4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Directory code for lustre client.
42 #include <linux/pagemap.h>
44 #include <linux/version.h>
45 #include <linux/smp_lock.h>
46 #include <asm/uaccess.h>
47 #include <linux/buffer_head.h> // for wait_on_buffer
48 #include <linux/pagevec.h>
50 #define DEBUG_SUBSYSTEM S_LLITE
52 #include <obd_support.h>
53 #include <obd_class.h>
54 #include <lustre_lib.h>
55 #include <lustre/lustre_idl.h>
56 #include <lustre_lite.h>
57 #include <lustre_dlm.h>
58 #include <lustre_fid.h>
59 #include "llite_internal.h"
61 #ifndef HAVE_PAGE_CHECKED
62 #ifdef HAVE_PG_FS_MISC
63 #define PageChecked(page) test_bit(PG_fs_misc, &(page)->flags)
64 #define SetPageChecked(page) set_bit(PG_fs_misc, &(page)->flags)
66 #error PageChecked or PageFsMisc not defined in kernel
71 * (new) readdir implementation overview.
73 * Original lustre readdir implementation cached exact copy of raw directory
74 * pages on the client. These pages were indexed in client page cache by
75 * logical offset in the directory file. This design, while very simple and
76 * intuitive had some inherent problems:
78 * . it implies that byte offset to the directory entry serves as a
79 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
80 * ext3/htree directory entries may move due to splits, and more
83 * . it is incompatible with the design of split directories for cmd3,
84 * that assumes that names are distributed across nodes based on their
85 * hash, and so readdir should be done in hash order.
87 * New readdir implementation does readdir in hash order, and uses hash of a
88 * file name as a telldir/seekdir cookie. This led to number of complications:
90 * . hash is not unique, so it cannot be used to index cached directory
91 * pages on the client (note, that it requires a whole pageful of hash
92 * collided entries to cause two pages to have identical hashes);
94 * . hash is not unique, so it cannot, strictly speaking, be used as an
95 * entry cookie. ext3/htree has the same problem and lustre implementation
96 * mimics their solution: seekdir(hash) positions directory at the first
97 * entry with the given hash.
103 * Client caches directory pages using hash of the first entry as an index. As
104 * noted above hash is not unique, so this solution doesn't work as is:
105 * special processing is needed for "page hash chains" (i.e., sequences of
106 * pages filled with entries all having the same hash value).
108 * First, such chains have to be detected. To this end, server returns to the
109 * client the hash of the first entry on the page next to one returned. When
110 * client detects that this hash is the same as hash of the first entry on the
111 * returned page, page hash collision has to be handled. Pages in the
112 * hash chain, except first one, are termed "overflow pages".
114 * Solution to index uniqueness problem is to not cache overflow
115 * pages. Instead, when page hash collision is detected, all overflow pages
116 * from emerging chain are immediately requested from the server and placed in
117 * a special data structure (struct ll_dir_chain). This data structure is used
118 * by ll_readdir() to process entries from overflow pages. When readdir
119 * invocation finishes, overflow pages are discarded. If page hash collision
120 * chain weren't completely processed, next call to readdir will again detect
121 * page hash collision, again read overflow pages in, process next portion of
122 * entries and again discard the pages. This is not as wasteful as it looks,
123 * because, given reasonable hash, page hash collisions are extremely rare.
125 * 1. directory positioning
127 * When seekdir(hash) is called, original
138 * identification of and access to overflow pages
142 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
143 * a header lu_dirpage which describes the start/end hash, and whether this
144 * page is empty (contains no dir entry) or hash collide with next page.
145 * After client receives reply, several pages will be integrated into dir page
146 * in CFS_PAGE_SIZE (if CFS_PAGE_SIZE greater than LU_PAGE_SIZE), and the
147 * lu_dirpage for this integrated page will be adjusted.
151 /* returns the page unlocked, but with a reference */
152 static int ll_dir_readpage(struct file *file, struct page *page0)
154 struct inode *inode = page0->mapping->host;
155 int hash64 = ll_i2sbi(inode)->ll_flags & LL_SBI_64BIT_HASH;
156 struct obd_export *exp = ll_i2sbi(inode)->ll_md_exp;
157 struct ptlrpc_request *request;
158 struct mdt_body *body;
159 struct md_op_data *op_data;
161 struct page **page_pool;
163 #ifndef HAVE_ADD_TO_PAGE_CACHE_LRU
164 struct pagevec lru_pvec;
166 struct lu_dirpage *dp;
167 int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> CFS_PAGE_SHIFT;
168 int nrdpgs = 0; /* number of pages read actually */
175 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
177 hash = fd->fd_dir.lfd_next;
179 hash = ll_i2info(inode)->lli_sa_pos;
181 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash "LPU64"\n",
182 inode->i_ino, inode->i_generation, inode, hash);
184 LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
186 OBD_ALLOC(page_pool, sizeof(page) * max_pages);
187 if (page_pool != NULL) {
188 page_pool[0] = page0;
193 for (npages = 1; npages < max_pages; npages++) {
194 page = page_cache_alloc_cold(inode->i_mapping);
197 page_pool[npages] = page;
200 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
201 LUSTRE_OPC_ANY, NULL);
202 op_data->op_npages = npages;
203 op_data->op_offset = hash;
204 rc = md_readpage(exp, op_data, page_pool, &request);
205 ll_finish_md_op_data(op_data);
207 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
208 /* Checked by mdc_readpage() */
209 LASSERT(body != NULL);
211 if (body->valid & OBD_MD_FLSIZE)
212 cl_isize_write(inode, body->size);
214 nrdpgs = (request->rq_bulk->bd_nob_transferred+CFS_PAGE_SIZE-1)
216 SetPageUptodate(page0);
219 ptlrpc_req_finished(request);
221 CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
223 ll_pagevec_init(&lru_pvec, 0);
224 for (i = 1; i < npages; i++) {
225 unsigned long offset;
230 if (rc < 0 || i >= nrdpgs) {
231 page_cache_release(page);
235 SetPageUptodate(page);
238 hash = le64_to_cpu(dp->ldp_hash_start);
241 offset = hash_x_index(hash, hash64);
243 prefetchw(&page->flags);
244 ret = ll_add_to_page_cache_lru(page, inode->i_mapping, offset,
248 if (ll_pagevec_add(&lru_pvec, page) == 0)
249 ll_pagevec_lru_add_file(&lru_pvec);
251 CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
252 " %d\n", offset, ret);
254 page_cache_release(page);
256 ll_pagevec_lru_add_file(&lru_pvec);
258 if (page_pool != &page0)
259 OBD_FREE(page_pool, sizeof(struct page *) * max_pages);
264 #ifndef MS_HAS_NEW_AOPS
265 struct address_space_operations ll_dir_aops = {
266 .readpage = ll_dir_readpage,
269 struct address_space_operations_ext ll_dir_aops = {
270 .orig_aops.readpage = ll_dir_readpage,
274 static void ll_check_page(struct inode *dir, struct page *page)
276 /* XXX: check page format later */
277 SetPageChecked(page);
280 void ll_release_page(struct page *page, int remove)
285 if (likely(page->mapping != NULL))
286 truncate_complete_page(page->mapping, page);
289 page_cache_release(page);
293 * Find, kmap and return page that contains given hash.
295 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
296 __u64 *start, __u64 *end)
298 int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
299 struct address_space *mapping = dir->i_mapping;
301 * Complement of hash is used as an index so that
302 * radix_tree_gang_lookup() can be used to find a page with starting
303 * hash _smaller_ than one we are looking for.
305 unsigned long offset = hash_x_index(*hash, hash64);
309 TREE_READ_LOCK_IRQ(mapping);
310 found = radix_tree_gang_lookup(&mapping->page_tree,
311 (void **)&page, offset, 1);
313 struct lu_dirpage *dp;
315 page_cache_get(page);
316 TREE_READ_UNLOCK_IRQ(mapping);
318 * In contrast to find_lock_page() we are sure that directory
319 * page cannot be truncated (while DLM lock is held) and,
320 * hence, can avoid restart.
322 * In fact, page cannot be locked here at all, because
323 * ll_dir_readpage() does synchronous io.
326 if (PageUptodate(page)) {
328 if (BITS_PER_LONG == 32 && hash64) {
329 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
330 *end = le64_to_cpu(dp->ldp_hash_end) >> 32;
333 *start = le64_to_cpu(dp->ldp_hash_start);
334 *end = le64_to_cpu(dp->ldp_hash_end);
336 LASSERTF(*start <= *hash, "start = "LPX64",end = "
337 LPX64",hash = "LPX64"\n", *start, *end, *hash);
338 CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash "LPU64"\n",
339 offset, *start, *end, *hash);
341 ll_release_page(page, 0);
343 } else if (*end != *start && *hash == *end) {
345 * upon hash collision, remove this page,
346 * otherwise put page reference, and
347 * ll_get_dir_page() will issue RPC to fetch
350 ll_release_page(page,
351 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
355 page_cache_release(page);
356 page = ERR_PTR(-EIO);
360 TREE_READ_UNLOCK_IRQ(mapping);
366 struct page *ll_get_dir_page(struct file *filp, struct inode *dir, __u64 hash,
367 struct ll_dir_chain *chain)
369 ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
370 struct address_space *mapping = dir->i_mapping;
371 struct lustre_handle lockh;
372 struct lu_dirpage *dp;
379 struct ll_inode_info *lli = ll_i2info(dir);
380 int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
383 rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
384 ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
386 struct ldlm_enqueue_info einfo = { LDLM_IBITS, mode,
387 ll_md_blocking_ast, ldlm_completion_ast,
389 struct lookup_intent it = { .it_op = IT_READDIR };
390 struct ptlrpc_request *request;
391 struct md_op_data *op_data;
393 op_data = ll_prep_md_op_data(NULL, dir, NULL, NULL, 0, 0,
394 LUSTRE_OPC_ANY, NULL);
396 return (void *)op_data;
398 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
399 op_data, &lockh, NULL, 0, NULL, 0);
401 ll_finish_md_op_data(op_data);
403 request = (struct ptlrpc_request *)it.d.lustre.it_data;
405 ptlrpc_req_finished(request);
407 CERROR("lock enqueue: "DFID" at "LPU64": rc %d\n",
408 PFID(ll_inode2fid(dir)), hash, rc);
412 /* for cross-ref object, l_ast_data of the lock may not be set,
413 * we reset it here */
414 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
417 ldlm_lock_dump_handle(D_OTHER, &lockh);
419 cfs_mutex_lock(&lli->lli_readdir_mutex);
420 page = ll_dir_page_locate(dir, &lhash, &start, &end);
422 CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
423 PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
424 GOTO(out_unlock, page);
425 } else if (page != NULL) {
427 * XXX nikita: not entirely correct handling of a corner case:
428 * suppose hash chain of entries with hash value HASH crosses
429 * border between pages P0 and P1. First both P0 and P1 are
430 * cached, seekdir() is called for some entry from the P0 part
431 * of the chain. Later P0 goes out of cache. telldir(HASH)
432 * happens and finds P1, as it starts with matching hash
433 * value. Remaining entries from P0 part of the chain are
434 * skipped. (Is that really a bug?)
436 * Possible solutions: 0. don't cache P1 is such case, handle
437 * it as an "overflow" page. 1. invalidate all pages at
438 * once. 2. use HASH|1 as an index for P1.
440 GOTO(hash_collision, page);
443 page = read_cache_page(mapping, hash_x_index(hash, hash64),
444 (filler_t*)mapping->a_ops->readpage, filp);
446 CERROR("read cache page: "DFID" at "LPU64": rc %ld\n",
447 PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
448 GOTO(out_unlock, page);
453 if (!PageUptodate(page)) {
454 CERROR("page not updated: "DFID" at "LPU64": rc %d\n",
455 PFID(ll_inode2fid(dir)), hash, -5);
458 if (!PageChecked(page))
459 ll_check_page(dir, page);
460 if (PageError(page)) {
461 CERROR("page error: "DFID" at "LPU64": rc %d\n",
462 PFID(ll_inode2fid(dir)), hash, -5);
466 dp = page_address(page);
467 if (BITS_PER_LONG == 32 && hash64) {
468 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
469 end = le64_to_cpu(dp->ldp_hash_end) >> 32;
472 start = le64_to_cpu(dp->ldp_hash_start);
473 end = le64_to_cpu(dp->ldp_hash_end);
477 LASSERT(start == lhash);
478 CWARN("Page-wide hash collision: "LPU64"\n", end);
479 if (BITS_PER_LONG == 32 && hash64)
480 CWARN("Real page-wide hash collision at ["LPU64" "LPU64
481 "] with hash "LPU64"\n",
482 le64_to_cpu(dp->ldp_hash_start),
483 le64_to_cpu(dp->ldp_hash_end), hash);
485 * Fetch whole overflow chain...
492 cfs_mutex_unlock(&lli->lli_readdir_mutex);
493 ldlm_lock_decref(&lockh, mode);
497 ll_release_page(page, 1);
498 page = ERR_PTR(-EIO);
502 int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
504 struct inode *inode = filp->f_dentry->d_inode;
505 struct ll_inode_info *info = ll_i2info(inode);
506 struct ll_sb_info *sbi = ll_i2sbi(inode);
507 struct ll_file_data *fd = LUSTRE_FPRIVATE(filp);
508 __u64 pos = fd->fd_dir.lfd_pos;
509 int api32 = ll_need_32bit_api(sbi);
510 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
512 struct ll_dir_chain chain;
517 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
518 inode->i_ino, inode->i_generation, inode,
519 (unsigned long)pos, i_size_read(inode), api32);
521 if (pos == MDS_DIR_END_OFF)
529 ll_dir_chain_init(&chain);
531 fd->fd_dir.lfd_next = pos;
532 page = ll_get_dir_page(filp, inode, pos, &chain);
534 while (rc == 0 && !done) {
535 struct lu_dirpage *dp;
536 struct lu_dirent *ent;
540 * If page is empty (end of directory is reached),
543 __u64 hash = MDS_DIR_END_OFF;
546 dp = page_address(page);
547 for (ent = lu_dirent_start(dp); ent != NULL && !done;
548 ent = lu_dirent_next(ent)) {
556 * XXX: implement correct swabbing here.
559 hash = le64_to_cpu(ent->lde_hash);
562 * Skip until we find target hash
567 namelen = le16_to_cpu(ent->lde_namelen);
578 fid_le_to_cpu(&fid, &ent->lde_fid);
579 ino = cl_fid_build_ino(&fid, api32);
580 type = ll_dirent_type_get(ent);
581 /* For 'll_nfs_get_name_filldir()', it will try
582 * to access the 'ent' through its 'lde_name',
583 * so the parameter 'name' for 'filldir()' must
584 * be part of the 'ent'. */
585 done = filldir(cookie, ent->lde_name, namelen,
588 next = le64_to_cpu(dp->ldp_hash_end);
591 if (pos == MDS_DIR_END_OFF) {
593 * End of directory reached.
596 ll_release_page(page, 0);
597 } else if (1 /* chain is exhausted*/) {
599 * Normal case: continue to the next
602 ll_release_page(page,
603 le32_to_cpu(dp->ldp_flags) &
605 fd->fd_dir.lfd_next = pos;
606 page = ll_get_dir_page(filp, inode, pos,
610 * go into overflow page.
612 LASSERT(le32_to_cpu(dp->ldp_flags) &
614 ll_release_page(page, 1);
618 ll_release_page(page, 0);
622 CERROR("error reading dir "DFID" at %lu: rc %d\n",
623 PFID(&info->lli_fid), (unsigned long)pos, rc);
627 fd->fd_dir.lfd_pos = pos;
628 if (pos == MDS_DIR_END_OFF) {
630 filp->f_pos = LL_DIR_END_OFF_32BIT;
632 filp->f_pos = LL_DIR_END_OFF;
635 filp->f_pos = pos >> 32;
639 filp->f_version = inode->i_version;
640 touch_atime(filp->f_vfsmnt, filp->f_dentry);
642 ll_dir_chain_fini(&chain);
646 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
651 int ll_send_mgc_param(struct obd_export *mgc, char *string)
653 struct mgs_send_param *msp;
660 strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
661 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
662 sizeof(struct mgs_send_param), msp, NULL);
664 CERROR("Failed to set parameter: %d\n", rc);
670 char *ll_get_fsname(struct inode *inode)
672 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
676 OBD_ALLOC(fsname, MGS_PARAM_MAXLEN);
677 len = strlen(lsi->lsi_lmd->lmd_profile);
678 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
679 if (ptr && (strcmp(ptr, "-client") == 0))
681 strncpy(fsname, lsi->lsi_lmd->lmd_profile, len);
687 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
690 struct ll_sb_info *sbi = ll_i2sbi(inode);
691 struct md_op_data *op_data;
692 struct ptlrpc_request *req = NULL;
694 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
695 struct obd_device *mgc = lsi->lsi_mgc;
696 char *fsname = NULL, *param = NULL;
701 * This is coming from userspace, so should be in
702 * local endian. But the MDS would like it in little
703 * endian, so we swab it before we send it.
705 switch (lump->lmm_magic) {
706 case LOV_USER_MAGIC_V1: {
707 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
708 lustre_swab_lov_user_md_v1(lump);
709 lum_size = sizeof(struct lov_user_md_v1);
712 case LOV_USER_MAGIC_V3: {
713 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
714 lustre_swab_lov_user_md_v3(
715 (struct lov_user_md_v3 *)lump);
716 lum_size = sizeof(struct lov_user_md_v3);
720 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
721 " %#08x != %#08x nor %#08x\n",
722 lump->lmm_magic, LOV_USER_MAGIC_V1,
728 lum_size = sizeof(struct lov_user_md_v1);
731 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
732 LUSTRE_OPC_ANY, NULL);
734 RETURN(PTR_ERR(op_data));
736 /* swabbing is done in lov_setstripe() on server side */
737 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
738 NULL, 0, &req, NULL);
739 ll_finish_md_op_data(op_data);
740 ptlrpc_req_finished(req);
742 if (rc != -EPERM && rc != -EACCES)
743 CERROR("mdc_setattr fails: rc = %d\n", rc);
746 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
747 LOV_USER_MAGIC_V3 have the same initial fields so we do not
748 need the make the distiction between the 2 versions */
749 if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
750 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
752 /* Get fsname and assume devname to be -MDT0000. */
753 fsname = ll_get_fsname(inode);
754 /* Set root stripesize */
755 sprintf(param, "%s-MDT0000.lov.stripesize=%u", fsname,
756 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
757 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
761 /* Set root stripecount */
762 sprintf(param, "%s-MDT0000.lov.stripecount=%hd", fsname,
763 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
764 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
768 /* Set root stripeoffset */
769 sprintf(param, "%s-MDT0000.lov.stripeoffset=%hd", fsname,
770 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
771 (typeof(lump->lmm_stripe_offset))(-1));
772 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
777 OBD_FREE(fsname, MGS_PARAM_MAXLEN);
779 OBD_FREE(param, MGS_PARAM_MAXLEN);
784 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
785 int *lmm_size, struct ptlrpc_request **request)
787 struct ll_sb_info *sbi = ll_i2sbi(inode);
788 struct mdt_body *body;
789 struct lov_mds_md *lmm = NULL;
790 struct ptlrpc_request *req = NULL;
792 struct md_op_data *op_data;
794 rc = ll_get_max_mdsize(sbi, &lmmsize);
798 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
799 0, lmmsize, LUSTRE_OPC_ANY,
802 RETURN(PTR_ERR(op_data));
804 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
805 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
806 ll_finish_md_op_data(op_data);
808 CDEBUG(D_INFO, "md_getattr failed on inode "
809 "%lu/%u: rc %d\n", inode->i_ino,
810 inode->i_generation, rc);
814 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
815 LASSERT(body != NULL);
817 lmmsize = body->eadatasize;
819 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
821 GOTO(out, rc = -ENODATA);
824 lmm = req_capsule_server_sized_get(&req->rq_pill,
825 &RMF_MDT_MD, lmmsize);
826 LASSERT(lmm != NULL);
829 * This is coming from the MDS, so is probably in
830 * little endian. We convert it to host endian before
831 * passing it to userspace.
833 /* We don't swab objects for directories */
834 switch (le32_to_cpu(lmm->lmm_magic)) {
836 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
837 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
840 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
841 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
844 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
855 * Get MDT index for the inode.
857 int ll_get_mdt_idx(struct inode *inode)
859 struct ll_sb_info *sbi = ll_i2sbi(inode);
860 struct md_op_data *op_data;
864 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
865 0, LUSTRE_OPC_ANY, NULL);
867 RETURN(PTR_ERR(op_data));
869 op_data->op_valid |= OBD_MD_MDTIDX;
870 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
871 mdtidx = op_data->op_mds;
872 ll_finish_md_op_data(op_data);
874 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
880 static int copy_and_ioctl(int cmd, struct obd_export *exp, void *data, int len)
888 if (cfs_copy_from_user(ptr, data, len)) {
892 rc = obd_iocontrol(cmd, exp, len, data, NULL);
897 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
899 int cmd = qctl->qc_cmd;
900 int type = qctl->qc_type;
901 int id = qctl->qc_id;
902 int valid = qctl->qc_valid;
907 case LUSTRE_Q_INVALIDATE:
908 case LUSTRE_Q_FINVALIDATE:
913 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
914 sbi->ll_flags & LL_SBI_RMT_CLIENT)
918 if (((type == USRQUOTA && cfs_curproc_euid() != id) ||
919 (type == GRPQUOTA && !in_egroup_p(id))) &&
920 (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
921 sbi->ll_flags & LL_SBI_RMT_CLIENT))
927 CERROR("unsupported quotactl op: %#x\n", cmd);
931 if (valid != QC_GENERAL) {
932 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
935 if (cmd == Q_GETINFO)
936 qctl->qc_cmd = Q_GETOINFO;
937 else if (cmd == Q_GETQUOTA)
938 qctl->qc_cmd = Q_GETOQUOTA;
944 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
945 sizeof(*qctl), qctl, NULL);
948 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
949 sizeof(*qctl), qctl, NULL);
952 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
953 sizeof(*qctl), qctl, NULL);
955 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
957 sizeof(*qctl), qctl, NULL);
969 struct obd_quotactl *oqctl;
971 OBD_ALLOC_PTR(oqctl);
975 QCTL_COPY(oqctl, qctl);
976 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
978 if (rc != -EALREADY && cmd == Q_QUOTAON) {
979 oqctl->qc_cmd = Q_QUOTAOFF;
980 obd_quotactl(sbi->ll_md_exp, oqctl);
985 /* If QIF_SPACE is not set, client should collect the
986 * space usage from OSSs by itself */
987 if (cmd == Q_GETQUOTA &&
988 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
989 !oqctl->qc_dqblk.dqb_curspace) {
990 struct obd_quotactl *oqctl_tmp;
992 OBD_ALLOC_PTR(oqctl_tmp);
993 if (oqctl_tmp == NULL)
994 GOTO(out, rc = -ENOMEM);
996 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
997 oqctl_tmp->qc_id = oqctl->qc_id;
998 oqctl_tmp->qc_type = oqctl->qc_type;
1000 /* collect space usage from OSTs */
1001 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1002 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1003 if (!rc || rc == -EREMOTEIO) {
1004 oqctl->qc_dqblk.dqb_curspace =
1005 oqctl_tmp->qc_dqblk.dqb_curspace;
1006 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1009 /* collect space & inode usage from MDTs */
1010 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1011 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1012 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1013 if (!rc || rc == -EREMOTEIO) {
1014 oqctl->qc_dqblk.dqb_curspace +=
1015 oqctl_tmp->qc_dqblk.dqb_curspace;
1016 oqctl->qc_dqblk.dqb_curinodes =
1017 oqctl_tmp->qc_dqblk.dqb_curinodes;
1018 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1020 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1023 OBD_FREE_PTR(oqctl_tmp);
1026 QCTL_COPY(qctl, oqctl);
1027 OBD_FREE_PTR(oqctl);
1033 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1035 struct inode *inode = file->f_dentry->d_inode;
1036 struct ll_sb_info *sbi = ll_i2sbi(inode);
1037 struct obd_ioctl_data *data;
1041 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1042 inode->i_ino, inode->i_generation, inode, cmd);
1044 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1045 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1048 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1050 case FSFILT_IOC_GETFLAGS:
1051 case FSFILT_IOC_SETFLAGS:
1052 RETURN(ll_iocontrol(inode, file, cmd, arg));
1053 case FSFILT_IOC_GETVERSION_OLD:
1054 case FSFILT_IOC_GETVERSION:
1055 RETURN(put_user(inode->i_generation, (int *)arg));
1056 /* We need to special case any other ioctls we want to handle,
1057 * to send them to the MDS/OST as appropriate and to properly
1058 * network encode the arg field.
1059 case FSFILT_IOC_SETVERSION_OLD:
1060 case FSFILT_IOC_SETVERSION:
1062 case LL_IOC_GET_MDTIDX: {
1065 mdtidx = ll_get_mdt_idx(inode);
1069 if (put_user((int)mdtidx, (int*)arg))
1074 case IOC_MDC_LOOKUP: {
1075 struct ptlrpc_request *request = NULL;
1076 int namelen, len = 0;
1079 struct md_op_data *op_data;
1081 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1086 filename = data->ioc_inlbuf1;
1087 namelen = strlen(filename);
1090 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1091 GOTO(out_free, rc = -EINVAL);
1094 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1095 0, LUSTRE_OPC_ANY, NULL);
1096 if (IS_ERR(op_data))
1097 GOTO(out_free, rc = PTR_ERR(op_data));
1099 op_data->op_valid = OBD_MD_FLID;
1100 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1101 ll_finish_md_op_data(op_data);
1103 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1106 ptlrpc_req_finished(request);
1109 obd_ioctl_freedata(buf, len);
1112 case LL_IOC_LOV_SETSTRIPE: {
1113 struct lov_user_md_v3 lumv3;
1114 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1115 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1116 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1118 int set_default = 0;
1120 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1121 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1122 sizeof(lumv3p->lmm_objects[0]));
1123 /* first try with v1 which is smaller than v3 */
1124 if (cfs_copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1127 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1128 if (cfs_copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1132 if (inode->i_sb->s_root == file->f_dentry)
1135 /* in v1 and v3 cases lumv1 points to data */
1136 rc = ll_dir_setstripe(inode, lumv1, set_default);
1140 case LL_IOC_OBD_STATFS:
1141 RETURN(ll_obd_statfs(inode, (void *)arg));
1142 case LL_IOC_LOV_GETSTRIPE:
1143 case LL_IOC_MDC_GETINFO:
1144 case IOC_MDC_GETFILEINFO:
1145 case IOC_MDC_GETFILESTRIPE: {
1146 struct ptlrpc_request *request = NULL;
1147 struct lov_user_md *lump;
1148 struct lov_mds_md *lmm = NULL;
1149 struct mdt_body *body;
1150 char *filename = NULL;
1153 if (cmd == IOC_MDC_GETFILEINFO ||
1154 cmd == IOC_MDC_GETFILESTRIPE) {
1155 filename = getname((const char *)arg);
1156 if (IS_ERR(filename))
1157 RETURN(PTR_ERR(filename));
1159 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1160 &lmmsize, &request);
1162 rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1166 body = req_capsule_server_get(&request->rq_pill,
1168 LASSERT(body != NULL);
1174 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1175 cmd == LL_IOC_MDC_GETINFO))
1176 GOTO(skip_lmm, rc = 0);
1181 if (cmd == IOC_MDC_GETFILESTRIPE ||
1182 cmd == LL_IOC_LOV_GETSTRIPE) {
1183 lump = (struct lov_user_md *)arg;
1185 struct lov_user_mds_data *lmdp;
1186 lmdp = (struct lov_user_mds_data *)arg;
1187 lump = &lmdp->lmd_lmm;
1189 if (cfs_copy_to_user(lump, lmm, lmmsize)) {
1190 if (cfs_copy_to_user(lump, lmm, sizeof(*lump)))
1191 GOTO(out_req, rc = -EFAULT);
1195 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1196 struct lov_user_mds_data *lmdp;
1199 st.st_dev = inode->i_sb->s_dev;
1200 st.st_mode = body->mode;
1201 st.st_nlink = body->nlink;
1202 st.st_uid = body->uid;
1203 st.st_gid = body->gid;
1204 st.st_rdev = body->rdev;
1205 st.st_size = body->size;
1206 st.st_blksize = CFS_PAGE_SIZE;
1207 st.st_blocks = body->blocks;
1208 st.st_atime = body->atime;
1209 st.st_mtime = body->mtime;
1210 st.st_ctime = body->ctime;
1211 st.st_ino = inode->i_ino;
1213 lmdp = (struct lov_user_mds_data *)arg;
1214 if (cfs_copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1215 GOTO(out_req, rc = -EFAULT);
1220 ptlrpc_req_finished(request);
1225 case IOC_LOV_GETINFO: {
1226 struct lov_user_mds_data *lumd;
1227 struct lov_stripe_md *lsm;
1228 struct lov_user_md *lum;
1229 struct lov_mds_md *lmm;
1233 lumd = (struct lov_user_mds_data *)arg;
1234 lum = &lumd->lmd_lmm;
1236 rc = ll_get_max_mdsize(sbi, &lmmsize);
1240 OBD_ALLOC_LARGE(lmm, lmmsize);
1241 if (cfs_copy_from_user(lmm, lum, lmmsize))
1242 GOTO(free_lmm, rc = -EFAULT);
1244 switch (lmm->lmm_magic) {
1245 case LOV_USER_MAGIC_V1:
1246 if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1248 /* swab objects first so that stripes num will be sane */
1249 lustre_swab_lov_user_md_objects(
1250 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1251 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1252 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1254 case LOV_USER_MAGIC_V3:
1255 if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1257 /* swab objects first so that stripes num will be sane */
1258 lustre_swab_lov_user_md_objects(
1259 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1260 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1261 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1264 GOTO(free_lmm, rc = -EINVAL);
1267 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1269 GOTO(free_lmm, rc = -ENOMEM);
1271 /* Perform glimpse_size operation. */
1272 memset(&st, 0, sizeof(st));
1274 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1278 if (cfs_copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1279 GOTO(free_lsm, rc = -EFAULT);
1283 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1285 OBD_FREE_LARGE(lmm, lmmsize);
1288 case OBD_IOC_LLOG_CATINFO: {
1289 RETURN(-EOPNOTSUPP);
1291 case OBD_IOC_QUOTACHECK: {
1292 struct obd_quotactl *oqctl;
1295 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1296 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1299 OBD_ALLOC_PTR(oqctl);
1302 oqctl->qc_type = arg;
1303 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1305 CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1309 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1311 CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1313 OBD_FREE_PTR(oqctl);
1316 case OBD_IOC_POLL_QUOTACHECK: {
1317 struct if_quotacheck *check;
1319 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1320 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1323 OBD_ALLOC_PTR(check);
1327 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1330 CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1331 if (cfs_copy_to_user((void *)arg, check,
1333 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1337 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1340 CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1341 if (cfs_copy_to_user((void *)arg, check,
1343 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1347 OBD_FREE_PTR(check);
1350 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0)
1351 case LL_IOC_QUOTACTL_18: {
1352 /* copy the old 1.x quota struct for internal use, then copy
1353 * back into old format struct. For 1.8 compatibility. */
1354 struct if_quotactl_18 *qctl_18;
1355 struct if_quotactl *qctl_20;
1357 OBD_ALLOC_PTR(qctl_18);
1361 OBD_ALLOC_PTR(qctl_20);
1363 GOTO(out_quotactl_18, rc = -ENOMEM);
1365 if (cfs_copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1366 GOTO(out_quotactl_20, rc = -ENOMEM);
1368 QCTL_COPY(qctl_20, qctl_18);
1369 qctl_20->qc_idx = 0;
1371 /* XXX: dqb_valid was borrowed as a flag to mark that
1372 * only mds quota is wanted */
1373 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1374 qctl_18->qc_dqblk.dqb_valid) {
1375 qctl_20->qc_valid = QC_MDTIDX;
1376 qctl_20->qc_dqblk.dqb_valid = 0;
1377 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1378 qctl_20->qc_valid = QC_UUID;
1379 qctl_20->obd_uuid = qctl_18->obd_uuid;
1381 qctl_20->qc_valid = QC_GENERAL;
1384 rc = quotactl_ioctl(sbi, qctl_20);
1387 QCTL_COPY(qctl_18, qctl_20);
1388 qctl_18->obd_uuid = qctl_20->obd_uuid;
1390 if (cfs_copy_to_user((void *)arg, qctl_18,
1396 OBD_FREE_PTR(qctl_20);
1398 OBD_FREE_PTR(qctl_18);
1402 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1403 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0) */
1404 case LL_IOC_QUOTACTL: {
1405 struct if_quotactl *qctl;
1407 OBD_ALLOC_PTR(qctl);
1411 if (cfs_copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1412 GOTO(out_quotactl, rc = -EFAULT);
1414 rc = quotactl_ioctl(sbi, qctl);
1416 if (rc == 0 && cfs_copy_to_user((void *)arg,qctl,sizeof(*qctl)))
1423 case OBD_IOC_GETDTNAME:
1424 case OBD_IOC_GETMDNAME:
1425 RETURN(ll_get_obd_name(inode, cmd, arg));
1426 case LL_IOC_FLUSHCTX:
1427 RETURN(ll_flush_ctx(inode));
1428 #ifdef CONFIG_FS_POSIX_ACL
1429 case LL_IOC_RMTACL: {
1430 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1431 inode == inode->i_sb->s_root->d_inode) {
1432 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1434 LASSERT(fd != NULL);
1435 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1437 fd->fd_flags |= LL_FILE_RMTACL;
1443 case LL_IOC_GETOBDCOUNT: {
1445 struct obd_export *exp;
1447 if (cfs_copy_from_user(&count, (int *)arg, sizeof(int)))
1450 /* get ost count when count is zero, get mdt count otherwise */
1451 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1452 vallen = sizeof(count);
1453 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1454 KEY_TGT_COUNT, &vallen, &count, NULL);
1456 CERROR("get target count failed: %d\n", rc);
1460 if (cfs_copy_to_user((int *)arg, &count, sizeof(int)))
1465 case LL_IOC_PATH2FID:
1466 if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
1467 sizeof(struct lu_fid)))
1470 case LL_IOC_GET_CONNECT_FLAGS: {
1471 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1473 case OBD_IOC_CHANGELOG_SEND:
1474 case OBD_IOC_CHANGELOG_CLEAR:
1475 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1476 sizeof(struct ioc_changelog));
1478 case OBD_IOC_FID2PATH:
1479 RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
1480 case LL_IOC_HSM_CT_START:
1481 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1482 sizeof(struct lustre_kernelcomm));
1486 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp,0,NULL,(void *)arg));
1490 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1492 struct inode *inode = file->f_mapping->host;
1493 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1494 struct ll_sb_info *sbi = ll_i2sbi(inode);
1495 int api32 = ll_need_32bit_api(sbi);
1496 loff_t ret = -EINVAL;
1499 cfs_mutex_lock(&inode->i_mutex);
1504 offset += file->f_pos;
1510 offset += LL_DIR_END_OFF_32BIT;
1512 offset += LL_DIR_END_OFF;
1519 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1520 (!api32 && offset <= LL_DIR_END_OFF))) {
1521 if (offset != file->f_pos) {
1522 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1523 (!api32 && offset == LL_DIR_END_OFF))
1524 fd->fd_dir.lfd_pos = MDS_DIR_END_OFF;
1525 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1526 fd->fd_dir.lfd_pos = offset << 32;
1528 fd->fd_dir.lfd_pos = offset;
1529 file->f_pos = offset;
1530 file->f_version = 0;
1537 cfs_mutex_unlock(&inode->i_mutex);
1541 int ll_dir_open(struct inode *inode, struct file *file)
1544 RETURN(ll_file_open(inode, file));
1547 int ll_dir_release(struct inode *inode, struct file *file)
1550 RETURN(ll_file_release(inode, file));
1553 struct file_operations ll_dir_operations = {
1554 .llseek = ll_dir_seek,
1555 .open = ll_dir_open,
1556 .release = ll_dir_release,
1557 .read = generic_read_dir,
1558 .readdir = ll_readdir,
1559 .unlocked_ioctl = ll_dir_ioctl,