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, 2015, Intel Corporation.
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/uaccess.h>
45 #include <linux/buffer_head.h> /* for wait_on_buffer */
46 #include <linux/pagevec.h>
47 #include <linux/prefetch.h>
49 #define DEBUG_SUBSYSTEM S_LLITE
51 #include "../include/obd_support.h"
52 #include "../include/obd_class.h"
53 #include "../include/lustre_lib.h"
54 #include "../include/lustre/lustre_idl.h"
55 #include "../include/lustre_lite.h"
56 #include "../include/lustre_dlm.h"
57 #include "../include/lustre_fid.h"
58 #include "../include/lustre_kernelcomm.h"
59 #include "llite_internal.h"
62 * (new) readdir implementation overview.
64 * Original lustre readdir implementation cached exact copy of raw directory
65 * pages on the client. These pages were indexed in client page cache by
66 * logical offset in the directory file. This design, while very simple and
67 * intuitive had some inherent problems:
69 * . it implies that byte offset to the directory entry serves as a
70 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
71 * ext3/htree directory entries may move due to splits, and more
74 * . it is incompatible with the design of split directories for cmd3,
75 * that assumes that names are distributed across nodes based on their
76 * hash, and so readdir should be done in hash order.
78 * New readdir implementation does readdir in hash order, and uses hash of a
79 * file name as a telldir/seekdir cookie. This led to number of complications:
81 * . hash is not unique, so it cannot be used to index cached directory
82 * pages on the client (note, that it requires a whole pageful of hash
83 * collided entries to cause two pages to have identical hashes);
85 * . hash is not unique, so it cannot, strictly speaking, be used as an
86 * entry cookie. ext3/htree has the same problem and lustre implementation
87 * mimics their solution: seekdir(hash) positions directory at the first
88 * entry with the given hash.
94 * Client caches directory pages using hash of the first entry as an index. As
95 * noted above hash is not unique, so this solution doesn't work as is:
96 * special processing is needed for "page hash chains" (i.e., sequences of
97 * pages filled with entries all having the same hash value).
99 * First, such chains have to be detected. To this end, server returns to the
100 * client the hash of the first entry on the page next to one returned. When
101 * client detects that this hash is the same as hash of the first entry on the
102 * returned page, page hash collision has to be handled. Pages in the
103 * hash chain, except first one, are termed "overflow pages".
105 * Solution to index uniqueness problem is to not cache overflow
106 * pages. Instead, when page hash collision is detected, all overflow pages
107 * from emerging chain are immediately requested from the server and placed in
108 * a special data structure (struct ll_dir_chain). This data structure is used
109 * by ll_readdir() to process entries from overflow pages. When readdir
110 * invocation finishes, overflow pages are discarded. If page hash collision
111 * chain weren't completely processed, next call to readdir will again detect
112 * page hash collision, again read overflow pages in, process next portion of
113 * entries and again discard the pages. This is not as wasteful as it looks,
114 * because, given reasonable hash, page hash collisions are extremely rare.
116 * 1. directory positioning
118 * When seekdir(hash) is called, original
129 * identification of and access to overflow pages
133 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
134 * a header lu_dirpage which describes the start/end hash, and whether this
135 * page is empty (contains no dir entry) or hash collide with next page.
136 * After client receives reply, several pages will be integrated into dir page
137 * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the lu_dirpage
138 * for this integrated page will be adjusted. See lmv_adjust_dirpages().
142 /* returns the page unlocked, but with a reference */
143 static int ll_dir_filler(void *_hash
, struct page
*page0
)
145 struct inode
*inode
= page0
->mapping
->host
;
146 int hash64
= ll_i2sbi(inode
)->ll_flags
& LL_SBI_64BIT_HASH
;
147 struct obd_export
*exp
= ll_i2sbi(inode
)->ll_md_exp
;
148 struct ptlrpc_request
*request
;
149 struct mdt_body
*body
;
150 struct md_op_data
*op_data
;
151 __u64 hash
= *((__u64
*)_hash
);
152 struct page
**page_pool
;
154 struct lu_dirpage
*dp
;
155 int max_pages
= ll_i2sbi(inode
)->ll_md_brw_size
>> PAGE_SHIFT
;
156 int nrdpgs
= 0; /* number of pages read actually */
161 CDEBUG(D_VFSTRACE
, "VFS Op:inode=%lu/%u(%p) hash %llu\n",
162 inode
->i_ino
, inode
->i_generation
, inode
, hash
);
164 LASSERT(max_pages
> 0 && max_pages
<= MD_MAX_BRW_PAGES
);
166 page_pool
= kcalloc(max_pages
, sizeof(page
), GFP_NOFS
);
168 page_pool
[0] = page0
;
173 for (npages
= 1; npages
< max_pages
; npages
++) {
174 page
= page_cache_alloc_cold(inode
->i_mapping
);
177 page_pool
[npages
] = page
;
180 op_data
= ll_prep_md_op_data(NULL
, inode
, NULL
, NULL
, 0, 0,
181 LUSTRE_OPC_ANY
, NULL
);
182 op_data
->op_npages
= npages
;
183 op_data
->op_offset
= hash
;
184 rc
= md_readpage(exp
, op_data
, page_pool
, &request
);
185 ll_finish_md_op_data(op_data
);
187 /* page0 is special, which was added into page cache early */
188 delete_from_page_cache(page0
);
189 } else if (rc
== 0) {
190 body
= req_capsule_server_get(&request
->rq_pill
, &RMF_MDT_BODY
);
191 /* Checked by mdc_readpage() */
192 if (body
->valid
& OBD_MD_FLSIZE
)
193 cl_isize_write(inode
, body
->size
);
195 nrdpgs
= (request
->rq_bulk
->bd_nob_transferred
+PAGE_SIZE
-1)
197 SetPageUptodate(page0
);
200 ptlrpc_req_finished(request
);
202 CDEBUG(D_VFSTRACE
, "read %d/%d pages\n", nrdpgs
, npages
);
204 for (i
= 1; i
< npages
; i
++) {
205 unsigned long offset
;
210 if (rc
< 0 || i
>= nrdpgs
) {
215 SetPageUptodate(page
);
218 hash
= le64_to_cpu(dp
->ldp_hash_start
);
221 offset
= hash_x_index(hash
, hash64
);
223 prefetchw(&page
->flags
);
224 ret
= add_to_page_cache_lru(page
, inode
->i_mapping
, offset
,
229 CDEBUG(D_VFSTRACE
, "page %lu add to page cache failed: %d\n",
235 if (page_pool
!= &page0
)
240 void ll_release_page(struct page
*page
, int remove
)
245 if (likely(page
->mapping
))
246 truncate_complete_page(page
->mapping
, page
);
253 * Find, kmap and return page that contains given hash.
255 static struct page
*ll_dir_page_locate(struct inode
*dir
, __u64
*hash
,
256 __u64
*start
, __u64
*end
)
258 int hash64
= ll_i2sbi(dir
)->ll_flags
& LL_SBI_64BIT_HASH
;
259 struct address_space
*mapping
= dir
->i_mapping
;
261 * Complement of hash is used as an index so that
262 * radix_tree_gang_lookup() can be used to find a page with starting
263 * hash _smaller_ than one we are looking for.
265 unsigned long offset
= hash_x_index(*hash
, hash64
);
269 spin_lock_irq(&mapping
->tree_lock
);
270 found
= radix_tree_gang_lookup(&mapping
->page_tree
,
271 (void **)&page
, offset
, 1);
272 if (found
> 0 && !radix_tree_exceptional_entry(page
)) {
273 struct lu_dirpage
*dp
;
276 spin_unlock_irq(&mapping
->tree_lock
);
278 * In contrast to find_lock_page() we are sure that directory
279 * page cannot be truncated (while DLM lock is held) and,
280 * hence, can avoid restart.
282 * In fact, page cannot be locked here at all, because
283 * ll_dir_filler() does synchronous io.
285 wait_on_page_locked(page
);
286 if (PageUptodate(page
)) {
288 if (BITS_PER_LONG
== 32 && hash64
) {
289 *start
= le64_to_cpu(dp
->ldp_hash_start
) >> 32;
290 *end
= le64_to_cpu(dp
->ldp_hash_end
) >> 32;
293 *start
= le64_to_cpu(dp
->ldp_hash_start
);
294 *end
= le64_to_cpu(dp
->ldp_hash_end
);
296 LASSERTF(*start
<= *hash
, "start = %#llx,end = %#llx,hash = %#llx\n",
297 *start
, *end
, *hash
);
298 CDEBUG(D_VFSTRACE
, "page %lu [%llu %llu], hash %llu\n",
299 offset
, *start
, *end
, *hash
);
301 ll_release_page(page
, 0);
303 } else if (*end
!= *start
&& *hash
== *end
) {
305 * upon hash collision, remove this page,
306 * otherwise put page reference, and
307 * ll_get_dir_page() will issue RPC to fetch
310 ll_release_page(page
,
311 le32_to_cpu(dp
->ldp_flags
) & LDF_COLLIDE
);
316 page
= ERR_PTR(-EIO
);
320 spin_unlock_irq(&mapping
->tree_lock
);
326 struct page
*ll_get_dir_page(struct inode
*dir
, __u64 hash
,
327 struct ll_dir_chain
*chain
)
329 ldlm_policy_data_t policy
= {.l_inodebits
= {MDS_INODELOCK_UPDATE
} };
330 struct address_space
*mapping
= dir
->i_mapping
;
331 struct lustre_handle lockh
;
332 struct lu_dirpage
*dp
;
339 struct ll_inode_info
*lli
= ll_i2info(dir
);
340 int hash64
= ll_i2sbi(dir
)->ll_flags
& LL_SBI_64BIT_HASH
;
343 rc
= md_lock_match(ll_i2sbi(dir
)->ll_md_exp
, LDLM_FL_BLOCK_GRANTED
,
344 ll_inode2fid(dir
), LDLM_IBITS
, &policy
, mode
, &lockh
);
346 struct ldlm_enqueue_info einfo
= {
347 .ei_type
= LDLM_IBITS
,
349 .ei_cb_bl
= ll_md_blocking_ast
,
350 .ei_cb_cp
= ldlm_completion_ast
,
352 struct lookup_intent it
= { .it_op
= IT_READDIR
};
353 struct ptlrpc_request
*request
;
354 struct md_op_data
*op_data
;
356 op_data
= ll_prep_md_op_data(NULL
, dir
, dir
, NULL
, 0, 0,
357 LUSTRE_OPC_ANY
, NULL
);
359 return (void *)op_data
;
361 rc
= md_enqueue(ll_i2sbi(dir
)->ll_md_exp
, &einfo
, &it
,
362 op_data
, &lockh
, NULL
, 0, NULL
, 0);
364 ll_finish_md_op_data(op_data
);
366 request
= (struct ptlrpc_request
*)it
.d
.lustre
.it_data
;
368 ptlrpc_req_finished(request
);
370 CERROR("lock enqueue: " DFID
" at %llu: rc %d\n",
371 PFID(ll_inode2fid(dir
)), hash
, rc
);
375 CDEBUG(D_INODE
, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
376 dir
, dir
->i_ino
, dir
->i_generation
);
377 md_set_lock_data(ll_i2sbi(dir
)->ll_md_exp
,
378 &it
.d
.lustre
.it_lock_handle
, dir
, NULL
);
380 /* for cross-ref object, l_ast_data of the lock may not be set,
383 md_set_lock_data(ll_i2sbi(dir
)->ll_md_exp
, &lockh
.cookie
,
386 ldlm_lock_dump_handle(D_OTHER
, &lockh
);
388 mutex_lock(&lli
->lli_readdir_mutex
);
389 page
= ll_dir_page_locate(dir
, &lhash
, &start
, &end
);
391 CERROR("dir page locate: "DFID
" at %llu: rc %ld\n",
392 PFID(ll_inode2fid(dir
)), lhash
, PTR_ERR(page
));
396 * XXX nikita: not entirely correct handling of a corner case:
397 * suppose hash chain of entries with hash value HASH crosses
398 * border between pages P0 and P1. First both P0 and P1 are
399 * cached, seekdir() is called for some entry from the P0 part
400 * of the chain. Later P0 goes out of cache. telldir(HASH)
401 * happens and finds P1, as it starts with matching hash
402 * value. Remaining entries from P0 part of the chain are
403 * skipped. (Is that really a bug?)
405 * Possible solutions: 0. don't cache P1 is such case, handle
406 * it as an "overflow" page. 1. invalidate all pages at
407 * once. 2. use HASH|1 as an index for P1.
412 page
= read_cache_page(mapping
, hash_x_index(hash
, hash64
),
413 ll_dir_filler
, &lhash
);
415 CERROR("read cache page: "DFID
" at %llu: rc %ld\n",
416 PFID(ll_inode2fid(dir
)), hash
, PTR_ERR(page
));
420 wait_on_page_locked(page
);
422 if (!PageUptodate(page
)) {
423 CERROR("page not updated: "DFID
" at %llu: rc %d\n",
424 PFID(ll_inode2fid(dir
)), hash
, -5);
427 if (!PageChecked(page
))
428 /* XXX: check page format later */
429 SetPageChecked(page
);
430 if (PageError(page
)) {
431 CERROR("page error: "DFID
" at %llu: rc %d\n",
432 PFID(ll_inode2fid(dir
)), hash
, -5);
436 dp
= page_address(page
);
437 if (BITS_PER_LONG
== 32 && hash64
) {
438 start
= le64_to_cpu(dp
->ldp_hash_start
) >> 32;
439 end
= le64_to_cpu(dp
->ldp_hash_end
) >> 32;
442 start
= le64_to_cpu(dp
->ldp_hash_start
);
443 end
= le64_to_cpu(dp
->ldp_hash_end
);
447 LASSERT(start
== lhash
);
448 CWARN("Page-wide hash collision: %llu\n", end
);
449 if (BITS_PER_LONG
== 32 && hash64
)
450 CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n",
451 le64_to_cpu(dp
->ldp_hash_start
),
452 le64_to_cpu(dp
->ldp_hash_end
), hash
);
454 * Fetch whole overflow chain...
461 mutex_unlock(&lli
->lli_readdir_mutex
);
462 ldlm_lock_decref(&lockh
, mode
);
466 ll_release_page(page
, 1);
467 page
= ERR_PTR(-EIO
);
471 int ll_dir_read(struct inode
*inode
, struct dir_context
*ctx
)
473 struct ll_inode_info
*info
= ll_i2info(inode
);
474 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
475 __u64 pos
= ctx
->pos
;
476 int api32
= ll_need_32bit_api(sbi
);
477 int hash64
= sbi
->ll_flags
& LL_SBI_64BIT_HASH
;
479 struct ll_dir_chain chain
;
483 ll_dir_chain_init(&chain
);
485 page
= ll_get_dir_page(inode
, pos
, &chain
);
487 while (rc
== 0 && !done
) {
488 struct lu_dirpage
*dp
;
489 struct lu_dirent
*ent
;
493 * If page is empty (end of directory is reached),
496 __u64 hash
= MDS_DIR_END_OFF
;
499 dp
= page_address(page
);
500 for (ent
= lu_dirent_start(dp
); ent
&& !done
;
501 ent
= lu_dirent_next(ent
)) {
509 * XXX: implement correct swabbing here.
512 hash
= le64_to_cpu(ent
->lde_hash
);
515 * Skip until we find target hash
520 namelen
= le16_to_cpu(ent
->lde_namelen
);
531 fid_le_to_cpu(&fid
, &ent
->lde_fid
);
532 ino
= cl_fid_build_ino(&fid
, api32
);
533 type
= ll_dirent_type_get(ent
);
535 /* For 'll_nfs_get_name_filldir()', it will try
536 * to access the 'ent' through its 'lde_name',
537 * so the parameter 'name' for 'ctx->actor()'
538 * must be part of the 'ent'.
540 done
= !dir_emit(ctx
, ent
->lde_name
,
543 next
= le64_to_cpu(dp
->ldp_hash_end
);
546 if (pos
== MDS_DIR_END_OFF
) {
548 * End of directory reached.
551 ll_release_page(page
, 0);
552 } else if (1 /* chain is exhausted*/) {
554 * Normal case: continue to the next
557 ll_release_page(page
,
558 le32_to_cpu(dp
->ldp_flags
) &
561 page
= ll_get_dir_page(inode
, pos
,
565 * go into overflow page.
567 LASSERT(le32_to_cpu(dp
->ldp_flags
) &
569 ll_release_page(page
, 1);
573 ll_release_page(page
, 0);
577 CERROR("error reading dir "DFID
" at %lu: rc %d\n",
578 PFID(&info
->lli_fid
), (unsigned long)pos
, rc
);
583 ll_dir_chain_fini(&chain
);
587 static int ll_readdir(struct file
*filp
, struct dir_context
*ctx
)
589 struct inode
*inode
= file_inode(filp
);
590 struct ll_file_data
*lfd
= LUSTRE_FPRIVATE(filp
);
591 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
592 int hash64
= sbi
->ll_flags
& LL_SBI_64BIT_HASH
;
593 int api32
= ll_need_32bit_api(sbi
);
596 CDEBUG(D_VFSTRACE
, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
597 inode
->i_ino
, inode
->i_generation
,
598 inode
, (unsigned long)lfd
->lfd_pos
, i_size_read(inode
), api32
);
600 if (lfd
->lfd_pos
== MDS_DIR_END_OFF
) {
608 ctx
->pos
= lfd
->lfd_pos
;
609 rc
= ll_dir_read(inode
, ctx
);
610 lfd
->lfd_pos
= ctx
->pos
;
611 if (ctx
->pos
== MDS_DIR_END_OFF
) {
613 ctx
->pos
= LL_DIR_END_OFF_32BIT
;
615 ctx
->pos
= LL_DIR_END_OFF
;
620 filp
->f_version
= inode
->i_version
;
624 ll_stats_ops_tally(sbi
, LPROC_LL_READDIR
, 1);
629 static int ll_send_mgc_param(struct obd_export
*mgc
, char *string
)
631 struct mgs_send_param
*msp
;
634 msp
= kzalloc(sizeof(*msp
), GFP_NOFS
);
638 strlcpy(msp
->mgs_param
, string
, sizeof(msp
->mgs_param
));
639 rc
= obd_set_info_async(NULL
, mgc
, sizeof(KEY_SET_INFO
), KEY_SET_INFO
,
640 sizeof(struct mgs_send_param
), msp
, NULL
);
642 CERROR("Failed to set parameter: %d\n", rc
);
648 static int ll_dir_setdirstripe(struct inode
*dir
, struct lmv_user_md
*lump
,
651 struct ptlrpc_request
*request
= NULL
;
652 struct md_op_data
*op_data
;
653 struct ll_sb_info
*sbi
= ll_i2sbi(dir
);
657 mode
= (~current_umask() & 0755) | S_IFDIR
;
658 op_data
= ll_prep_md_op_data(NULL
, dir
, NULL
, filename
,
659 strlen(filename
), mode
, LUSTRE_OPC_MKDIR
,
661 if (IS_ERR(op_data
)) {
662 err
= PTR_ERR(op_data
);
666 op_data
->op_cli_flags
|= CLI_SET_MEA
;
667 err
= md_create(sbi
->ll_md_exp
, op_data
, lump
, sizeof(*lump
), mode
,
668 from_kuid(&init_user_ns
, current_fsuid()),
669 from_kgid(&init_user_ns
, current_fsgid()),
670 cfs_curproc_cap_pack(), 0, &request
);
671 ll_finish_md_op_data(op_data
);
675 ptlrpc_req_finished(request
);
679 int ll_dir_setstripe(struct inode
*inode
, struct lov_user_md
*lump
,
682 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
683 struct md_op_data
*op_data
;
684 struct ptlrpc_request
*req
= NULL
;
686 struct lustre_sb_info
*lsi
= s2lsi(inode
->i_sb
);
687 struct obd_device
*mgc
= lsi
->lsi_mgc
;
692 * This is coming from userspace, so should be in
693 * local endian. But the MDS would like it in little
694 * endian, so we swab it before we send it.
696 switch (lump
->lmm_magic
) {
697 case LOV_USER_MAGIC_V1
: {
698 if (lump
->lmm_magic
!= cpu_to_le32(LOV_USER_MAGIC_V1
))
699 lustre_swab_lov_user_md_v1(lump
);
700 lum_size
= sizeof(struct lov_user_md_v1
);
703 case LOV_USER_MAGIC_V3
: {
704 if (lump
->lmm_magic
!= cpu_to_le32(LOV_USER_MAGIC_V3
))
705 lustre_swab_lov_user_md_v3(
706 (struct lov_user_md_v3
*)lump
);
707 lum_size
= sizeof(struct lov_user_md_v3
);
711 CDEBUG(D_IOCTL
, "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
712 lump
->lmm_magic
, LOV_USER_MAGIC_V1
,
718 lum_size
= sizeof(struct lov_user_md_v1
);
721 op_data
= ll_prep_md_op_data(NULL
, inode
, NULL
, NULL
, 0, 0,
722 LUSTRE_OPC_ANY
, NULL
);
724 return PTR_ERR(op_data
);
726 if (lump
&& lump
->lmm_magic
== cpu_to_le32(LMV_USER_MAGIC
))
727 op_data
->op_cli_flags
|= CLI_SET_MEA
;
729 /* swabbing is done in lov_setstripe() on server side */
730 rc
= md_setattr(sbi
->ll_md_exp
, op_data
, lump
, lum_size
,
731 NULL
, 0, &req
, NULL
);
732 ll_finish_md_op_data(op_data
);
733 ptlrpc_req_finished(req
);
735 if (rc
!= -EPERM
&& rc
!= -EACCES
)
736 CERROR("mdc_setattr fails: rc = %d\n", rc
);
739 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
740 * LOV_USER_MAGIC_V3 have the same initial fields so we do not
741 * need to make the distinction between the 2 versions
743 if (set_default
&& mgc
->u
.cli
.cl_mgc_mgsexp
) {
747 param
= kzalloc(MGS_PARAM_MAXLEN
, GFP_NOFS
);
752 /* Get fsname and assume devname to be -MDT0000. */
753 ll_get_fsname(inode
->i_sb
, buf
, MTI_NAME_MAXLEN
);
754 strcat(buf
, "-MDT0000.lov");
757 /* Set root stripesize */
758 sprintf(buf
, ".stripesize=%u",
759 lump
? le32_to_cpu(lump
->lmm_stripe_size
) : 0);
760 rc
= ll_send_mgc_param(mgc
->u
.cli
.cl_mgc_mgsexp
, param
);
764 /* Set root stripecount */
765 sprintf(buf
, ".stripecount=%hd",
766 lump
? le16_to_cpu(lump
->lmm_stripe_count
) : 0);
767 rc
= ll_send_mgc_param(mgc
->u
.cli
.cl_mgc_mgsexp
, param
);
771 /* Set root stripeoffset */
772 sprintf(buf
, ".stripeoffset=%hd",
773 lump
? le16_to_cpu(lump
->lmm_stripe_offset
) :
774 (typeof(lump
->lmm_stripe_offset
))(-1));
775 rc
= ll_send_mgc_param(mgc
->u
.cli
.cl_mgc_mgsexp
, param
);
783 int ll_dir_getstripe(struct inode
*inode
, struct lov_mds_md
**lmmp
,
784 int *lmm_size
, struct ptlrpc_request
**request
)
786 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
787 struct mdt_body
*body
;
788 struct lov_mds_md
*lmm
= NULL
;
789 struct ptlrpc_request
*req
= NULL
;
791 struct md_op_data
*op_data
;
793 rc
= ll_get_default_mdsize(sbi
, &lmmsize
);
797 op_data
= ll_prep_md_op_data(NULL
, inode
, NULL
, NULL
,
798 0, lmmsize
, LUSTRE_OPC_ANY
,
801 return PTR_ERR(op_data
);
803 op_data
->op_valid
= OBD_MD_FLEASIZE
| OBD_MD_FLDIREA
;
804 rc
= md_getattr(sbi
->ll_md_exp
, op_data
, &req
);
805 ll_finish_md_op_data(op_data
);
807 CDEBUG(D_INFO
, "md_getattr failed on inode %lu/%u: rc %d\n",
809 inode
->i_generation
, rc
);
813 body
= req_capsule_server_get(&req
->rq_pill
, &RMF_MDT_BODY
);
815 lmmsize
= body
->eadatasize
;
817 if (!(body
->valid
& (OBD_MD_FLEASIZE
| OBD_MD_FLDIREA
)) ||
823 lmm
= req_capsule_server_sized_get(&req
->rq_pill
,
824 &RMF_MDT_MD
, lmmsize
);
827 * This is coming from the MDS, so is probably in
828 * little endian. We convert it to host endian before
829 * passing it to userspace.
831 /* We don't swab objects for directories */
832 switch (le32_to_cpu(lmm
->lmm_magic
)) {
834 if (cpu_to_le32(LOV_MAGIC
) != LOV_MAGIC
)
835 lustre_swab_lov_user_md_v1((struct lov_user_md_v1
*)lmm
);
838 if (cpu_to_le32(LOV_MAGIC
) != LOV_MAGIC
)
839 lustre_swab_lov_user_md_v3((struct lov_user_md_v3
*)lmm
);
842 CERROR("unknown magic: %lX\n", (unsigned long)lmm
->lmm_magic
);
853 * Get MDT index for the inode.
855 int ll_get_mdt_idx(struct inode
*inode
)
857 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
858 struct md_op_data
*op_data
;
861 op_data
= ll_prep_md_op_data(NULL
, inode
, NULL
, NULL
, 0,
862 0, LUSTRE_OPC_ANY
, NULL
);
864 return PTR_ERR(op_data
);
866 op_data
->op_flags
|= MF_GET_MDT_IDX
;
867 rc
= md_getattr(sbi
->ll_md_exp
, op_data
, NULL
);
868 mdtidx
= op_data
->op_mds
;
869 ll_finish_md_op_data(op_data
);
871 CDEBUG(D_INFO
, "md_getattr_name: %d\n", rc
);
878 * Generic handler to do any pre-copy work.
880 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
881 * first information for it that real work has started.
883 * Moreover, for a ARCHIVE request, it will sample the file data version and
884 * store it in \a copy.
886 * \return 0 on success.
888 static int ll_ioc_copy_start(struct super_block
*sb
, struct hsm_copy
*copy
)
890 struct ll_sb_info
*sbi
= ll_s2sbi(sb
);
891 struct hsm_progress_kernel hpk
;
894 /* Forge a hsm_progress based on data from copy. */
895 hpk
.hpk_fid
= copy
->hc_hai
.hai_fid
;
896 hpk
.hpk_cookie
= copy
->hc_hai
.hai_cookie
;
897 hpk
.hpk_extent
.offset
= copy
->hc_hai
.hai_extent
.offset
;
898 hpk
.hpk_extent
.length
= 0;
901 hpk
.hpk_data_version
= 0;
903 /* For archive request, we need to read the current file version. */
904 if (copy
->hc_hai
.hai_action
== HSMA_ARCHIVE
) {
906 __u64 data_version
= 0;
908 /* Get inode for this fid */
909 inode
= search_inode_for_lustre(sb
, ©
->hc_hai
.hai_fid
);
911 hpk
.hpk_flags
|= HP_FLAG_RETRY
;
912 /* hpk_errval is >= 0 */
913 hpk
.hpk_errval
= -PTR_ERR(inode
);
918 /* Read current file data version */
919 rc
= ll_data_version(inode
, &data_version
, 1);
922 CDEBUG(D_HSM
, "Could not read file data version of "
923 DFID
" (rc = %d). Archive request (%#llx) could not be done.\n",
924 PFID(©
->hc_hai
.hai_fid
), rc
,
925 copy
->hc_hai
.hai_cookie
);
926 hpk
.hpk_flags
|= HP_FLAG_RETRY
;
927 /* hpk_errval must be >= 0 */
928 hpk
.hpk_errval
= -rc
;
932 /* Store in the hsm_copy for later copytool use.
933 * Always modified even if no lsm.
935 copy
->hc_data_version
= data_version
;
939 rc
= obd_iocontrol(LL_IOC_HSM_PROGRESS
, sbi
->ll_md_exp
, sizeof(hpk
),
946 * Generic handler to do any post-copy work.
948 * It will send the last hsm_progress update to coordinator to inform it
949 * that copy is finished and whether it was successful or not.
952 * - for ARCHIVE request, it will sample the file data version and compare it
953 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
954 * will be considered as failed.
955 * - for RESTORE request, it will sample the file data version and send it to
956 * coordinator which is useful if the file was imported as 'released'.
958 * \return 0 on success.
960 static int ll_ioc_copy_end(struct super_block
*sb
, struct hsm_copy
*copy
)
962 struct ll_sb_info
*sbi
= ll_s2sbi(sb
);
963 struct hsm_progress_kernel hpk
;
966 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
967 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
968 * initialized if copy_end was called with copy == NULL.
971 /* Forge a hsm_progress based on data from copy. */
972 hpk
.hpk_fid
= copy
->hc_hai
.hai_fid
;
973 hpk
.hpk_cookie
= copy
->hc_hai
.hai_cookie
;
974 hpk
.hpk_extent
= copy
->hc_hai
.hai_extent
;
975 hpk
.hpk_flags
= copy
->hc_flags
| HP_FLAG_COMPLETED
;
976 hpk
.hpk_errval
= copy
->hc_errval
;
977 hpk
.hpk_data_version
= 0;
979 /* For archive request, we need to check the file data was not changed.
981 * For restore request, we need to send the file data version, this is
982 * useful when the file was created using hsm_import.
984 if (((copy
->hc_hai
.hai_action
== HSMA_ARCHIVE
) ||
985 (copy
->hc_hai
.hai_action
== HSMA_RESTORE
)) &&
986 (copy
->hc_errval
== 0)) {
988 __u64 data_version
= 0;
990 /* Get lsm for this fid */
991 inode
= search_inode_for_lustre(sb
, ©
->hc_hai
.hai_fid
);
993 hpk
.hpk_flags
|= HP_FLAG_RETRY
;
994 /* hpk_errval must be >= 0 */
995 hpk
.hpk_errval
= -PTR_ERR(inode
);
1000 rc
= ll_data_version(inode
, &data_version
,
1001 copy
->hc_hai
.hai_action
== HSMA_ARCHIVE
);
1004 CDEBUG(D_HSM
, "Could not read file data version. Request could not be confirmed.\n");
1005 if (hpk
.hpk_errval
== 0)
1006 hpk
.hpk_errval
= -rc
;
1010 /* Store in the hsm_copy for later copytool use.
1011 * Always modified even if no lsm.
1013 hpk
.hpk_data_version
= data_version
;
1015 /* File could have been stripped during archiving, so we need
1018 if ((copy
->hc_hai
.hai_action
== HSMA_ARCHIVE
) &&
1019 (copy
->hc_data_version
!= data_version
)) {
1020 CDEBUG(D_HSM
, "File data version mismatched. File content was changed during archiving. "
1021 DFID
", start:%#llx current:%#llx\n",
1022 PFID(©
->hc_hai
.hai_fid
),
1023 copy
->hc_data_version
, data_version
);
1024 /* File was changed, send error to cdt. Do not ask for
1025 * retry because if a file is modified frequently,
1026 * the cdt will loop on retried archive requests.
1027 * The policy engine will ask for a new archive later
1028 * when the file will not be modified for some tunable
1031 /* we do not notify caller */
1032 hpk
.hpk_flags
&= ~HP_FLAG_RETRY
;
1033 /* hpk_errval must be >= 0 */
1034 hpk
.hpk_errval
= EBUSY
;
1040 rc
= obd_iocontrol(LL_IOC_HSM_PROGRESS
, sbi
->ll_md_exp
, sizeof(hpk
),
1046 static int copy_and_ioctl(int cmd
, struct obd_export
*exp
,
1047 const void __user
*data
, size_t size
)
1052 copy
= kzalloc(size
, GFP_NOFS
);
1056 if (copy_from_user(copy
, data
, size
)) {
1061 rc
= obd_iocontrol(cmd
, exp
, size
, copy
, NULL
);
1068 static int quotactl_ioctl(struct ll_sb_info
*sbi
, struct if_quotactl
*qctl
)
1070 int cmd
= qctl
->qc_cmd
;
1071 int type
= qctl
->qc_type
;
1072 int id
= qctl
->qc_id
;
1073 int valid
= qctl
->qc_valid
;
1077 case LUSTRE_Q_INVALIDATE
:
1078 case LUSTRE_Q_FINVALIDATE
:
1083 if (!capable(CFS_CAP_SYS_ADMIN
) ||
1084 sbi
->ll_flags
& LL_SBI_RMT_CLIENT
)
1088 if (((type
== USRQUOTA
&&
1089 !uid_eq(current_euid(), make_kuid(&init_user_ns
, id
))) ||
1090 (type
== GRPQUOTA
&&
1091 !in_egroup_p(make_kgid(&init_user_ns
, id
)))) &&
1092 (!capable(CFS_CAP_SYS_ADMIN
) ||
1093 sbi
->ll_flags
& LL_SBI_RMT_CLIENT
))
1099 CERROR("unsupported quotactl op: %#x\n", cmd
);
1103 if (valid
!= QC_GENERAL
) {
1104 if (sbi
->ll_flags
& LL_SBI_RMT_CLIENT
)
1107 if (cmd
== Q_GETINFO
)
1108 qctl
->qc_cmd
= Q_GETOINFO
;
1109 else if (cmd
== Q_GETQUOTA
)
1110 qctl
->qc_cmd
= Q_GETOQUOTA
;
1116 rc
= obd_iocontrol(OBD_IOC_QUOTACTL
, sbi
->ll_md_exp
,
1117 sizeof(*qctl
), qctl
, NULL
);
1120 rc
= obd_iocontrol(OBD_IOC_QUOTACTL
, sbi
->ll_dt_exp
,
1121 sizeof(*qctl
), qctl
, NULL
);
1124 rc
= obd_iocontrol(OBD_IOC_QUOTACTL
, sbi
->ll_md_exp
,
1125 sizeof(*qctl
), qctl
, NULL
);
1127 rc
= obd_iocontrol(OBD_IOC_QUOTACTL
,
1129 sizeof(*qctl
), qctl
, NULL
);
1141 struct obd_quotactl
*oqctl
;
1143 oqctl
= kzalloc(sizeof(*oqctl
), GFP_NOFS
);
1147 QCTL_COPY(oqctl
, qctl
);
1148 rc
= obd_quotactl(sbi
->ll_md_exp
, oqctl
);
1150 if (rc
!= -EALREADY
&& cmd
== Q_QUOTAON
) {
1151 oqctl
->qc_cmd
= Q_QUOTAOFF
;
1152 obd_quotactl(sbi
->ll_md_exp
, oqctl
);
1157 /* If QIF_SPACE is not set, client should collect the
1158 * space usage from OSSs by itself
1160 if (cmd
== Q_GETQUOTA
&&
1161 !(oqctl
->qc_dqblk
.dqb_valid
& QIF_SPACE
) &&
1162 !oqctl
->qc_dqblk
.dqb_curspace
) {
1163 struct obd_quotactl
*oqctl_tmp
;
1165 oqctl_tmp
= kzalloc(sizeof(*oqctl_tmp
), GFP_NOFS
);
1171 oqctl_tmp
->qc_cmd
= Q_GETOQUOTA
;
1172 oqctl_tmp
->qc_id
= oqctl
->qc_id
;
1173 oqctl_tmp
->qc_type
= oqctl
->qc_type
;
1175 /* collect space usage from OSTs */
1176 oqctl_tmp
->qc_dqblk
.dqb_curspace
= 0;
1177 rc
= obd_quotactl(sbi
->ll_dt_exp
, oqctl_tmp
);
1178 if (!rc
|| rc
== -EREMOTEIO
) {
1179 oqctl
->qc_dqblk
.dqb_curspace
=
1180 oqctl_tmp
->qc_dqblk
.dqb_curspace
;
1181 oqctl
->qc_dqblk
.dqb_valid
|= QIF_SPACE
;
1184 /* collect space & inode usage from MDTs */
1185 oqctl_tmp
->qc_dqblk
.dqb_curspace
= 0;
1186 oqctl_tmp
->qc_dqblk
.dqb_curinodes
= 0;
1187 rc
= obd_quotactl(sbi
->ll_md_exp
, oqctl_tmp
);
1188 if (!rc
|| rc
== -EREMOTEIO
) {
1189 oqctl
->qc_dqblk
.dqb_curspace
+=
1190 oqctl_tmp
->qc_dqblk
.dqb_curspace
;
1191 oqctl
->qc_dqblk
.dqb_curinodes
=
1192 oqctl_tmp
->qc_dqblk
.dqb_curinodes
;
1193 oqctl
->qc_dqblk
.dqb_valid
|= QIF_INODES
;
1195 oqctl
->qc_dqblk
.dqb_valid
&= ~QIF_SPACE
;
1201 QCTL_COPY(qctl
, oqctl
);
1208 /* This function tries to get a single name component,
1209 * to send to the server. No actual path traversal involved,
1210 * so we limit to NAME_MAX
1212 static char *ll_getname(const char __user
*filename
)
1217 tmp
= kzalloc(NAME_MAX
+ 1, GFP_KERNEL
);
1219 return ERR_PTR(-ENOMEM
);
1221 len
= strncpy_from_user(tmp
, filename
, NAME_MAX
+ 1);
1226 else if (len
> NAME_MAX
&& tmp
[NAME_MAX
] != 0)
1227 ret
= -ENAMETOOLONG
;
1236 #define ll_putname(filename) kfree(filename)
1238 static long ll_dir_ioctl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
1240 struct inode
*inode
= file_inode(file
);
1241 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
1242 struct obd_ioctl_data
*data
;
1245 CDEBUG(D_VFSTRACE
, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1246 inode
->i_ino
, inode
->i_generation
, inode
, cmd
);
1248 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1249 if (_IOC_TYPE(cmd
) == 'T' || _IOC_TYPE(cmd
) == 't') /* tty ioctls */
1252 ll_stats_ops_tally(ll_i2sbi(inode
), LPROC_LL_IOCTL
, 1);
1254 case FSFILT_IOC_GETFLAGS
:
1255 case FSFILT_IOC_SETFLAGS
:
1256 return ll_iocontrol(inode
, file
, cmd
, arg
);
1257 case FSFILT_IOC_GETVERSION_OLD
:
1258 case FSFILT_IOC_GETVERSION
:
1259 return put_user(inode
->i_generation
, (int __user
*)arg
);
1260 /* We need to special case any other ioctls we want to handle,
1261 * to send them to the MDS/OST as appropriate and to properly
1262 * network encode the arg field.
1263 case FSFILT_IOC_SETVERSION_OLD:
1264 case FSFILT_IOC_SETVERSION:
1266 case LL_IOC_GET_MDTIDX
: {
1269 mdtidx
= ll_get_mdt_idx(inode
);
1273 if (put_user((int)mdtidx
, (int __user
*)arg
))
1278 case IOC_MDC_LOOKUP
: {
1279 struct ptlrpc_request
*request
= NULL
;
1280 int namelen
, len
= 0;
1283 struct md_op_data
*op_data
;
1285 rc
= obd_ioctl_getdata(&buf
, &len
, (void __user
*)arg
);
1290 filename
= data
->ioc_inlbuf1
;
1291 namelen
= strlen(filename
);
1294 CDEBUG(D_INFO
, "IOC_MDC_LOOKUP missing filename\n");
1299 op_data
= ll_prep_md_op_data(NULL
, inode
, NULL
, filename
, namelen
,
1300 0, LUSTRE_OPC_ANY
, NULL
);
1301 if (IS_ERR(op_data
)) {
1302 rc
= PTR_ERR(op_data
);
1306 op_data
->op_valid
= OBD_MD_FLID
;
1307 rc
= md_getattr_name(sbi
->ll_md_exp
, op_data
, &request
);
1308 ll_finish_md_op_data(op_data
);
1310 CDEBUG(D_INFO
, "md_getattr_name: %d\n", rc
);
1313 ptlrpc_req_finished(request
);
1315 obd_ioctl_freedata(buf
, len
);
1318 case LL_IOC_LMV_SETSTRIPE
: {
1319 struct lmv_user_md
*lum
;
1327 rc
= obd_ioctl_getdata(&buf
, &len
, (void __user
*)arg
);
1332 if (!data
->ioc_inlbuf1
|| !data
->ioc_inlbuf2
||
1333 data
->ioc_inllen1
== 0 || data
->ioc_inllen2
== 0) {
1338 filename
= data
->ioc_inlbuf1
;
1339 namelen
= data
->ioc_inllen1
;
1342 CDEBUG(D_INFO
, "IOC_MDC_LOOKUP missing filename\n");
1346 lum
= (struct lmv_user_md
*)data
->ioc_inlbuf2
;
1347 lumlen
= data
->ioc_inllen2
;
1349 if (lum
->lum_magic
!= LMV_USER_MAGIC
||
1350 lumlen
!= sizeof(*lum
)) {
1351 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1352 filename
, lum
->lum_magic
, lumlen
, -EFAULT
);
1358 * ll_dir_setdirstripe will be used to set dir stripe
1359 * mdc_create--->mdt_reint_create (with dirstripe)
1361 rc
= ll_dir_setdirstripe(inode
, lum
, filename
);
1363 obd_ioctl_freedata(buf
, len
);
1367 case LL_IOC_LOV_SETSTRIPE
: {
1368 struct lov_user_md_v3 lumv3
;
1369 struct lov_user_md_v1
*lumv1
= (struct lov_user_md_v1
*)&lumv3
;
1370 struct lov_user_md_v1 __user
*lumv1p
= (void __user
*)arg
;
1371 struct lov_user_md_v3 __user
*lumv3p
= (void __user
*)arg
;
1373 int set_default
= 0;
1375 LASSERT(sizeof(lumv3
) == sizeof(*lumv3p
));
1376 LASSERT(sizeof(lumv3
.lmm_objects
[0]) ==
1377 sizeof(lumv3p
->lmm_objects
[0]));
1378 /* first try with v1 which is smaller than v3 */
1379 if (copy_from_user(lumv1
, lumv1p
, sizeof(*lumv1
)))
1382 if (lumv1
->lmm_magic
== LOV_USER_MAGIC_V3
) {
1383 if (copy_from_user(&lumv3
, lumv3p
, sizeof(lumv3
)))
1387 if (is_root_inode(inode
))
1390 /* in v1 and v3 cases lumv1 points to data */
1391 rc
= ll_dir_setstripe(inode
, lumv1
, set_default
);
1395 case LL_IOC_LMV_GETSTRIPE
: {
1396 struct lmv_user_md __user
*lump
= (void __user
*)arg
;
1397 struct lmv_user_md lum
;
1398 struct lmv_user_md
*tmp
;
1403 if (copy_from_user(&lum
, lump
, sizeof(struct lmv_user_md
)))
1406 if (lum
.lum_magic
!= LMV_MAGIC_V1
)
1409 lum_size
= lmv_user_md_size(1, LMV_MAGIC_V1
);
1410 tmp
= kzalloc(lum_size
, GFP_NOFS
);
1417 tmp
->lum_type
= LMV_STRIPE_TYPE
;
1418 tmp
->lum_stripe_count
= 1;
1419 mdtindex
= ll_get_mdt_idx(inode
);
1425 tmp
->lum_stripe_offset
= mdtindex
;
1426 tmp
->lum_objects
[0].lum_mds
= mdtindex
;
1427 memcpy(&tmp
->lum_objects
[0].lum_fid
, ll_inode2fid(inode
),
1428 sizeof(struct lu_fid
));
1429 if (copy_to_user((void __user
*)arg
, tmp
, lum_size
)) {
1437 case LL_IOC_LOV_SWAP_LAYOUTS
:
1439 case LL_IOC_OBD_STATFS
:
1440 return ll_obd_statfs(inode
, (void __user
*)arg
);
1441 case LL_IOC_LOV_GETSTRIPE
:
1442 case LL_IOC_MDC_GETINFO
:
1443 case IOC_MDC_GETFILEINFO
:
1444 case IOC_MDC_GETFILESTRIPE
: {
1445 struct ptlrpc_request
*request
= NULL
;
1446 struct lov_user_md __user
*lump
;
1447 struct lov_mds_md
*lmm
= NULL
;
1448 struct mdt_body
*body
;
1449 char *filename
= NULL
;
1452 if (cmd
== IOC_MDC_GETFILEINFO
||
1453 cmd
== IOC_MDC_GETFILESTRIPE
) {
1454 filename
= ll_getname((const char __user
*)arg
);
1455 if (IS_ERR(filename
))
1456 return PTR_ERR(filename
);
1458 rc
= ll_lov_getstripe_ea_info(inode
, filename
, &lmm
,
1459 &lmmsize
, &request
);
1461 rc
= ll_dir_getstripe(inode
, &lmm
, &lmmsize
, &request
);
1465 body
= req_capsule_server_get(&request
->rq_pill
,
1473 if (rc
== -ENODATA
&& (cmd
== IOC_MDC_GETFILEINFO
||
1474 cmd
== LL_IOC_MDC_GETINFO
)) {
1481 if (cmd
== IOC_MDC_GETFILESTRIPE
||
1482 cmd
== LL_IOC_LOV_GETSTRIPE
) {
1483 lump
= (struct lov_user_md __user
*)arg
;
1485 struct lov_user_mds_data __user
*lmdp
;
1487 lmdp
= (struct lov_user_mds_data __user
*)arg
;
1488 lump
= &lmdp
->lmd_lmm
;
1490 if (copy_to_user(lump
, lmm
, lmmsize
)) {
1491 if (copy_to_user(lump
, lmm
, sizeof(*lump
))) {
1498 if (cmd
== IOC_MDC_GETFILEINFO
|| cmd
== LL_IOC_MDC_GETINFO
) {
1499 struct lov_user_mds_data __user
*lmdp
;
1502 st
.st_dev
= inode
->i_sb
->s_dev
;
1503 st
.st_mode
= body
->mode
;
1504 st
.st_nlink
= body
->nlink
;
1505 st
.st_uid
= body
->uid
;
1506 st
.st_gid
= body
->gid
;
1507 st
.st_rdev
= body
->rdev
;
1508 st
.st_size
= body
->size
;
1509 st
.st_blksize
= PAGE_SIZE
;
1510 st
.st_blocks
= body
->blocks
;
1511 st
.st_atime
= body
->atime
;
1512 st
.st_mtime
= body
->mtime
;
1513 st
.st_ctime
= body
->ctime
;
1514 st
.st_ino
= inode
->i_ino
;
1516 lmdp
= (struct lov_user_mds_data __user
*)arg
;
1517 if (copy_to_user(&lmdp
->lmd_st
, &st
, sizeof(st
))) {
1524 ptlrpc_req_finished(request
);
1526 ll_putname(filename
);
1529 case IOC_LOV_GETINFO
: {
1530 struct lov_user_mds_data __user
*lumd
;
1531 struct lov_stripe_md
*lsm
;
1532 struct lov_user_md __user
*lum
;
1533 struct lov_mds_md
*lmm
;
1537 lumd
= (struct lov_user_mds_data __user
*)arg
;
1538 lum
= &lumd
->lmd_lmm
;
1540 rc
= ll_get_max_mdsize(sbi
, &lmmsize
);
1544 lmm
= libcfs_kvzalloc(lmmsize
, GFP_NOFS
);
1547 if (copy_from_user(lmm
, lum
, lmmsize
)) {
1552 switch (lmm
->lmm_magic
) {
1553 case LOV_USER_MAGIC_V1
:
1554 if (cpu_to_le32(LOV_USER_MAGIC_V1
) == LOV_USER_MAGIC_V1
)
1556 /* swab objects first so that stripes num will be sane */
1557 lustre_swab_lov_user_md_objects(
1558 ((struct lov_user_md_v1
*)lmm
)->lmm_objects
,
1559 ((struct lov_user_md_v1
*)lmm
)->lmm_stripe_count
);
1560 lustre_swab_lov_user_md_v1((struct lov_user_md_v1
*)lmm
);
1562 case LOV_USER_MAGIC_V3
:
1563 if (cpu_to_le32(LOV_USER_MAGIC_V3
) == LOV_USER_MAGIC_V3
)
1565 /* swab objects first so that stripes num will be sane */
1566 lustre_swab_lov_user_md_objects(
1567 ((struct lov_user_md_v3
*)lmm
)->lmm_objects
,
1568 ((struct lov_user_md_v3
*)lmm
)->lmm_stripe_count
);
1569 lustre_swab_lov_user_md_v3((struct lov_user_md_v3
*)lmm
);
1576 rc
= obd_unpackmd(sbi
->ll_dt_exp
, &lsm
, lmm
, lmmsize
);
1582 /* Perform glimpse_size operation. */
1583 memset(&st
, 0, sizeof(st
));
1585 rc
= ll_glimpse_ioctl(sbi
, lsm
, &st
);
1589 if (copy_to_user(&lumd
->lmd_st
, &st
, sizeof(st
))) {
1595 obd_free_memmd(sbi
->ll_dt_exp
, &lsm
);
1600 case OBD_IOC_LLOG_CATINFO
: {
1603 case OBD_IOC_QUOTACHECK
: {
1604 struct obd_quotactl
*oqctl
;
1607 if (!capable(CFS_CAP_SYS_ADMIN
) ||
1608 sbi
->ll_flags
& LL_SBI_RMT_CLIENT
)
1611 oqctl
= kzalloc(sizeof(*oqctl
), GFP_NOFS
);
1614 oqctl
->qc_type
= arg
;
1615 rc
= obd_quotacheck(sbi
->ll_md_exp
, oqctl
);
1617 CDEBUG(D_INFO
, "md_quotacheck failed: rc %d\n", rc
);
1621 rc
= obd_quotacheck(sbi
->ll_dt_exp
, oqctl
);
1623 CDEBUG(D_INFO
, "obd_quotacheck failed: rc %d\n", rc
);
1628 case OBD_IOC_POLL_QUOTACHECK
: {
1629 struct if_quotacheck
*check
;
1631 if (!capable(CFS_CAP_SYS_ADMIN
) ||
1632 sbi
->ll_flags
& LL_SBI_RMT_CLIENT
)
1635 check
= kzalloc(sizeof(*check
), GFP_NOFS
);
1639 rc
= obd_iocontrol(cmd
, sbi
->ll_md_exp
, 0, (void *)check
,
1642 CDEBUG(D_QUOTA
, "mdc ioctl %d failed: %d\n", cmd
, rc
);
1643 if (copy_to_user((void __user
*)arg
, check
,
1645 CDEBUG(D_QUOTA
, "copy_to_user failed\n");
1649 rc
= obd_iocontrol(cmd
, sbi
->ll_dt_exp
, 0, (void *)check
,
1652 CDEBUG(D_QUOTA
, "osc ioctl %d failed: %d\n", cmd
, rc
);
1653 if (copy_to_user((void __user
*)arg
, check
,
1655 CDEBUG(D_QUOTA
, "copy_to_user failed\n");
1662 case LL_IOC_QUOTACTL
: {
1663 struct if_quotactl
*qctl
;
1665 qctl
= kzalloc(sizeof(*qctl
), GFP_NOFS
);
1669 if (copy_from_user(qctl
, (void __user
*)arg
, sizeof(*qctl
))) {
1674 rc
= quotactl_ioctl(sbi
, qctl
);
1676 if (rc
== 0 && copy_to_user((void __user
*)arg
, qctl
,
1684 case OBD_IOC_GETDTNAME
:
1685 case OBD_IOC_GETMDNAME
:
1686 return ll_get_obd_name(inode
, cmd
, arg
);
1687 case LL_IOC_FLUSHCTX
:
1688 return ll_flush_ctx(inode
);
1689 #ifdef CONFIG_FS_POSIX_ACL
1690 case LL_IOC_RMTACL
: {
1691 if (sbi
->ll_flags
& LL_SBI_RMT_CLIENT
&& is_root_inode(inode
)) {
1692 struct ll_file_data
*fd
= LUSTRE_FPRIVATE(file
);
1694 rc
= rct_add(&sbi
->ll_rct
, current_pid(), arg
);
1696 fd
->fd_flags
|= LL_FILE_RMTACL
;
1702 case LL_IOC_GETOBDCOUNT
: {
1704 struct obd_export
*exp
;
1706 if (copy_from_user(&count
, (int __user
*)arg
, sizeof(int)))
1709 /* get ost count when count is zero, get mdt count otherwise */
1710 exp
= count
? sbi
->ll_md_exp
: sbi
->ll_dt_exp
;
1711 vallen
= sizeof(count
);
1712 rc
= obd_get_info(NULL
, exp
, sizeof(KEY_TGT_COUNT
),
1713 KEY_TGT_COUNT
, &vallen
, &count
, NULL
);
1715 CERROR("get target count failed: %d\n", rc
);
1719 if (copy_to_user((int __user
*)arg
, &count
, sizeof(int)))
1724 case LL_IOC_PATH2FID
:
1725 if (copy_to_user((void __user
*)arg
, ll_inode2fid(inode
),
1726 sizeof(struct lu_fid
)))
1729 case LL_IOC_GET_CONNECT_FLAGS
: {
1730 return obd_iocontrol(cmd
, sbi
->ll_md_exp
, 0, NULL
,
1731 (void __user
*)arg
);
1733 case OBD_IOC_CHANGELOG_SEND
:
1734 case OBD_IOC_CHANGELOG_CLEAR
:
1735 if (!capable(CFS_CAP_SYS_ADMIN
))
1738 rc
= copy_and_ioctl(cmd
, sbi
->ll_md_exp
, (void __user
*)arg
,
1739 sizeof(struct ioc_changelog
));
1741 case OBD_IOC_FID2PATH
:
1742 return ll_fid2path(inode
, (void __user
*)arg
);
1743 case LL_IOC_HSM_REQUEST
: {
1744 struct hsm_user_request
*hur
;
1747 hur
= memdup_user((void __user
*)arg
, sizeof(*hur
));
1749 return PTR_ERR(hur
);
1751 /* Compute the whole struct size */
1752 totalsize
= hur_len(hur
);
1757 /* Final size will be more than double totalsize */
1758 if (totalsize
>= MDS_MAXREQSIZE
/ 3)
1761 hur
= libcfs_kvzalloc(totalsize
, GFP_NOFS
);
1765 /* Copy the whole struct */
1766 if (copy_from_user(hur
, (void __user
*)arg
, totalsize
)) {
1771 if (hur
->hur_request
.hr_action
== HUA_RELEASE
) {
1772 const struct lu_fid
*fid
;
1776 for (i
= 0; i
< hur
->hur_request
.hr_itemcount
; i
++) {
1777 fid
= &hur
->hur_user_item
[i
].hui_fid
;
1778 f
= search_inode_for_lustre(inode
->i_sb
, fid
);
1784 rc
= ll_hsm_release(f
);
1790 rc
= obd_iocontrol(cmd
, ll_i2mdexp(inode
), totalsize
,
1798 case LL_IOC_HSM_PROGRESS
: {
1799 struct hsm_progress_kernel hpk
;
1800 struct hsm_progress hp
;
1802 if (copy_from_user(&hp
, (void __user
*)arg
, sizeof(hp
)))
1805 hpk
.hpk_fid
= hp
.hp_fid
;
1806 hpk
.hpk_cookie
= hp
.hp_cookie
;
1807 hpk
.hpk_extent
= hp
.hp_extent
;
1808 hpk
.hpk_flags
= hp
.hp_flags
;
1809 hpk
.hpk_errval
= hp
.hp_errval
;
1810 hpk
.hpk_data_version
= 0;
1812 /* File may not exist in Lustre; all progress
1813 * reported to Lustre root
1815 rc
= obd_iocontrol(cmd
, sbi
->ll_md_exp
, sizeof(hpk
), &hpk
,
1819 case LL_IOC_HSM_CT_START
:
1820 rc
= copy_and_ioctl(cmd
, sbi
->ll_md_exp
, (void __user
*)arg
,
1821 sizeof(struct lustre_kernelcomm
));
1824 case LL_IOC_HSM_COPY_START
: {
1825 struct hsm_copy
*copy
;
1828 copy
= memdup_user((char __user
*)arg
, sizeof(*copy
));
1830 return PTR_ERR(copy
);
1832 rc
= ll_ioc_copy_start(inode
->i_sb
, copy
);
1833 if (copy_to_user((char __user
*)arg
, copy
, sizeof(*copy
)))
1839 case LL_IOC_HSM_COPY_END
: {
1840 struct hsm_copy
*copy
;
1843 copy
= memdup_user((char __user
*)arg
, sizeof(*copy
));
1845 return PTR_ERR(copy
);
1847 rc
= ll_ioc_copy_end(inode
->i_sb
, copy
);
1848 if (copy_to_user((char __user
*)arg
, copy
, sizeof(*copy
)))
1855 return obd_iocontrol(cmd
, sbi
->ll_dt_exp
, 0, NULL
,
1856 (void __user
*)arg
);
1860 static loff_t
ll_dir_seek(struct file
*file
, loff_t offset
, int origin
)
1862 struct inode
*inode
= file
->f_mapping
->host
;
1863 struct ll_file_data
*fd
= LUSTRE_FPRIVATE(file
);
1864 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
1865 int api32
= ll_need_32bit_api(sbi
);
1866 loff_t ret
= -EINVAL
;
1873 offset
+= file
->f_pos
;
1879 offset
+= LL_DIR_END_OFF_32BIT
;
1881 offset
+= LL_DIR_END_OFF
;
1888 ((api32
&& offset
<= LL_DIR_END_OFF_32BIT
) ||
1889 (!api32
&& offset
<= LL_DIR_END_OFF
))) {
1890 if (offset
!= file
->f_pos
) {
1891 if ((api32
&& offset
== LL_DIR_END_OFF_32BIT
) ||
1892 (!api32
&& offset
== LL_DIR_END_OFF
))
1893 fd
->lfd_pos
= MDS_DIR_END_OFF
;
1894 else if (api32
&& sbi
->ll_flags
& LL_SBI_64BIT_HASH
)
1895 fd
->lfd_pos
= offset
<< 32;
1897 fd
->lfd_pos
= offset
;
1898 file
->f_pos
= offset
;
1899 file
->f_version
= 0;
1906 inode_unlock(inode
);
1910 static int ll_dir_open(struct inode
*inode
, struct file
*file
)
1912 return ll_file_open(inode
, file
);
1915 static int ll_dir_release(struct inode
*inode
, struct file
*file
)
1917 return ll_file_release(inode
, file
);
1920 const struct file_operations ll_dir_operations
= {
1921 .llseek
= ll_dir_seek
,
1922 .open
= ll_dir_open
,
1923 .release
= ll_dir_release
,
1924 .read
= generic_read_dir
,
1925 .iterate
= ll_readdir
,
1926 .unlocked_ioctl
= ll_dir_ioctl
,