ORANGEFS_ATTR_SYS_MTIME_SET;
315 if (iattr->ia_valid & ATTR_CTIME)
316 attrs->mask |= ORANGEFS_ATTR_SYS_CTIME;
319 * ORANGEFS cannot set size with a setattr operation. Probably not likely
320 * to be requested through the VFS, but just in case, don't worry about
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)) {
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
335 gossip_debug(GOSSIP_UTILS_DEBUG,
336 "User attempted to set sticky bit on non-root directory; returning EINVAL.\n");
341 if (tmp_mode & (S_ISUID)) {
342 gossip_debug(GOSSIP_UTILS_DEBUG,
343 "Attempting to set setuid bit (not supported); returning EINVAL.\n");
347 attrs->perms = ORANGEFS_util_translate_mode(tmp_mode);
348 attrs->mask |= ORANGEFS_ATTR_SYS_PERM;
354 static int compare_attributes_to_inode(struct inode *inode,
355 struct ORANGEFS_sys_attr_s *attrs,
359 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
360 loff_t inode_size, rounded_up_size;
362 /* Much of what happens below relies on the type being around. */
363 if (!(mask & ORANGEFS_ATTR_SYS_TYPE))
366 if (attrs->objtype == ORANGEFS_TYPE_METAFILE &&
367 inode->i_flags != orangefs_inode_flags(attrs))
370 /* Compare file size. */
372 switch (attrs->objtype) {
373 case ORANGEFS_TYPE_METAFILE:
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)
383 case ORANGEFS_TYPE_SYMLINK:
384 if (mask & ORANGEFS_ATTR_SYS_SIZE)
385 if (symname && strlen(symname) != inode->i_size)
389 if (inode->i_size != PAGE_CACHE_SIZE &&
390 inode_get_bytes(inode) != PAGE_CACHE_SIZE)
394 /* Compare general attributes. */
396 if (mask & ORANGEFS_ATTR_SYS_UID &&
397 !uid_eq(inode->i_uid, make_kuid(&init_user_ns, attrs->owner)))
399 if (mask & ORANGEFS_ATTR_SYS_GID &&
400 !gid_eq(inode->i_gid, make_kgid(&init_user_ns, attrs->group)))
402 if (mask & ORANGEFS_ATTR_SYS_ATIME &&
403 inode->i_atime.tv_sec != attrs->atime)
405 if (mask & ORANGEFS_ATTR_SYS_MTIME &&
406 inode->i_atime.tv_sec != attrs->mtime)
408 if (mask & ORANGEFS_ATTR_SYS_CTIME &&
409 inode->i_atime.tv_sec != attrs->ctime)
411 if (inode->i_atime.tv_nsec != 0 ||
412 inode->i_mtime.tv_nsec != 0 ||
413 inode->i_ctime.tv_nsec != 0)
416 if (mask & ORANGEFS_ATTR_SYS_PERM &&
417 (inode->i_mode & ~(S_ISVTX|S_IFREG|S_IFDIR|S_IFLNK)) !=
418 orangefs_inode_perms(attrs))
421 if (is_root_handle(inode))
422 if (!(inode->i_mode & S_ISVTX))
425 /* Compare file type. */
427 switch (attrs->objtype) {
428 case ORANGEFS_TYPE_METAFILE:
429 if (!S_ISREG(inode->i_mode))
432 case ORANGEFS_TYPE_DIRECTORY:
433 if (!S_ISDIR(inode->i_mode))
435 if (inode->i_nlink != 1)
438 case ORANGEFS_TYPE_SYMLINK:
439 if (!S_ISLNK(inode->i_mode))
441 if (orangefs_inode && symname &&
442 mask & ORANGEFS_ATTR_SYS_LNK_TARGET)
443 if (strcmp(orangefs_inode->link_target, symname))
447 gossip_err("orangefs: compare_attributes_to_inode: got invalid attribute type %x\n",
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.
461 int orangefs_inode_old_getattr(struct inode *inode, __u32 getattr_mask,
464 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
465 struct orangefs_kernel_op_s *new_op;
468 gossip_debug(GOSSIP_UTILS_DEBUG,
469 "%s: called on inode %pU\n",
471 get_khandle_from_ino(inode));
473 new_op = op_alloc(ORANGEFS_VFS_OP_GETATTR);
476 new_op->upcall.req.getattr.refn = orangefs_inode->refn;
477 new_op->upcall.req.getattr.mask = getattr_mask;
479 ret = service_operation(new_op, __func__,
480 get_interruptible_flag(inode));
485 ret = compare_attributes_to_inode(inode,
486 &new_op->downcall.resp.getattr.attributes,
487 new_op->downcall.resp.getattr.link_target,
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)
496 if (orangefs_inode->blksize != 1 << inode->i_blkbits)
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__);
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.
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;
519 * mimic behavior of generic_fillattr() for other file
522 orangefs_inode->blksize = (1 << inode->i_blkbits);
528 gossip_debug(GOSSIP_UTILS_DEBUG,
529 "Getattr on handle %pU, "
530 "fsid %d\n (inode ct = %d) returned %d\n",
531 &orangefs_inode->refn.khandle,
532 orangefs_inode->refn.fs_id,
533 (int)atomic_read(&inode->i_count),
540 static int orangefs_inode_type(enum orangefs_ds_type objtype)
542 if (objtype == ORANGEFS_TYPE_METAFILE)
544 else if (objtype == ORANGEFS_TYPE_DIRECTORY)
546 else if (objtype == ORANGEFS_TYPE_SYMLINK)
552 int orangefs_inode_getattr(struct inode *inode, int new, int size)
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;
559 gossip_debug(GOSSIP_UTILS_DEBUG, "%s: called on inode %pU\n", __func__,
560 get_khandle_from_ino(inode));
562 new_op = op_alloc(ORANGEFS_VFS_OP_GETATTR);
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;
569 ret = service_operation(new_op, __func__,
570 get_interruptible_flag(inode));
574 ret = orangefs_inode_type(new_op->
575 downcall.resp.getattr.attributes.objtype);
578 * If the inode type or symlink target have changed then this
581 if (ret == -1 || !(inode->i_mode & ret)) {
582 orangefs_make_bad_inode(inode);
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);
597 inode->i_flags = orangefs_inode_flags(&new_op->
598 downcall.resp.getattr.attributes);
600 inode_size = (loff_t)new_op->
601 downcall.resp.getattr.attributes.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;
610 (unsigned long)(rounded_up_size / 512);
611 spin_unlock(&inode->i_lock);
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);
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,
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;
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);
659 * issues a orangefs setattr request to make sure the new attribute values
660 * take effect if successful. returns 0 on success; -errno otherwise
662 int orangefs_inode_setattr(struct inode *inode, struct iattr *iattr)
664 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
665 struct orangefs_kernel_op_s *new_op;
668 new_op = op_alloc(ORANGEFS_VFS_OP_SETATTR);
672 new_op->upcall.req.setattr.refn = orangefs_inode->refn;
673 ret = copy_attributes_from_inode(inode,
674 &new_op->upcall.req.setattr.attributes,
677 ret = service_operation(new_op, __func__,
678 get_interruptible_flag(inode));
680 gossip_debug(GOSSIP_UTILS_DEBUG,
681 "orangefs_inode_setattr: returning %d\n",
688 * successful setattr should clear the atime, mtime and
692 ClearAtimeFlag(orangefs_inode);
693 ClearMtimeFlag(orangefs_inode);
694 ClearCtimeFlag(orangefs_inode);
695 ClearModeFlag(orangefs_inode);
701 int orangefs_flush_inode(struct inode *inode)
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
715 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
717 memset(&wbattr, 0, sizeof(wbattr));
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
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);
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
737 * Also mode updates are being handled now..
741 wbattr.ia_valid |= ATTR_MTIME;
743 wbattr.ia_valid |= ATTR_CTIME;
745 wbattr.ia_valid |= ATTR_ATIME;
748 wbattr.ia_mode = inode->i_mode;
749 wbattr.ia_valid |= ATTR_MODE;
752 gossip_debug(GOSSIP_UTILS_DEBUG,
753 "*********** orangefs_flush_inode: %pU "
755 get_khandle_from_ino(inode),
757 if (wbattr.ia_valid == 0) {
758 gossip_debug(GOSSIP_UTILS_DEBUG,
759 "orangefs_flush_inode skipping setattr()\n");
763 gossip_debug(GOSSIP_UTILS_DEBUG,
764 "orangefs_flush_inode (%pU) writing mode %o\n",
765 get_khandle_from_ino(inode),
768 ret = orangefs_inode_setattr(inode, &wbattr);
773 int orangefs_unmount_sb(struct super_block *sb)
776 struct orangefs_kernel_op_s *new_op = NULL;
778 gossip_debug(GOSSIP_UTILS_DEBUG,
779 "orangefs_unmount_sb called on sb %p\n",
782 new_op = op_alloc(ORANGEFS_VFS_OP_FS_UMOUNT);
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);
791 gossip_debug(GOSSIP_UTILS_DEBUG,
792 "Attempting ORANGEFS Unmount via host %s\n",
793 new_op->upcall.req.fs_umount.orangefs_config_server);
795 ret = service_operation(new_op, "orangefs_fs_umount", 0);
797 gossip_debug(GOSSIP_UTILS_DEBUG,
798 "orangefs_unmount: got return value of %d\n", ret);
802 ORANGEFS_SB(sb)->mount_pending = 1;
808 void orangefs_make_bad_inode(struct inode *inode)
810 if (is_root_handle(inode)) {
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.
816 gossip_debug(GOSSIP_UTILS_DEBUG,
817 "*** NOT making bad root inode %pU\n",
818 get_khandle_from_ino(inode));
820 gossip_debug(GOSSIP_UTILS_DEBUG,
821 "*** making bad inode %pU\n",
822 get_khandle_from_ino(inode));
823 make_bad_inode(inode);
828 * The following is a very dirty hack that is now a permanent part of the
829 * ORANGEFS protocol. See protocol.h for more error definitions.
832 /* The order matches include/orangefs-types.h in the OrangeFS source. */
833 static int PINT_errno_mapping[] = {
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
846 int orangefs_normalize_to_errno(__s32 error_code)
851 if (error_code == 0) {
854 * This shouldn't ever happen. If it does it should be fixed on the
857 } else if (error_code > 0) {
858 gossip_err("orangefs: error status receieved.\n");
859 gossip_err("orangefs: assuming error code is inverted.\n");
860 error_code = -error_code;
864 * XXX: This is very bad since error codes from ORANGEFS may not be
865 * suitable for return into userspace.
869 * Convert ORANGEFS error values into errno values suitable for return
872 if ((-error_code) & ORANGEFS_NON_ERRNO_ERROR_BIT) {
874 (ORANGEFS_ERROR_NUMBER_BITS|ORANGEFS_NON_ERRNO_ERROR_BIT|
875 ORANGEFS_ERROR_BIT)) == ORANGEFS_ECANCEL) {
877 * cancellation error codes generally correspond to
878 * a timeout from the client's perspective
880 error_code = -ETIMEDOUT;
882 /* assume a default error code */
883 gossip_err("orangefs: warning: got error code without errno equivalent: %d.\n", error_code);
884 error_code = -EINVAL;
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);
890 if (i < sizeof(PINT_errno_mapping)/sizeof(*PINT_errno_mapping))
891 error_code = -PINT_errno_mapping[i];
893 error_code = -EINVAL;
896 * Only ORANGEFS protocol error codes should ever come here. Otherwise
897 * there is a bug somewhere.
900 gossip_err("orangefs: orangefs_normalize_to_errno: got error code which is not from ORANGEFS.\n");
906 __s32 ORANGEFS_util_translate_mode(int mode)
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,
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
923 for (i = 0; i < NUM_MODES; i++)
925 ret |= orangefs_modes[i];
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.
936 int orangefs_prepare_cdm_array(char *debug_array_string)
940 char *cds_head = NULL;
941 char *cds_delimiter = NULL;
944 gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__);
947 * figure out how many elements the cdm_array needs.
949 for (i = 0; i < strlen(debug_array_string); i++)
950 if (debug_array_string[i] == '\n')
953 if (!cdm_element_count) {
954 pr_info("No elements in client debug array string!\n");
959 kzalloc(cdm_element_count * sizeof(struct client_debug_mask),
962 pr_info("malloc failed for cdm_array!\n");
967 cds_head = debug_array_string;
969 for (i = 0; i < cdm_element_count; i++) {
970 cds_delimiter = strchr(cds_head, '\n');
971 *cds_delimiter = '\0';
973 keyword_len = strcspn(cds_head, " ");
975 cdm_array[i].keyword = kzalloc(keyword_len + 1, GFP_KERNEL);
976 if (!cdm_array[i].keyword) {
983 cdm_array[i].keyword,
984 (unsigned long long *)&(cdm_array[i].mask1),
985 (unsigned long long *)&(cdm_array[i].mask2));
987 if (!strcmp(cdm_array[i].keyword, ORANGEFS_VERBOSE))
988 client_verbose_index = i;
990 if (!strcmp(cdm_array[i].keyword, ORANGEFS_ALL))
991 client_all_index = i;
993 cds_head = cds_delimiter + 1;
996 rc = cdm_element_count;
998 gossip_debug(GOSSIP_UTILS_DEBUG, "%s: rc:%d:\n", __func__, rc);
1007 * /sys/kernel/debug/orangefs/debug-help can be catted to
1008 * see all the available kernel and client debug keywords.
1010 * When the kernel boots, we have no idea what keywords the
1011 * client supports, nor their associated masks.
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.
1018 * The client might be restarted any number of times between
1019 * reboots, we only build the debug-help file the first time.
1021 int orangefs_prepare_debugfs_help_string(int at_boot)
1026 char *client_title = "Client Debug Keywords:\n";
1027 char *kernel_title = "Kernel Debug Keywords:\n";
1029 gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__);
1032 byte_count += strlen(HELP_STRING_UNINITIALIZED);
1033 client_title = HELP_STRING_UNINITIALIZED;
1036 * fill the client keyword/mask array and remember
1037 * how many elements there were.
1040 orangefs_prepare_cdm_array(client_debug_array_string);
1041 if (cdm_element_count <= 0)
1044 /* Count the bytes destined for debug_help_string. */
1045 byte_count += strlen(client_title);
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__);
1055 gossip_debug(GOSSIP_UTILS_DEBUG,
1056 "%s: cdm_element_count:%d:\n",
1061 byte_count += strlen(kernel_title);
1062 for (i = 0; i < num_kmod_keyword_mask_map; i++) {
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__);
1071 /* build debug_help_string. */
1072 debug_help_string = kzalloc(DEBUG_HELP_STRING_SIZE, GFP_KERNEL);
1073 if (!debug_help_string) {
1078 strcat(debug_help_string, client_title);
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");
1088 strcat(debug_help_string, "\n");
1089 strcat(debug_help_string, kernel_title);
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");
1109 void debug_mask_to_string(void *mask, int type)
1114 int element_count = 0;
1116 gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__);
1119 debug_string = client_debug_string;
1120 element_count = cdm_element_count;
1122 debug_string = kernel_debug_string;
1123 element_count = num_kmod_keyword_mask_map;
1126 memset(debug_string, 0, ORANGEFS_MAX_DEBUG_STRING_LEN);
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
1134 if (check_amalgam_keyword(mask, type))
1137 /* Build the debug string. */
1138 for (i = 0; i < element_count; i++)
1140 do_c_string(mask, i);
1142 do_k_string(mask, i);
1144 len = strlen(debug_string);
1146 if ((len) && (type))
1147 client_debug_string[len - 1] = '\0';
1149 kernel_debug_string[len - 1] = '\0';
1151 strcpy(client_debug_string, "none");
1153 strcpy(kernel_debug_string, "none");
1156 gossip_debug(GOSSIP_UTILS_DEBUG, "%s: string:%s:\n", __func__, debug_string);
1162 void do_k_string(void *k_mask, int index)
1164 __u64 *mask = (__u64 *) k_mask;
1166 if (keyword_is_amalgam((char *) s_kmod_keyword_mask_map[index].keyword))
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))
1172 < ORANGEFS_MAX_DEBUG_STRING_LEN - 1) {
1173 strcat(kernel_debug_string,
1174 s_kmod_keyword_mask_map[index].keyword);
1175 strcat(kernel_debug_string, ",");
1177 gossip_err("%s: overflow!\n", __func__);
1178 strcpy(kernel_debug_string, ORANGEFS_ALL);
1188 void do_c_string(void *c_mask, int index)
1190 struct client_debug_mask *mask = (struct client_debug_mask *) c_mask;
1192 if (keyword_is_amalgam(cdm_array[index].keyword))
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)
1199 < ORANGEFS_MAX_DEBUG_STRING_LEN - 2) {
1200 strcat(client_debug_string,
1201 cdm_array[index].keyword);
1202 strcat(client_debug_string, ",");
1204 gossip_err("%s: overflow!\n", __func__);
1205 strcpy(client_debug_string, ORANGEFS_ALL);
1213 int keyword_is_amalgam(char *keyword)
1217 if ((!strcmp(keyword, ORANGEFS_ALL)) || (!strcmp(keyword, ORANGEFS_VERBOSE)))
1227 * return 1 if we found an amalgam.
1229 int check_amalgam_keyword(void *mask, int type)
1232 struct client_debug_mask *c_mask;
1233 int k_all_index = num_kmod_keyword_mask_map - 1;
1237 c_mask = (struct client_debug_mask *) mask;
1239 if ((c_mask->mask1 == cdm_array[client_all_index].mask1) &&
1240 (c_mask->mask2 == cdm_array[client_all_index].mask2)) {
1241 strcpy(client_debug_string, ORANGEFS_ALL);
1246 if ((c_mask->mask1 == cdm_array[client_verbose_index].mask1) &&
1247 (c_mask->mask2 == cdm_array[client_verbose_index].mask2)) {
1248 strcpy(client_debug_string, ORANGEFS_VERBOSE);
1254 k_mask = (__u64 *) mask;
1256 if (*k_mask >= s_kmod_keyword_mask_map[k_all_index].mask_val) {
1257 strcpy(kernel_debug_string, ORANGEFS_ALL);
1272 void debug_string_to_mask(char *debug_string, void *mask, int type)
1274 char *unchecked_keyword;
1276 char *strsep_fodder = kstrdup(debug_string, GFP_KERNEL);
1277 char *original_pointer;
1278 int element_count = 0;
1279 struct client_debug_mask *c_mask;
1282 gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__);
1285 c_mask = (struct client_debug_mask *)mask;
1286 element_count = cdm_element_count;
1288 k_mask = (__u64 *)mask;
1290 element_count = num_kmod_keyword_mask_map;
1293 original_pointer = strsep_fodder;
1294 while ((unchecked_keyword = strsep(&strsep_fodder, ",")))
1295 if (strlen(unchecked_keyword)) {
1296 for (i = 0; i < element_count; i++)
1307 kfree(original_pointer);
1310 void do_c_mask(int i,
1311 char *unchecked_keyword,
1312 struct client_debug_mask **sane_mask)
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;
1321 void do_k_mask(int i, char *unchecked_keyword, __u64 **sane_mask)
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;