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_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
138 * lu_dirpage for this integrated page will be adjusted. See
139 * lmv_adjust_dirpages().
143 /* returns the page unlocked, but with a reference */
144 static int ll_dir_filler(void *_hash
, struct page
*page0
)
146 struct inode
*inode
= page0
->mapping
->host
;
147 int hash64
= ll_i2sbi(inode
)->ll_flags
& LL_SBI_64BIT_HASH
;
148 struct obd_export
*exp
= ll_i2sbi(inode
)->ll_md_exp
;
149 struct ptlrpc_request
*request
;
150 struct mdt_body
*body
;
151 struct md_op_data
*op_data
;
152 __u64 hash
= *((__u64
*)_hash
);
153 struct page
**page_pool
;
155 struct lu_dirpage
*dp
;
156 int max_pages
= ll_i2sbi(inode
)->ll_md_brw_size
>> PAGE_CACHE_SHIFT
;
157 int nrdpgs
= 0; /* number of pages read actually */
162 CDEBUG(D_VFSTRACE
, "VFS Op:inode=%lu/%u(%p) hash %llu\n",
163 inode
->i_ino
, inode
->i_generation
, inode
, hash
);
165 LASSERT(max_pages
> 0 && max_pages
<= MD_MAX_BRW_PAGES
);
167 page_pool
= kcalloc(max_pages
, sizeof(page
), GFP_NOFS
);
169 page_pool
[0] = page0
;
174 for (npages
= 1; npages
< max_pages
; npages
++) {
175 page
= page_cache_alloc_cold(inode
->i_mapping
);
178 page_pool
[npages
] = page
;
181 op_data
= ll_prep_md_op_data(NULL
, inode
, NULL
, NULL
, 0, 0,
182 LUSTRE_OPC_ANY
, NULL
);
183 op_data
->op_npages
= npages
;
184 op_data
->op_offset
= hash
;
185 rc
= md_readpage(exp
, op_data
, page_pool
, &request
);
186 ll_finish_md_op_data(op_data
);
188 /* page0 is special, which was added into page cache early */
189 delete_from_page_cache(page0
);
190 } else if (rc
== 0) {
191 body
= req_capsule_server_get(&request
->rq_pill
, &RMF_MDT_BODY
);
192 /* Checked by mdc_readpage() */
193 if (body
->valid
& OBD_MD_FLSIZE
)
194 cl_isize_write(inode
, body
->size
);
196 nrdpgs
= (request
->rq_bulk
->bd_nob_transferred
+PAGE_CACHE_SIZE
-1)
198 SetPageUptodate(page0
);
201 ptlrpc_req_finished(request
);
203 CDEBUG(D_VFSTRACE
, "read %d/%d pages\n", nrdpgs
, npages
);
205 for (i
= 1; i
< npages
; i
++) {
206 unsigned long offset
;
211 if (rc
< 0 || i
>= nrdpgs
) {
212 page_cache_release(page
);
216 SetPageUptodate(page
);
219 hash
= le64_to_cpu(dp
->ldp_hash_start
);
222 offset
= hash_x_index(hash
, hash64
);
224 prefetchw(&page
->flags
);
225 ret
= add_to_page_cache_lru(page
, inode
->i_mapping
, offset
,
230 CDEBUG(D_VFSTRACE
, "page %lu add to page cache failed: %d\n",
233 page_cache_release(page
);
236 if (page_pool
!= &page0
)
241 void ll_release_page(struct page
*page
, int remove
)
246 if (likely(page
->mapping
))
247 truncate_complete_page(page
->mapping
, page
);
250 page_cache_release(page
);
254 * Find, kmap and return page that contains given hash.
256 static struct page
*ll_dir_page_locate(struct inode
*dir
, __u64
*hash
,
257 __u64
*start
, __u64
*end
)
259 int hash64
= ll_i2sbi(dir
)->ll_flags
& LL_SBI_64BIT_HASH
;
260 struct address_space
*mapping
= dir
->i_mapping
;
262 * Complement of hash is used as an index so that
263 * radix_tree_gang_lookup() can be used to find a page with starting
264 * hash _smaller_ than one we are looking for.
266 unsigned long offset
= hash_x_index(*hash
, hash64
);
270 spin_lock_irq(&mapping
->tree_lock
);
271 found
= radix_tree_gang_lookup(&mapping
->page_tree
,
272 (void **)&page
, offset
, 1);
273 if (found
> 0 && !radix_tree_exceptional_entry(page
)) {
274 struct lu_dirpage
*dp
;
276 page_cache_get(page
);
277 spin_unlock_irq(&mapping
->tree_lock
);
279 * In contrast to find_lock_page() we are sure that directory
280 * page cannot be truncated (while DLM lock is held) and,
281 * hence, can avoid restart.
283 * In fact, page cannot be locked here at all, because
284 * ll_dir_filler() does synchronous io.
286 wait_on_page_locked(page
);
287 if (PageUptodate(page
)) {
289 if (BITS_PER_LONG
== 32 && hash64
) {
290 *start
= le64_to_cpu(dp
->ldp_hash_start
) >> 32;
291 *end
= le64_to_cpu(dp
->ldp_hash_end
) >> 32;
294 *start
= le64_to_cpu(dp
->ldp_hash_start
);
295 *end
= le64_to_cpu(dp
->ldp_hash_end
);
297 LASSERTF(*start
<= *hash
, "start = %#llx,end = %#llx,hash = %#llx\n",
298 *start
, *end
, *hash
);
299 CDEBUG(D_VFSTRACE
, "page %lu [%llu %llu], hash %llu\n",
300 offset
, *start
, *end
, *hash
);
302 ll_release_page(page
, 0);
304 } else if (*end
!= *start
&& *hash
== *end
) {
306 * upon hash collision, remove this page,
307 * otherwise put page reference, and
308 * ll_get_dir_page() will issue RPC to fetch
311 ll_release_page(page
,
312 le32_to_cpu(dp
->ldp_flags
) & LDF_COLLIDE
);
316 page_cache_release(page
);
317 page
= ERR_PTR(-EIO
);
321 spin_unlock_irq(&mapping
->tree_lock
);
327 struct page
*ll_get_dir_page(struct inode
*dir
, __u64 hash
,
328 struct ll_dir_chain
*chain
)
330 ldlm_policy_data_t policy
= {.l_inodebits
= {MDS_INODELOCK_UPDATE
} };
331 struct address_space
*mapping
= dir
->i_mapping
;
332 struct lustre_handle lockh
;
333 struct lu_dirpage
*dp
;
340 struct ll_inode_info
*lli
= ll_i2info(dir
);
341 int hash64
= ll_i2sbi(dir
)->ll_flags
& LL_SBI_64BIT_HASH
;
344 rc
= md_lock_match(ll_i2sbi(dir
)->ll_md_exp
, LDLM_FL_BLOCK_GRANTED
,
345 ll_inode2fid(dir
), LDLM_IBITS
, &policy
, mode
, &lockh
);
347 struct ldlm_enqueue_info einfo
= {
348 .ei_type
= LDLM_IBITS
,
350 .ei_cb_bl
= ll_md_blocking_ast
,
351 .ei_cb_cp
= ldlm_completion_ast
,
353 struct lookup_intent it
= { .it_op
= IT_READDIR
};
354 struct ptlrpc_request
*request
;
355 struct md_op_data
*op_data
;
357 op_data
= ll_prep_md_op_data(NULL
, dir
, dir
, NULL
, 0, 0,
358 LUSTRE_OPC_ANY
, NULL
);
360 return (void *)op_data
;
362 rc
= md_enqueue(ll_i2sbi(dir
)->ll_md_exp
, &einfo
, &it
,
363 op_data
, &lockh
, NULL
, 0, NULL
, 0);
365 ll_finish_md_op_data(op_data
);
367 request
= (struct ptlrpc_request
*)it
.d
.lustre
.it_data
;
369 ptlrpc_req_finished(request
);
371 CERROR("lock enqueue: " DFID
" at %llu: rc %d\n",
372 PFID(ll_inode2fid(dir
)), hash
, rc
);
376 CDEBUG(D_INODE
, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
377 dir
, dir
->i_ino
, dir
->i_generation
);
378 md_set_lock_data(ll_i2sbi(dir
)->ll_md_exp
,
379 &it
.d
.lustre
.it_lock_handle
, dir
, NULL
);
381 /* for cross-ref object, l_ast_data of the lock may not be set,
384 md_set_lock_data(ll_i2sbi(dir
)->ll_md_exp
, &lockh
.cookie
,
387 ldlm_lock_dump_handle(D_OTHER
, &lockh
);
389 mutex_lock(&lli
->lli_readdir_mutex
);
390 page
= ll_dir_page_locate(dir
, &lhash
, &start
, &end
);
392 CERROR("dir page locate: "DFID
" at %llu: rc %ld\n",
393 PFID(ll_inode2fid(dir
)), lhash
, PTR_ERR(page
));
397 * XXX nikita: not entirely correct handling of a corner case:
398 * suppose hash chain of entries with hash value HASH crosses
399 * border between pages P0 and P1. First both P0 and P1 are
400 * cached, seekdir() is called for some entry from the P0 part
401 * of the chain. Later P0 goes out of cache. telldir(HASH)
402 * happens and finds P1, as it starts with matching hash
403 * value. Remaining entries from P0 part of the chain are
404 * skipped. (Is that really a bug?)
406 * Possible solutions: 0. don't cache P1 is such case, handle
407 * it as an "overflow" page. 1. invalidate all pages at
408 * once. 2. use HASH|1 as an index for P1.
413 page
= read_cache_page(mapping
, hash_x_index(hash
, hash64
),
414 ll_dir_filler
, &lhash
);
416 CERROR("read cache page: "DFID
" at %llu: rc %ld\n",
417 PFID(ll_inode2fid(dir
)), hash
, PTR_ERR(page
));
421 wait_on_page_locked(page
);
423 if (!PageUptodate(page
)) {
424 CERROR("page not updated: "DFID
" at %llu: rc %d\n",
425 PFID(ll_inode2fid(dir
)), hash
, -5);
428 if (!PageChecked(page
))
429 /* XXX: check page format later */
430 SetPageChecked(page
);
431 if (PageError(page
)) {
432 CERROR("page error: "DFID
" at %llu: rc %d\n",
433 PFID(ll_inode2fid(dir
)), hash
, -5);
437 dp
= page_address(page
);
438 if (BITS_PER_LONG
== 32 && hash64
) {
439 start
= le64_to_cpu(dp
->ldp_hash_start
) >> 32;
440 end
= le64_to_cpu(dp
->ldp_hash_end
) >> 32;
443 start
= le64_to_cpu(dp
->ldp_hash_start
);
444 end
= le64_to_cpu(dp
->ldp_hash_end
);
448 LASSERT(start
== lhash
);
449 CWARN("Page-wide hash collision: %llu\n", end
);
450 if (BITS_PER_LONG
== 32 && hash64
)
451 CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n",
452 le64_to_cpu(dp
->ldp_hash_start
),
453 le64_to_cpu(dp
->ldp_hash_end
), hash
);
455 * Fetch whole overflow chain...
462 mutex_unlock(&lli
->lli_readdir_mutex
);
463 ldlm_lock_decref(&lockh
, mode
);
467 ll_release_page(page
, 1);
468 page
= ERR_PTR(-EIO
);
472 int ll_dir_read(struct inode
*inode
, struct dir_context
*ctx
)
474 struct ll_inode_info
*info
= ll_i2info(inode
);
475 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
476 __u64 pos
= ctx
->pos
;
477 int api32
= ll_need_32bit_api(sbi
);
478 int hash64
= sbi
->ll_flags
& LL_SBI_64BIT_HASH
;
480 struct ll_dir_chain chain
;
484 ll_dir_chain_init(&chain
);
486 page
= ll_get_dir_page(inode
, pos
, &chain
);
488 while (rc
== 0 && !done
) {
489 struct lu_dirpage
*dp
;
490 struct lu_dirent
*ent
;
494 * If page is empty (end of directory is reached),
497 __u64 hash
= MDS_DIR_END_OFF
;
500 dp
= page_address(page
);
501 for (ent
= lu_dirent_start(dp
); ent
&& !done
;
502 ent
= lu_dirent_next(ent
)) {
510 * XXX: implement correct swabbing here.
513 hash
= le64_to_cpu(ent
->lde_hash
);
516 * Skip until we find target hash
521 namelen
= le16_to_cpu(ent
->lde_namelen
);
532 fid_le_to_cpu(&fid
, &ent
->lde_fid
);
533 ino
= cl_fid_build_ino(&fid
, api32
);
534 type
= ll_dirent_type_get(ent
);
536 /* For 'll_nfs_get_name_filldir()', it will try
537 * to access the 'ent' through its 'lde_name',
538 * so the parameter 'name' for 'ctx->actor()'
539 * must be part of the 'ent'.
541 done
= !dir_emit(ctx
, ent
->lde_name
,
544 next
= le64_to_cpu(dp
->ldp_hash_end
);
547 if (pos
== MDS_DIR_END_OFF
) {
549 * End of directory reached.
552 ll_release_page(page
, 0);
553 } else if (1 /* chain is exhausted*/) {
555 * Normal case: continue to the next
558 ll_release_page(page
,
559 le32_to_cpu(dp
->ldp_flags
) &
562 page
= ll_get_dir_page(inode
, pos
,
566 * go into overflow page.
568 LASSERT(le32_to_cpu(dp
->ldp_flags
) &
570 ll_release_page(page
, 1);
574 ll_release_page(page
, 0);
578 CERROR("error reading dir "DFID
" at %lu: rc %d\n",
579 PFID(&info
->lli_fid
), (unsigned long)pos
, rc
);
584 ll_dir_chain_fini(&chain
);
588 static int ll_readdir(struct file
*filp
, struct dir_context
*ctx
)
590 struct inode
*inode
= file_inode(filp
);
591 struct ll_file_data
*lfd
= LUSTRE_FPRIVATE(filp
);
592 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
593 int hash64
= sbi
->ll_flags
& LL_SBI_64BIT_HASH
;
594 int api32
= ll_need_32bit_api(sbi
);
597 CDEBUG(D_VFSTRACE
, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
598 inode
->i_ino
, inode
->i_generation
,
599 inode
, (unsigned long)lfd
->lfd_pos
, i_size_read(inode
), api32
);
601 if (lfd
->lfd_pos
== MDS_DIR_END_OFF
) {
609 ctx
->pos
= lfd
->lfd_pos
;
610 rc
= ll_dir_read(inode
, ctx
);
611 lfd
->lfd_pos
= ctx
->pos
;
612 if (ctx
->pos
== MDS_DIR_END_OFF
) {
614 ctx
->pos
= LL_DIR_END_OFF_32BIT
;
616 ctx
->pos
= LL_DIR_END_OFF
;
621 filp
->f_version
= inode
->i_version
;
625 ll_stats_ops_tally(sbi
, LPROC_LL_READDIR
, 1);
630 static int ll_send_mgc_param(struct obd_export
*mgc
, char *string
)
632 struct mgs_send_param
*msp
;
635 msp
= kzalloc(sizeof(*msp
), GFP_NOFS
);
639 strlcpy(msp
->mgs_param
, string
, sizeof(msp
->mgs_param
));
640 rc
= obd_set_info_async(NULL
, mgc
, sizeof(KEY_SET_INFO
), KEY_SET_INFO
,
641 sizeof(struct mgs_send_param
), msp
, NULL
);
643 CERROR("Failed to set parameter: %d\n", rc
);
649 static int ll_dir_setdirstripe(struct inode
*dir
, struct lmv_user_md
*lump
,
652 struct ptlrpc_request
*request
= NULL
;
653 struct md_op_data
*op_data
;
654 struct ll_sb_info
*sbi
= ll_i2sbi(dir
);
658 mode
= (~current_umask() & 0755) | S_IFDIR
;
659 op_data
= ll_prep_md_op_data(NULL
, dir
, NULL
, filename
,
660 strlen(filename
), mode
, LUSTRE_OPC_MKDIR
,
662 if (IS_ERR(op_data
)) {
663 err
= PTR_ERR(op_data
);
667 op_data
->op_cli_flags
|= CLI_SET_MEA
;
668 err
= md_create(sbi
->ll_md_exp
, op_data
, lump
, sizeof(*lump
), mode
,
669 from_kuid(&init_user_ns
, current_fsuid()),
670 from_kgid(&init_user_ns
, current_fsgid()),
671 cfs_curproc_cap_pack(), 0, &request
);
672 ll_finish_md_op_data(op_data
);
676 ptlrpc_req_finished(request
);
680 int ll_dir_setstripe(struct inode
*inode
, struct lov_user_md
*lump
,
683 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
684 struct md_op_data
*op_data
;
685 struct ptlrpc_request
*req
= NULL
;
687 struct lustre_sb_info
*lsi
= s2lsi(inode
->i_sb
);
688 struct obd_device
*mgc
= lsi
->lsi_mgc
;
693 * This is coming from userspace, so should be in
694 * local endian. But the MDS would like it in little
695 * endian, so we swab it before we send it.
697 switch (lump
->lmm_magic
) {
698 case LOV_USER_MAGIC_V1
: {
699 if (lump
->lmm_magic
!= cpu_to_le32(LOV_USER_MAGIC_V1
))
700 lustre_swab_lov_user_md_v1(lump
);
701 lum_size
= sizeof(struct lov_user_md_v1
);
704 case LOV_USER_MAGIC_V3
: {
705 if (lump
->lmm_magic
!= cpu_to_le32(LOV_USER_MAGIC_V3
))
706 lustre_swab_lov_user_md_v3(
707 (struct lov_user_md_v3
*)lump
);
708 lum_size
= sizeof(struct lov_user_md_v3
);
712 CDEBUG(D_IOCTL
, "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
713 lump
->lmm_magic
, LOV_USER_MAGIC_V1
,
719 lum_size
= sizeof(struct lov_user_md_v1
);
722 op_data
= ll_prep_md_op_data(NULL
, inode
, NULL
, NULL
, 0, 0,
723 LUSTRE_OPC_ANY
, NULL
);
725 return PTR_ERR(op_data
);
727 if (lump
&& lump
->lmm_magic
== cpu_to_le32(LMV_USER_MAGIC
))
728 op_data
->op_cli_flags
|= CLI_SET_MEA
;
730 /* swabbing is done in lov_setstripe() on server side */
731 rc
= md_setattr(sbi
->ll_md_exp
, op_data
, lump
, lum_size
,
732 NULL
, 0, &req
, NULL
);
733 ll_finish_md_op_data(op_data
);
734 ptlrpc_req_finished(req
);
736 if (rc
!= -EPERM
&& rc
!= -EACCES
)
737 CERROR("mdc_setattr fails: rc = %d\n", rc
);
740 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
741 * LOV_USER_MAGIC_V3 have the same initial fields so we do not
742 * need to make the distinction between the 2 versions
744 if (set_default
&& mgc
->u
.cli
.cl_mgc_mgsexp
) {
748 param
= kzalloc(MGS_PARAM_MAXLEN
, GFP_NOFS
);
753 /* Get fsname and assume devname to be -MDT0000. */
754 ll_get_fsname(inode
->i_sb
, buf
, MTI_NAME_MAXLEN
);
755 strcat(buf
, "-MDT0000.lov");
758 /* Set root stripesize */
759 sprintf(buf
, ".stripesize=%u",
760 lump
? le32_to_cpu(lump
->lmm_stripe_size
) : 0);
761 rc
= ll_send_mgc_param(mgc
->u
.cli
.cl_mgc_mgsexp
, param
);
765 /* Set root stripecount */
766 sprintf(buf
, ".stripecount=%hd",
767 lump
? le16_to_cpu(lump
->lmm_stripe_count
) : 0);
768 rc
= ll_send_mgc_param(mgc
->u
.cli
.cl_mgc_mgsexp
, param
);
772 /* Set root stripeoffset */
773 sprintf(buf
, ".stripeoffset=%hd",
774 lump
? le16_to_cpu(lump
->lmm_stripe_offset
) :
775 (typeof(lump
->lmm_stripe_offset
))(-1));
776 rc
= ll_send_mgc_param(mgc
->u
.cli
.cl_mgc_mgsexp
, param
);
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_default_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 %lu/%u: rc %d\n",
810 inode
->i_generation
, rc
);
814 body
= req_capsule_server_get(&req
->rq_pill
, &RMF_MDT_BODY
);
816 lmmsize
= body
->eadatasize
;
818 if (!(body
->valid
& (OBD_MD_FLEASIZE
| OBD_MD_FLDIREA
)) ||
824 lmm
= req_capsule_server_sized_get(&req
->rq_pill
,
825 &RMF_MDT_MD
, lmmsize
);
828 * This is coming from the MDS, so is probably in
829 * little endian. We convert it to host endian before
830 * passing it to userspace.
832 /* We don't swab objects for directories */
833 switch (le32_to_cpu(lmm
->lmm_magic
)) {
835 if (cpu_to_le32(LOV_MAGIC
) != LOV_MAGIC
)
836 lustre_swab_lov_user_md_v1((struct lov_user_md_v1
*)lmm
);
839 if (cpu_to_le32(LOV_MAGIC
) != LOV_MAGIC
)
840 lustre_swab_lov_user_md_v3((struct lov_user_md_v3
*)lmm
);
843 CERROR("unknown magic: %lX\n", (unsigned long)lmm
->lmm_magic
);
854 * Get MDT index for the inode.
856 int ll_get_mdt_idx(struct inode
*inode
)
858 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
859 struct md_op_data
*op_data
;
862 op_data
= ll_prep_md_op_data(NULL
, inode
, NULL
, NULL
, 0,
863 0, LUSTRE_OPC_ANY
, NULL
);
865 return PTR_ERR(op_data
);
867 op_data
->op_flags
|= MF_GET_MDT_IDX
;
868 rc
= md_getattr(sbi
->ll_md_exp
, op_data
, NULL
);
869 mdtidx
= op_data
->op_mds
;
870 ll_finish_md_op_data(op_data
);
872 CDEBUG(D_INFO
, "md_getattr_name: %d\n", rc
);
879 * Generic handler to do any pre-copy work.
881 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
882 * first information for it that real work has started.
884 * Moreover, for a ARCHIVE request, it will sample the file data version and
885 * store it in \a copy.
887 * \return 0 on success.
889 static int ll_ioc_copy_start(struct super_block
*sb
, struct hsm_copy
*copy
)
891 struct ll_sb_info
*sbi
= ll_s2sbi(sb
);
892 struct hsm_progress_kernel hpk
;
895 /* Forge a hsm_progress based on data from copy. */
896 hpk
.hpk_fid
= copy
->hc_hai
.hai_fid
;
897 hpk
.hpk_cookie
= copy
->hc_hai
.hai_cookie
;
898 hpk
.hpk_extent
.offset
= copy
->hc_hai
.hai_extent
.offset
;
899 hpk
.hpk_extent
.length
= 0;
902 hpk
.hpk_data_version
= 0;
904 /* For archive request, we need to read the current file version. */
905 if (copy
->hc_hai
.hai_action
== HSMA_ARCHIVE
) {
907 __u64 data_version
= 0;
909 /* Get inode for this fid */
910 inode
= search_inode_for_lustre(sb
, ©
->hc_hai
.hai_fid
);
912 hpk
.hpk_flags
|= HP_FLAG_RETRY
;
913 /* hpk_errval is >= 0 */
914 hpk
.hpk_errval
= -PTR_ERR(inode
);
919 /* Read current file data version */
920 rc
= ll_data_version(inode
, &data_version
, 1);
923 CDEBUG(D_HSM
, "Could not read file data version of "
924 DFID
" (rc = %d). Archive request (%#llx) could not be done.\n",
925 PFID(©
->hc_hai
.hai_fid
), rc
,
926 copy
->hc_hai
.hai_cookie
);
927 hpk
.hpk_flags
|= HP_FLAG_RETRY
;
928 /* hpk_errval must be >= 0 */
929 hpk
.hpk_errval
= -rc
;
933 /* Store in the hsm_copy for later copytool use.
934 * Always modified even if no lsm.
936 copy
->hc_data_version
= data_version
;
940 rc
= obd_iocontrol(LL_IOC_HSM_PROGRESS
, sbi
->ll_md_exp
, sizeof(hpk
),
947 * Generic handler to do any post-copy work.
949 * It will send the last hsm_progress update to coordinator to inform it
950 * that copy is finished and whether it was successful or not.
953 * - for ARCHIVE request, it will sample the file data version and compare it
954 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
955 * will be considered as failed.
956 * - for RESTORE request, it will sample the file data version and send it to
957 * coordinator which is useful if the file was imported as 'released'.
959 * \return 0 on success.
961 static int ll_ioc_copy_end(struct super_block
*sb
, struct hsm_copy
*copy
)
963 struct ll_sb_info
*sbi
= ll_s2sbi(sb
);
964 struct hsm_progress_kernel hpk
;
967 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
968 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
969 * initialized if copy_end was called with copy == NULL.
972 /* Forge a hsm_progress based on data from copy. */
973 hpk
.hpk_fid
= copy
->hc_hai
.hai_fid
;
974 hpk
.hpk_cookie
= copy
->hc_hai
.hai_cookie
;
975 hpk
.hpk_extent
= copy
->hc_hai
.hai_extent
;
976 hpk
.hpk_flags
= copy
->hc_flags
| HP_FLAG_COMPLETED
;
977 hpk
.hpk_errval
= copy
->hc_errval
;
978 hpk
.hpk_data_version
= 0;
980 /* For archive request, we need to check the file data was not changed.
982 * For restore request, we need to send the file data version, this is
983 * useful when the file was created using hsm_import.
985 if (((copy
->hc_hai
.hai_action
== HSMA_ARCHIVE
) ||
986 (copy
->hc_hai
.hai_action
== HSMA_RESTORE
)) &&
987 (copy
->hc_errval
== 0)) {
989 __u64 data_version
= 0;
991 /* Get lsm for this fid */
992 inode
= search_inode_for_lustre(sb
, ©
->hc_hai
.hai_fid
);
994 hpk
.hpk_flags
|= HP_FLAG_RETRY
;
995 /* hpk_errval must be >= 0 */
996 hpk
.hpk_errval
= -PTR_ERR(inode
);
1001 rc
= ll_data_version(inode
, &data_version
,
1002 copy
->hc_hai
.hai_action
== HSMA_ARCHIVE
);
1005 CDEBUG(D_HSM
, "Could not read file data version. Request could not be confirmed.\n");
1006 if (hpk
.hpk_errval
== 0)
1007 hpk
.hpk_errval
= -rc
;
1011 /* Store in the hsm_copy for later copytool use.
1012 * Always modified even if no lsm.
1014 hpk
.hpk_data_version
= data_version
;
1016 /* File could have been stripped during archiving, so we need
1019 if ((copy
->hc_hai
.hai_action
== HSMA_ARCHIVE
) &&
1020 (copy
->hc_data_version
!= data_version
)) {
1021 CDEBUG(D_HSM
, "File data version mismatched. File content was changed during archiving. "
1022 DFID
", start:%#llx current:%#llx\n",
1023 PFID(©
->hc_hai
.hai_fid
),
1024 copy
->hc_data_version
, data_version
);
1025 /* File was changed, send error to cdt. Do not ask for
1026 * retry because if a file is modified frequently,
1027 * the cdt will loop on retried archive requests.
1028 * The policy engine will ask for a new archive later
1029 * when the file will not be modified for some tunable
1032 /* we do not notify caller */
1033 hpk
.hpk_flags
&= ~HP_FLAG_RETRY
;
1034 /* hpk_errval must be >= 0 */
1035 hpk
.hpk_errval
= EBUSY
;
1041 rc
= obd_iocontrol(LL_IOC_HSM_PROGRESS
, sbi
->ll_md_exp
, sizeof(hpk
),
1047 static int copy_and_ioctl(int cmd
, struct obd_export
*exp
,
1048 const void __user
*data
, size_t size
)
1053 copy
= kzalloc(size
, GFP_NOFS
);
1057 if (copy_from_user(copy
, data
, size
)) {
1062 rc
= obd_iocontrol(cmd
, exp
, size
, copy
, NULL
);
1069 static int quotactl_ioctl(struct ll_sb_info
*sbi
, struct if_quotactl
*qctl
)
1071 int cmd
= qctl
->qc_cmd
;
1072 int type
= qctl
->qc_type
;
1073 int id
= qctl
->qc_id
;
1074 int valid
= qctl
->qc_valid
;
1078 case LUSTRE_Q_INVALIDATE
:
1079 case LUSTRE_Q_FINVALIDATE
:
1084 if (!capable(CFS_CAP_SYS_ADMIN
) ||
1085 sbi
->ll_flags
& LL_SBI_RMT_CLIENT
)
1089 if (((type
== USRQUOTA
&&
1090 !uid_eq(current_euid(), make_kuid(&init_user_ns
, id
))) ||
1091 (type
== GRPQUOTA
&&
1092 !in_egroup_p(make_kgid(&init_user_ns
, id
)))) &&
1093 (!capable(CFS_CAP_SYS_ADMIN
) ||
1094 sbi
->ll_flags
& LL_SBI_RMT_CLIENT
))
1100 CERROR("unsupported quotactl op: %#x\n", cmd
);
1104 if (valid
!= QC_GENERAL
) {
1105 if (sbi
->ll_flags
& LL_SBI_RMT_CLIENT
)
1108 if (cmd
== Q_GETINFO
)
1109 qctl
->qc_cmd
= Q_GETOINFO
;
1110 else if (cmd
== Q_GETQUOTA
)
1111 qctl
->qc_cmd
= Q_GETOQUOTA
;
1117 rc
= obd_iocontrol(OBD_IOC_QUOTACTL
, sbi
->ll_md_exp
,
1118 sizeof(*qctl
), qctl
, NULL
);
1121 rc
= obd_iocontrol(OBD_IOC_QUOTACTL
, sbi
->ll_dt_exp
,
1122 sizeof(*qctl
), qctl
, NULL
);
1125 rc
= obd_iocontrol(OBD_IOC_QUOTACTL
, sbi
->ll_md_exp
,
1126 sizeof(*qctl
), qctl
, NULL
);
1128 rc
= obd_iocontrol(OBD_IOC_QUOTACTL
,
1130 sizeof(*qctl
), qctl
, NULL
);
1142 struct obd_quotactl
*oqctl
;
1144 oqctl
= kzalloc(sizeof(*oqctl
), GFP_NOFS
);
1148 QCTL_COPY(oqctl
, qctl
);
1149 rc
= obd_quotactl(sbi
->ll_md_exp
, oqctl
);
1151 if (rc
!= -EALREADY
&& cmd
== Q_QUOTAON
) {
1152 oqctl
->qc_cmd
= Q_QUOTAOFF
;
1153 obd_quotactl(sbi
->ll_md_exp
, oqctl
);
1158 /* If QIF_SPACE is not set, client should collect the
1159 * space usage from OSSs by itself
1161 if (cmd
== Q_GETQUOTA
&&
1162 !(oqctl
->qc_dqblk
.dqb_valid
& QIF_SPACE
) &&
1163 !oqctl
->qc_dqblk
.dqb_curspace
) {
1164 struct obd_quotactl
*oqctl_tmp
;
1166 oqctl_tmp
= kzalloc(sizeof(*oqctl_tmp
), GFP_NOFS
);
1172 oqctl_tmp
->qc_cmd
= Q_GETOQUOTA
;
1173 oqctl_tmp
->qc_id
= oqctl
->qc_id
;
1174 oqctl_tmp
->qc_type
= oqctl
->qc_type
;
1176 /* collect space usage from OSTs */
1177 oqctl_tmp
->qc_dqblk
.dqb_curspace
= 0;
1178 rc
= obd_quotactl(sbi
->ll_dt_exp
, oqctl_tmp
);
1179 if (!rc
|| rc
== -EREMOTEIO
) {
1180 oqctl
->qc_dqblk
.dqb_curspace
=
1181 oqctl_tmp
->qc_dqblk
.dqb_curspace
;
1182 oqctl
->qc_dqblk
.dqb_valid
|= QIF_SPACE
;
1185 /* collect space & inode usage from MDTs */
1186 oqctl_tmp
->qc_dqblk
.dqb_curspace
= 0;
1187 oqctl_tmp
->qc_dqblk
.dqb_curinodes
= 0;
1188 rc
= obd_quotactl(sbi
->ll_md_exp
, oqctl_tmp
);
1189 if (!rc
|| rc
== -EREMOTEIO
) {
1190 oqctl
->qc_dqblk
.dqb_curspace
+=
1191 oqctl_tmp
->qc_dqblk
.dqb_curspace
;
1192 oqctl
->qc_dqblk
.dqb_curinodes
=
1193 oqctl_tmp
->qc_dqblk
.dqb_curinodes
;
1194 oqctl
->qc_dqblk
.dqb_valid
|= QIF_INODES
;
1196 oqctl
->qc_dqblk
.dqb_valid
&= ~QIF_SPACE
;
1202 QCTL_COPY(qctl
, oqctl
);
1209 /* This function tries to get a single name component,
1210 * to send to the server. No actual path traversal involved,
1211 * so we limit to NAME_MAX
1213 static char *ll_getname(const char __user
*filename
)
1218 tmp
= kzalloc(NAME_MAX
+ 1, GFP_KERNEL
);
1220 return ERR_PTR(-ENOMEM
);
1222 len
= strncpy_from_user(tmp
, filename
, NAME_MAX
+ 1);
1227 else if (len
> NAME_MAX
&& tmp
[NAME_MAX
] != 0)
1228 ret
= -ENAMETOOLONG
;
1237 #define ll_putname(filename) kfree(filename)
1239 static long ll_dir_ioctl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
1241 struct inode
*inode
= file_inode(file
);
1242 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
1243 struct obd_ioctl_data
*data
;
1246 CDEBUG(D_VFSTRACE
, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1247 inode
->i_ino
, inode
->i_generation
, inode
, cmd
);
1249 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1250 if (_IOC_TYPE(cmd
) == 'T' || _IOC_TYPE(cmd
) == 't') /* tty ioctls */
1253 ll_stats_ops_tally(ll_i2sbi(inode
), LPROC_LL_IOCTL
, 1);
1255 case FSFILT_IOC_GETFLAGS
:
1256 case FSFILT_IOC_SETFLAGS
:
1257 return ll_iocontrol(inode
, file
, cmd
, arg
);
1258 case FSFILT_IOC_GETVERSION_OLD
:
1259 case FSFILT_IOC_GETVERSION
:
1260 return put_user(inode
->i_generation
, (int __user
*)arg
);
1261 /* We need to special case any other ioctls we want to handle,
1262 * to send them to the MDS/OST as appropriate and to properly
1263 * network encode the arg field.
1264 case FSFILT_IOC_SETVERSION_OLD:
1265 case FSFILT_IOC_SETVERSION:
1267 case LL_IOC_GET_MDTIDX
: {
1270 mdtidx
= ll_get_mdt_idx(inode
);
1274 if (put_user((int)mdtidx
, (int __user
*)arg
))
1279 case IOC_MDC_LOOKUP
: {
1280 struct ptlrpc_request
*request
= NULL
;
1281 int namelen
, len
= 0;
1284 struct md_op_data
*op_data
;
1286 rc
= obd_ioctl_getdata(&buf
, &len
, (void __user
*)arg
);
1291 filename
= data
->ioc_inlbuf1
;
1292 namelen
= strlen(filename
);
1295 CDEBUG(D_INFO
, "IOC_MDC_LOOKUP missing filename\n");
1300 op_data
= ll_prep_md_op_data(NULL
, inode
, NULL
, filename
, namelen
,
1301 0, LUSTRE_OPC_ANY
, NULL
);
1302 if (IS_ERR(op_data
)) {
1303 rc
= PTR_ERR(op_data
);
1307 op_data
->op_valid
= OBD_MD_FLID
;
1308 rc
= md_getattr_name(sbi
->ll_md_exp
, op_data
, &request
);
1309 ll_finish_md_op_data(op_data
);
1311 CDEBUG(D_INFO
, "md_getattr_name: %d\n", rc
);
1314 ptlrpc_req_finished(request
);
1316 obd_ioctl_freedata(buf
, len
);
1319 case LL_IOC_LMV_SETSTRIPE
: {
1320 struct lmv_user_md
*lum
;
1328 rc
= obd_ioctl_getdata(&buf
, &len
, (void __user
*)arg
);
1333 if (!data
->ioc_inlbuf1
|| !data
->ioc_inlbuf2
||
1334 data
->ioc_inllen1
== 0 || data
->ioc_inllen2
== 0) {
1339 filename
= data
->ioc_inlbuf1
;
1340 namelen
= data
->ioc_inllen1
;
1343 CDEBUG(D_INFO
, "IOC_MDC_LOOKUP missing filename\n");
1347 lum
= (struct lmv_user_md
*)data
->ioc_inlbuf2
;
1348 lumlen
= data
->ioc_inllen2
;
1350 if (lum
->lum_magic
!= LMV_USER_MAGIC
||
1351 lumlen
!= sizeof(*lum
)) {
1352 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1353 filename
, lum
->lum_magic
, lumlen
, -EFAULT
);
1359 * ll_dir_setdirstripe will be used to set dir stripe
1360 * mdc_create--->mdt_reint_create (with dirstripe)
1362 rc
= ll_dir_setdirstripe(inode
, lum
, filename
);
1364 obd_ioctl_freedata(buf
, len
);
1368 case LL_IOC_LOV_SETSTRIPE
: {
1369 struct lov_user_md_v3 lumv3
;
1370 struct lov_user_md_v1
*lumv1
= (struct lov_user_md_v1
*)&lumv3
;
1371 struct lov_user_md_v1 __user
*lumv1p
= (void __user
*)arg
;
1372 struct lov_user_md_v3 __user
*lumv3p
= (void __user
*)arg
;
1374 int set_default
= 0;
1376 LASSERT(sizeof(lumv3
) == sizeof(*lumv3p
));
1377 LASSERT(sizeof(lumv3
.lmm_objects
[0]) ==
1378 sizeof(lumv3p
->lmm_objects
[0]));
1379 /* first try with v1 which is smaller than v3 */
1380 if (copy_from_user(lumv1
, lumv1p
, sizeof(*lumv1
)))
1383 if (lumv1
->lmm_magic
== LOV_USER_MAGIC_V3
) {
1384 if (copy_from_user(&lumv3
, lumv3p
, sizeof(lumv3
)))
1388 if (is_root_inode(inode
))
1391 /* in v1 and v3 cases lumv1 points to data */
1392 rc
= ll_dir_setstripe(inode
, lumv1
, set_default
);
1396 case LL_IOC_LMV_GETSTRIPE
: {
1397 struct lmv_user_md __user
*lump
= (void __user
*)arg
;
1398 struct lmv_user_md lum
;
1399 struct lmv_user_md
*tmp
;
1404 if (copy_from_user(&lum
, lump
, sizeof(struct lmv_user_md
)))
1407 if (lum
.lum_magic
!= LMV_MAGIC_V1
)
1410 lum_size
= lmv_user_md_size(1, LMV_MAGIC_V1
);
1411 tmp
= kzalloc(lum_size
, GFP_NOFS
);
1418 tmp
->lum_type
= LMV_STRIPE_TYPE
;
1419 tmp
->lum_stripe_count
= 1;
1420 mdtindex
= ll_get_mdt_idx(inode
);
1426 tmp
->lum_stripe_offset
= mdtindex
;
1427 tmp
->lum_objects
[0].lum_mds
= mdtindex
;
1428 memcpy(&tmp
->lum_objects
[0].lum_fid
, ll_inode2fid(inode
),
1429 sizeof(struct lu_fid
));
1430 if (copy_to_user((void __user
*)arg
, tmp
, lum_size
)) {
1438 case LL_IOC_LOV_SWAP_LAYOUTS
:
1440 case LL_IOC_OBD_STATFS
:
1441 return ll_obd_statfs(inode
, (void __user
*)arg
);
1442 case LL_IOC_LOV_GETSTRIPE
:
1443 case LL_IOC_MDC_GETINFO
:
1444 case IOC_MDC_GETFILEINFO
:
1445 case IOC_MDC_GETFILESTRIPE
: {
1446 struct ptlrpc_request
*request
= NULL
;
1447 struct lov_user_md __user
*lump
;
1448 struct lov_mds_md
*lmm
= NULL
;
1449 struct mdt_body
*body
;
1450 char *filename
= NULL
;
1453 if (cmd
== IOC_MDC_GETFILEINFO
||
1454 cmd
== IOC_MDC_GETFILESTRIPE
) {
1455 filename
= ll_getname((const char __user
*)arg
);
1456 if (IS_ERR(filename
))
1457 return PTR_ERR(filename
);
1459 rc
= ll_lov_getstripe_ea_info(inode
, filename
, &lmm
,
1460 &lmmsize
, &request
);
1462 rc
= ll_dir_getstripe(inode
, &lmm
, &lmmsize
, &request
);
1466 body
= req_capsule_server_get(&request
->rq_pill
,
1474 if (rc
== -ENODATA
&& (cmd
== IOC_MDC_GETFILEINFO
||
1475 cmd
== LL_IOC_MDC_GETINFO
)) {
1482 if (cmd
== IOC_MDC_GETFILESTRIPE
||
1483 cmd
== LL_IOC_LOV_GETSTRIPE
) {
1484 lump
= (struct lov_user_md __user
*)arg
;
1486 struct lov_user_mds_data __user
*lmdp
;
1488 lmdp
= (struct lov_user_mds_data __user
*)arg
;
1489 lump
= &lmdp
->lmd_lmm
;
1491 if (copy_to_user(lump
, lmm
, lmmsize
)) {
1492 if (copy_to_user(lump
, lmm
, sizeof(*lump
))) {
1499 if (cmd
== IOC_MDC_GETFILEINFO
|| cmd
== LL_IOC_MDC_GETINFO
) {
1500 struct lov_user_mds_data __user
*lmdp
;
1503 st
.st_dev
= inode
->i_sb
->s_dev
;
1504 st
.st_mode
= body
->mode
;
1505 st
.st_nlink
= body
->nlink
;
1506 st
.st_uid
= body
->uid
;
1507 st
.st_gid
= body
->gid
;
1508 st
.st_rdev
= body
->rdev
;
1509 st
.st_size
= body
->size
;
1510 st
.st_blksize
= PAGE_CACHE_SIZE
;
1511 st
.st_blocks
= body
->blocks
;
1512 st
.st_atime
= body
->atime
;
1513 st
.st_mtime
= body
->mtime
;
1514 st
.st_ctime
= body
->ctime
;
1515 st
.st_ino
= inode
->i_ino
;
1517 lmdp
= (struct lov_user_mds_data __user
*)arg
;
1518 if (copy_to_user(&lmdp
->lmd_st
, &st
, sizeof(st
))) {
1525 ptlrpc_req_finished(request
);
1527 ll_putname(filename
);
1530 case IOC_LOV_GETINFO
: {
1531 struct lov_user_mds_data __user
*lumd
;
1532 struct lov_stripe_md
*lsm
;
1533 struct lov_user_md __user
*lum
;
1534 struct lov_mds_md
*lmm
;
1538 lumd
= (struct lov_user_mds_data __user
*)arg
;
1539 lum
= &lumd
->lmd_lmm
;
1541 rc
= ll_get_max_mdsize(sbi
, &lmmsize
);
1545 lmm
= libcfs_kvzalloc(lmmsize
, GFP_NOFS
);
1548 if (copy_from_user(lmm
, lum
, lmmsize
)) {
1553 switch (lmm
->lmm_magic
) {
1554 case LOV_USER_MAGIC_V1
:
1555 if (cpu_to_le32(LOV_USER_MAGIC_V1
) == LOV_USER_MAGIC_V1
)
1557 /* swab objects first so that stripes num will be sane */
1558 lustre_swab_lov_user_md_objects(
1559 ((struct lov_user_md_v1
*)lmm
)->lmm_objects
,
1560 ((struct lov_user_md_v1
*)lmm
)->lmm_stripe_count
);
1561 lustre_swab_lov_user_md_v1((struct lov_user_md_v1
*)lmm
);
1563 case LOV_USER_MAGIC_V3
:
1564 if (cpu_to_le32(LOV_USER_MAGIC_V3
) == LOV_USER_MAGIC_V3
)
1566 /* swab objects first so that stripes num will be sane */
1567 lustre_swab_lov_user_md_objects(
1568 ((struct lov_user_md_v3
*)lmm
)->lmm_objects
,
1569 ((struct lov_user_md_v3
*)lmm
)->lmm_stripe_count
);
1570 lustre_swab_lov_user_md_v3((struct lov_user_md_v3
*)lmm
);
1577 rc
= obd_unpackmd(sbi
->ll_dt_exp
, &lsm
, lmm
, lmmsize
);
1583 /* Perform glimpse_size operation. */
1584 memset(&st
, 0, sizeof(st
));
1586 rc
= ll_glimpse_ioctl(sbi
, lsm
, &st
);
1590 if (copy_to_user(&lumd
->lmd_st
, &st
, sizeof(st
))) {
1596 obd_free_memmd(sbi
->ll_dt_exp
, &lsm
);
1601 case OBD_IOC_LLOG_CATINFO
: {
1604 case OBD_IOC_QUOTACHECK
: {
1605 struct obd_quotactl
*oqctl
;
1608 if (!capable(CFS_CAP_SYS_ADMIN
) ||
1609 sbi
->ll_flags
& LL_SBI_RMT_CLIENT
)
1612 oqctl
= kzalloc(sizeof(*oqctl
), GFP_NOFS
);
1615 oqctl
->qc_type
= arg
;
1616 rc
= obd_quotacheck(sbi
->ll_md_exp
, oqctl
);
1618 CDEBUG(D_INFO
, "md_quotacheck failed: rc %d\n", rc
);
1622 rc
= obd_quotacheck(sbi
->ll_dt_exp
, oqctl
);
1624 CDEBUG(D_INFO
, "obd_quotacheck failed: rc %d\n", rc
);
1629 case OBD_IOC_POLL_QUOTACHECK
: {
1630 struct if_quotacheck
*check
;
1632 if (!capable(CFS_CAP_SYS_ADMIN
) ||
1633 sbi
->ll_flags
& LL_SBI_RMT_CLIENT
)
1636 check
= kzalloc(sizeof(*check
), GFP_NOFS
);
1640 rc
= obd_iocontrol(cmd
, sbi
->ll_md_exp
, 0, (void *)check
,
1643 CDEBUG(D_QUOTA
, "mdc ioctl %d failed: %d\n", cmd
, rc
);
1644 if (copy_to_user((void __user
*)arg
, check
,
1646 CDEBUG(D_QUOTA
, "copy_to_user failed\n");
1650 rc
= obd_iocontrol(cmd
, sbi
->ll_dt_exp
, 0, (void *)check
,
1653 CDEBUG(D_QUOTA
, "osc ioctl %d failed: %d\n", cmd
, rc
);
1654 if (copy_to_user((void __user
*)arg
, check
,
1656 CDEBUG(D_QUOTA
, "copy_to_user failed\n");
1663 case LL_IOC_QUOTACTL
: {
1664 struct if_quotactl
*qctl
;
1666 qctl
= kzalloc(sizeof(*qctl
), GFP_NOFS
);
1670 if (copy_from_user(qctl
, (void __user
*)arg
, sizeof(*qctl
))) {
1675 rc
= quotactl_ioctl(sbi
, qctl
);
1677 if (rc
== 0 && copy_to_user((void __user
*)arg
, qctl
,
1685 case OBD_IOC_GETDTNAME
:
1686 case OBD_IOC_GETMDNAME
:
1687 return ll_get_obd_name(inode
, cmd
, arg
);
1688 case LL_IOC_FLUSHCTX
:
1689 return ll_flush_ctx(inode
);
1690 #ifdef CONFIG_FS_POSIX_ACL
1691 case LL_IOC_RMTACL
: {
1692 if (sbi
->ll_flags
& LL_SBI_RMT_CLIENT
&& is_root_inode(inode
)) {
1693 struct ll_file_data
*fd
= LUSTRE_FPRIVATE(file
);
1695 rc
= rct_add(&sbi
->ll_rct
, current_pid(), arg
);
1697 fd
->fd_flags
|= LL_FILE_RMTACL
;
1703 case LL_IOC_GETOBDCOUNT
: {
1705 struct obd_export
*exp
;
1707 if (copy_from_user(&count
, (int __user
*)arg
, sizeof(int)))
1710 /* get ost count when count is zero, get mdt count otherwise */
1711 exp
= count
? sbi
->ll_md_exp
: sbi
->ll_dt_exp
;
1712 vallen
= sizeof(count
);
1713 rc
= obd_get_info(NULL
, exp
, sizeof(KEY_TGT_COUNT
),
1714 KEY_TGT_COUNT
, &vallen
, &count
, NULL
);
1716 CERROR("get target count failed: %d\n", rc
);
1720 if (copy_to_user((int __user
*)arg
, &count
, sizeof(int)))
1725 case LL_IOC_PATH2FID
:
1726 if (copy_to_user((void __user
*)arg
, ll_inode2fid(inode
),
1727 sizeof(struct lu_fid
)))
1730 case LL_IOC_GET_CONNECT_FLAGS
: {
1731 return obd_iocontrol(cmd
, sbi
->ll_md_exp
, 0, NULL
,
1732 (void __user
*)arg
);
1734 case OBD_IOC_CHANGELOG_SEND
:
1735 case OBD_IOC_CHANGELOG_CLEAR
:
1736 if (!capable(CFS_CAP_SYS_ADMIN
))
1739 rc
= copy_and_ioctl(cmd
, sbi
->ll_md_exp
, (void __user
*)arg
,
1740 sizeof(struct ioc_changelog
));
1742 case OBD_IOC_FID2PATH
:
1743 return ll_fid2path(inode
, (void __user
*)arg
);
1744 case LL_IOC_HSM_REQUEST
: {
1745 struct hsm_user_request
*hur
;
1748 hur
= memdup_user((void __user
*)arg
, sizeof(*hur
));
1750 return PTR_ERR(hur
);
1752 /* Compute the whole struct size */
1753 totalsize
= hur_len(hur
);
1758 /* Final size will be more than double totalsize */
1759 if (totalsize
>= MDS_MAXREQSIZE
/ 3)
1762 hur
= libcfs_kvzalloc(totalsize
, GFP_NOFS
);
1766 /* Copy the whole struct */
1767 if (copy_from_user(hur
, (void __user
*)arg
, totalsize
)) {
1772 if (hur
->hur_request
.hr_action
== HUA_RELEASE
) {
1773 const struct lu_fid
*fid
;
1777 for (i
= 0; i
< hur
->hur_request
.hr_itemcount
; i
++) {
1778 fid
= &hur
->hur_user_item
[i
].hui_fid
;
1779 f
= search_inode_for_lustre(inode
->i_sb
, fid
);
1785 rc
= ll_hsm_release(f
);
1791 rc
= obd_iocontrol(cmd
, ll_i2mdexp(inode
), totalsize
,
1799 case LL_IOC_HSM_PROGRESS
: {
1800 struct hsm_progress_kernel hpk
;
1801 struct hsm_progress hp
;
1803 if (copy_from_user(&hp
, (void __user
*)arg
, sizeof(hp
)))
1806 hpk
.hpk_fid
= hp
.hp_fid
;
1807 hpk
.hpk_cookie
= hp
.hp_cookie
;
1808 hpk
.hpk_extent
= hp
.hp_extent
;
1809 hpk
.hpk_flags
= hp
.hp_flags
;
1810 hpk
.hpk_errval
= hp
.hp_errval
;
1811 hpk
.hpk_data_version
= 0;
1813 /* File may not exist in Lustre; all progress
1814 * reported to Lustre root
1816 rc
= obd_iocontrol(cmd
, sbi
->ll_md_exp
, sizeof(hpk
), &hpk
,
1820 case LL_IOC_HSM_CT_START
:
1821 rc
= copy_and_ioctl(cmd
, sbi
->ll_md_exp
, (void __user
*)arg
,
1822 sizeof(struct lustre_kernelcomm
));
1825 case LL_IOC_HSM_COPY_START
: {
1826 struct hsm_copy
*copy
;
1829 copy
= memdup_user((char __user
*)arg
, sizeof(*copy
));
1831 return PTR_ERR(copy
);
1833 rc
= ll_ioc_copy_start(inode
->i_sb
, copy
);
1834 if (copy_to_user((char __user
*)arg
, copy
, sizeof(*copy
)))
1840 case LL_IOC_HSM_COPY_END
: {
1841 struct hsm_copy
*copy
;
1844 copy
= memdup_user((char __user
*)arg
, sizeof(*copy
));
1846 return PTR_ERR(copy
);
1848 rc
= ll_ioc_copy_end(inode
->i_sb
, copy
);
1849 if (copy_to_user((char __user
*)arg
, copy
, sizeof(*copy
)))
1856 return obd_iocontrol(cmd
, sbi
->ll_dt_exp
, 0, NULL
,
1857 (void __user
*)arg
);
1861 static loff_t
ll_dir_seek(struct file
*file
, loff_t offset
, int origin
)
1863 struct inode
*inode
= file
->f_mapping
->host
;
1864 struct ll_file_data
*fd
= LUSTRE_FPRIVATE(file
);
1865 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
1866 int api32
= ll_need_32bit_api(sbi
);
1867 loff_t ret
= -EINVAL
;
1874 offset
+= file
->f_pos
;
1880 offset
+= LL_DIR_END_OFF_32BIT
;
1882 offset
+= LL_DIR_END_OFF
;
1889 ((api32
&& offset
<= LL_DIR_END_OFF_32BIT
) ||
1890 (!api32
&& offset
<= LL_DIR_END_OFF
))) {
1891 if (offset
!= file
->f_pos
) {
1892 if ((api32
&& offset
== LL_DIR_END_OFF_32BIT
) ||
1893 (!api32
&& offset
== LL_DIR_END_OFF
))
1894 fd
->lfd_pos
= MDS_DIR_END_OFF
;
1895 else if (api32
&& sbi
->ll_flags
& LL_SBI_64BIT_HASH
)
1896 fd
->lfd_pos
= offset
<< 32;
1898 fd
->lfd_pos
= offset
;
1899 file
->f_pos
= offset
;
1900 file
->f_version
= 0;
1907 inode_unlock(inode
);
1911 static int ll_dir_open(struct inode
*inode
, struct file
*file
)
1913 return ll_file_open(inode
, file
);
1916 static int ll_dir_release(struct inode
*inode
, struct file
*file
)
1918 return ll_file_release(inode
, file
);
1921 const struct file_operations ll_dir_operations
= {
1922 .llseek
= ll_dir_seek
,
1923 .open
= ll_dir_open
,
1924 .release
= ll_dir_release
,
1925 .read
= generic_read_dir
,
1926 .iterate
= ll_readdir
,
1927 .unlocked_ioctl
= ll_dir_ioctl
,