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, 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 <asm/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 <obd_support.h>
52 #include <obd_class.h>
53 #include <lustre_lib.h>
54 #include <lustre/lustre_idl.h>
55 #include <lustre_lite.h>
56 #include <lustre_dlm.h>
57 #include <lustre_fid.h>
58 #include "llite_internal.h"
61 * (new) readdir implementation overview.
63 * Original lustre readdir implementation cached exact copy of raw directory
64 * pages on the client. These pages were indexed in client page cache by
65 * logical offset in the directory file. This design, while very simple and
66 * intuitive had some inherent problems:
68 * . it implies that byte offset to the directory entry serves as a
69 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
70 * ext3/htree directory entries may move due to splits, and more
73 * . it is incompatible with the design of split directories for cmd3,
74 * that assumes that names are distributed across nodes based on their
75 * hash, and so readdir should be done in hash order.
77 * New readdir implementation does readdir in hash order, and uses hash of a
78 * file name as a telldir/seekdir cookie. This led to number of complications:
80 * . hash is not unique, so it cannot be used to index cached directory
81 * pages on the client (note, that it requires a whole pageful of hash
82 * collided entries to cause two pages to have identical hashes);
84 * . hash is not unique, so it cannot, strictly speaking, be used as an
85 * entry cookie. ext3/htree has the same problem and lustre implementation
86 * mimics their solution: seekdir(hash) positions directory at the first
87 * entry with the given hash.
93 * Client caches directory pages using hash of the first entry as an index. As
94 * noted above hash is not unique, so this solution doesn't work as is:
95 * special processing is needed for "page hash chains" (i.e., sequences of
96 * pages filled with entries all having the same hash value).
98 * First, such chains have to be detected. To this end, server returns to the
99 * client the hash of the first entry on the page next to one returned. When
100 * client detects that this hash is the same as hash of the first entry on the
101 * returned page, page hash collision has to be handled. Pages in the
102 * hash chain, except first one, are termed "overflow pages".
104 * Solution to index uniqueness problem is to not cache overflow
105 * pages. Instead, when page hash collision is detected, all overflow pages
106 * from emerging chain are immediately requested from the server and placed in
107 * a special data structure (struct ll_dir_chain). This data structure is used
108 * by ll_readdir() to process entries from overflow pages. When readdir
109 * invocation finishes, overflow pages are discarded. If page hash collision
110 * chain weren't completely processed, next call to readdir will again detect
111 * page hash collision, again read overflow pages in, process next portion of
112 * entries and again discard the pages. This is not as wasteful as it looks,
113 * because, given reasonable hash, page hash collisions are extremely rare.
115 * 1. directory positioning
117 * When seekdir(hash) is called, original
128 * identification of and access to overflow pages
132 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
133 * a header lu_dirpage which describes the start/end hash, and whether this
134 * page is empty (contains no dir entry) or hash collide with next page.
135 * After client receives reply, several pages will be integrated into dir page
136 * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
137 * lu_dirpage for this integrated page will be adjusted. See
138 * 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_CACHE_SHIFT
;
156 int nrdpgs
= 0; /* number of pages read actually */
161 CDEBUG(D_VFSTRACE
, "VFS Op:inode=%lu/%u(%p) hash "LPU64
"\n",
162 inode
->i_ino
, inode
->i_generation
, inode
, hash
);
164 LASSERT(max_pages
> 0 && max_pages
<= MD_MAX_BRW_PAGES
);
166 OBD_ALLOC(page_pool
, sizeof(page
) * max_pages
);
167 if (page_pool
!= NULL
) {
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 body
= req_capsule_server_get(&request
->rq_pill
, &RMF_MDT_BODY
);
188 /* Checked by mdc_readpage() */
189 LASSERT(body
!= NULL
);
191 if (body
->valid
& OBD_MD_FLSIZE
)
192 cl_isize_write(inode
, body
->size
);
194 nrdpgs
= (request
->rq_bulk
->bd_nob_transferred
+PAGE_CACHE_SIZE
-1)
196 SetPageUptodate(page0
);
199 ptlrpc_req_finished(request
);
201 CDEBUG(D_VFSTRACE
, "read %d/%d pages\n", nrdpgs
, npages
);
203 ll_pagevec_init(&lru_pvec
, 0);
204 for (i
= 1; i
< npages
; i
++) {
205 unsigned long offset
;
210 if (rc
< 0 || i
>= nrdpgs
) {
211 page_cache_release(page
);
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
,
228 if (ll_pagevec_add(&lru_pvec
, page
) == 0)
229 ll_pagevec_lru_add_file(&lru_pvec
);
231 CDEBUG(D_VFSTRACE
, "page %lu add to page cache failed:"
232 " %d\n", offset
, ret
);
234 page_cache_release(page
);
236 ll_pagevec_lru_add_file(&lru_pvec
);
238 if (page_pool
!= &page0
)
239 OBD_FREE(page_pool
, sizeof(struct page
*) * max_pages
);
243 static void ll_check_page(struct inode
*dir
, struct page
*page
)
245 /* XXX: check page format later */
246 SetPageChecked(page
);
249 void ll_release_page(struct page
*page
, int remove
)
254 if (likely(page
->mapping
!= NULL
))
255 truncate_complete_page(page
->mapping
, page
);
258 page_cache_release(page
);
262 * Find, kmap and return page that contains given hash.
264 static struct page
*ll_dir_page_locate(struct inode
*dir
, __u64
*hash
,
265 __u64
*start
, __u64
*end
)
267 int hash64
= ll_i2sbi(dir
)->ll_flags
& LL_SBI_64BIT_HASH
;
268 struct address_space
*mapping
= dir
->i_mapping
;
270 * Complement of hash is used as an index so that
271 * radix_tree_gang_lookup() can be used to find a page with starting
272 * hash _smaller_ than one we are looking for.
274 unsigned long offset
= hash_x_index(*hash
, hash64
);
278 TREE_READ_LOCK_IRQ(mapping
);
279 found
= radix_tree_gang_lookup(&mapping
->page_tree
,
280 (void **)&page
, offset
, 1);
282 struct lu_dirpage
*dp
;
284 page_cache_get(page
);
285 TREE_READ_UNLOCK_IRQ(mapping
);
287 * In contrast to find_lock_page() we are sure that directory
288 * page cannot be truncated (while DLM lock is held) and,
289 * hence, can avoid restart.
291 * In fact, page cannot be locked here at all, because
292 * ll_dir_filler() does synchronous io.
294 wait_on_page_locked(page
);
295 if (PageUptodate(page
)) {
297 if (BITS_PER_LONG
== 32 && hash64
) {
298 *start
= le64_to_cpu(dp
->ldp_hash_start
) >> 32;
299 *end
= le64_to_cpu(dp
->ldp_hash_end
) >> 32;
302 *start
= le64_to_cpu(dp
->ldp_hash_start
);
303 *end
= le64_to_cpu(dp
->ldp_hash_end
);
305 LASSERTF(*start
<= *hash
, "start = "LPX64
",end = "
306 LPX64
",hash = "LPX64
"\n", *start
, *end
, *hash
);
307 CDEBUG(D_VFSTRACE
, "page %lu [%llu %llu], hash "LPU64
"\n",
308 offset
, *start
, *end
, *hash
);
310 ll_release_page(page
, 0);
312 } else if (*end
!= *start
&& *hash
== *end
) {
314 * upon hash collision, remove this page,
315 * otherwise put page reference, and
316 * ll_get_dir_page() will issue RPC to fetch
319 ll_release_page(page
,
320 le32_to_cpu(dp
->ldp_flags
) & LDF_COLLIDE
);
324 page_cache_release(page
);
325 page
= ERR_PTR(-EIO
);
329 TREE_READ_UNLOCK_IRQ(mapping
);
335 struct page
*ll_get_dir_page(struct inode
*dir
, __u64 hash
,
336 struct ll_dir_chain
*chain
)
338 ldlm_policy_data_t policy
= {.l_inodebits
= {MDS_INODELOCK_UPDATE
} };
339 struct address_space
*mapping
= dir
->i_mapping
;
340 struct lustre_handle lockh
;
341 struct lu_dirpage
*dp
;
348 struct ll_inode_info
*lli
= ll_i2info(dir
);
349 int hash64
= ll_i2sbi(dir
)->ll_flags
& LL_SBI_64BIT_HASH
;
352 rc
= md_lock_match(ll_i2sbi(dir
)->ll_md_exp
, LDLM_FL_BLOCK_GRANTED
,
353 ll_inode2fid(dir
), LDLM_IBITS
, &policy
, mode
, &lockh
);
355 struct ldlm_enqueue_info einfo
= {
356 .ei_type
= LDLM_IBITS
,
358 .ei_cb_bl
= ll_md_blocking_ast
,
359 .ei_cb_cp
= ldlm_completion_ast
,
361 struct lookup_intent it
= { .it_op
= IT_READDIR
};
362 struct ptlrpc_request
*request
;
363 struct md_op_data
*op_data
;
365 op_data
= ll_prep_md_op_data(NULL
, dir
, NULL
, NULL
, 0, 0,
366 LUSTRE_OPC_ANY
, NULL
);
368 return (void *)op_data
;
370 rc
= md_enqueue(ll_i2sbi(dir
)->ll_md_exp
, &einfo
, &it
,
371 op_data
, &lockh
, NULL
, 0, NULL
, 0);
373 ll_finish_md_op_data(op_data
);
375 request
= (struct ptlrpc_request
*)it
.d
.lustre
.it_data
;
377 ptlrpc_req_finished(request
);
379 CERROR("lock enqueue: "DFID
" at "LPU64
": rc %d\n",
380 PFID(ll_inode2fid(dir
)), hash
, rc
);
384 CDEBUG(D_INODE
, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
385 dir
, dir
->i_ino
, dir
->i_generation
);
386 md_set_lock_data(ll_i2sbi(dir
)->ll_md_exp
,
387 &it
.d
.lustre
.it_lock_handle
, dir
, NULL
);
389 /* for cross-ref object, l_ast_data of the lock may not be set,
390 * we reset it here */
391 md_set_lock_data(ll_i2sbi(dir
)->ll_md_exp
, &lockh
.cookie
,
394 ldlm_lock_dump_handle(D_OTHER
, &lockh
);
396 mutex_lock(&lli
->lli_readdir_mutex
);
397 page
= ll_dir_page_locate(dir
, &lhash
, &start
, &end
);
399 CERROR("dir page locate: "DFID
" at "LPU64
": rc %ld\n",
400 PFID(ll_inode2fid(dir
)), lhash
, PTR_ERR(page
));
401 GOTO(out_unlock
, page
);
402 } else if (page
!= NULL
) {
404 * XXX nikita: not entirely correct handling of a corner case:
405 * suppose hash chain of entries with hash value HASH crosses
406 * border between pages P0 and P1. First both P0 and P1 are
407 * cached, seekdir() is called for some entry from the P0 part
408 * of the chain. Later P0 goes out of cache. telldir(HASH)
409 * happens and finds P1, as it starts with matching hash
410 * value. Remaining entries from P0 part of the chain are
411 * skipped. (Is that really a bug?)
413 * Possible solutions: 0. don't cache P1 is such case, handle
414 * it as an "overflow" page. 1. invalidate all pages at
415 * once. 2. use HASH|1 as an index for P1.
417 GOTO(hash_collision
, page
);
420 page
= read_cache_page(mapping
, hash_x_index(hash
, hash64
),
421 ll_dir_filler
, &lhash
);
423 CERROR("read cache page: "DFID
" at "LPU64
": rc %ld\n",
424 PFID(ll_inode2fid(dir
)), hash
, PTR_ERR(page
));
425 GOTO(out_unlock
, page
);
428 wait_on_page_locked(page
);
430 if (!PageUptodate(page
)) {
431 CERROR("page not updated: "DFID
" at "LPU64
": rc %d\n",
432 PFID(ll_inode2fid(dir
)), hash
, -5);
435 if (!PageChecked(page
))
436 ll_check_page(dir
, page
);
437 if (PageError(page
)) {
438 CERROR("page error: "DFID
" at "LPU64
": rc %d\n",
439 PFID(ll_inode2fid(dir
)), hash
, -5);
443 dp
= page_address(page
);
444 if (BITS_PER_LONG
== 32 && hash64
) {
445 start
= le64_to_cpu(dp
->ldp_hash_start
) >> 32;
446 end
= le64_to_cpu(dp
->ldp_hash_end
) >> 32;
449 start
= le64_to_cpu(dp
->ldp_hash_start
);
450 end
= le64_to_cpu(dp
->ldp_hash_end
);
454 LASSERT(start
== lhash
);
455 CWARN("Page-wide hash collision: "LPU64
"\n", end
);
456 if (BITS_PER_LONG
== 32 && hash64
)
457 CWARN("Real page-wide hash collision at ["LPU64
" "LPU64
458 "] with hash "LPU64
"\n",
459 le64_to_cpu(dp
->ldp_hash_start
),
460 le64_to_cpu(dp
->ldp_hash_end
), hash
);
462 * Fetch whole overflow chain...
469 mutex_unlock(&lli
->lli_readdir_mutex
);
470 ldlm_lock_decref(&lockh
, mode
);
474 ll_release_page(page
, 1);
475 page
= ERR_PTR(-EIO
);
479 int ll_dir_read(struct inode
*inode
, struct dir_context
*ctx
)
481 struct ll_inode_info
*info
= ll_i2info(inode
);
482 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
483 __u64 pos
= ctx
->pos
;
484 int api32
= ll_need_32bit_api(sbi
);
485 int hash64
= sbi
->ll_flags
& LL_SBI_64BIT_HASH
;
487 struct ll_dir_chain chain
;
491 ll_dir_chain_init(&chain
);
493 page
= ll_get_dir_page(inode
, pos
, &chain
);
495 while (rc
== 0 && !done
) {
496 struct lu_dirpage
*dp
;
497 struct lu_dirent
*ent
;
501 * If page is empty (end of directory is reached),
504 __u64 hash
= MDS_DIR_END_OFF
;
507 dp
= page_address(page
);
508 for (ent
= lu_dirent_start(dp
); ent
!= NULL
&& !done
;
509 ent
= lu_dirent_next(ent
)) {
517 * XXX: implement correct swabbing here.
520 hash
= le64_to_cpu(ent
->lde_hash
);
523 * Skip until we find target hash
528 namelen
= le16_to_cpu(ent
->lde_namelen
);
539 fid_le_to_cpu(&fid
, &ent
->lde_fid
);
540 ino
= cl_fid_build_ino(&fid
, api32
);
541 type
= ll_dirent_type_get(ent
);
543 /* For 'll_nfs_get_name_filldir()', it will try
544 * to access the 'ent' through its 'lde_name',
545 * so the parameter 'name' for 'ctx->actor()'
546 * must be part of the 'ent'.
548 done
= !dir_emit(ctx
, ent
->lde_name
,
551 next
= le64_to_cpu(dp
->ldp_hash_end
);
554 if (pos
== MDS_DIR_END_OFF
) {
556 * End of directory reached.
559 ll_release_page(page
, 0);
560 } else if (1 /* chain is exhausted*/) {
562 * Normal case: continue to the next
565 ll_release_page(page
,
566 le32_to_cpu(dp
->ldp_flags
) &
569 page
= ll_get_dir_page(inode
, pos
,
573 * go into overflow page.
575 LASSERT(le32_to_cpu(dp
->ldp_flags
) &
577 ll_release_page(page
, 1);
581 ll_release_page(page
, 0);
585 CERROR("error reading dir "DFID
" at %lu: rc %d\n",
586 PFID(&info
->lli_fid
), (unsigned long)pos
, rc
);
591 ll_dir_chain_fini(&chain
);
595 static int ll_readdir(struct file
*filp
, struct dir_context
*ctx
)
597 struct inode
*inode
= filp
->f_dentry
->d_inode
;
598 struct ll_file_data
*lfd
= LUSTRE_FPRIVATE(filp
);
599 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
600 int hash64
= sbi
->ll_flags
& LL_SBI_64BIT_HASH
;
601 int api32
= ll_need_32bit_api(sbi
);
604 CDEBUG(D_VFSTRACE
, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu "
605 " 32bit_api %d\n", inode
->i_ino
, inode
->i_generation
,
606 inode
, (unsigned long)lfd
->lfd_pos
, i_size_read(inode
), api32
);
608 if (lfd
->lfd_pos
== MDS_DIR_END_OFF
)
614 ctx
->pos
= lfd
->lfd_pos
;
615 rc
= ll_dir_read(inode
, ctx
);
616 lfd
->lfd_pos
= ctx
->pos
;
617 if (ctx
->pos
== MDS_DIR_END_OFF
) {
619 ctx
->pos
= LL_DIR_END_OFF_32BIT
;
621 ctx
->pos
= LL_DIR_END_OFF
;
626 filp
->f_version
= inode
->i_version
;
630 ll_stats_ops_tally(sbi
, LPROC_LL_READDIR
, 1);
635 int ll_send_mgc_param(struct obd_export
*mgc
, char *string
)
637 struct mgs_send_param
*msp
;
644 strncpy(msp
->mgs_param
, string
, MGS_PARAM_MAXLEN
);
645 rc
= obd_set_info_async(NULL
, mgc
, sizeof(KEY_SET_INFO
), KEY_SET_INFO
,
646 sizeof(struct mgs_send_param
), msp
, NULL
);
648 CERROR("Failed to set parameter: %d\n", rc
);
654 int ll_dir_setdirstripe(struct inode
*dir
, struct lmv_user_md
*lump
,
657 struct ptlrpc_request
*request
= NULL
;
658 struct md_op_data
*op_data
;
659 struct ll_sb_info
*sbi
= ll_i2sbi(dir
);
663 mode
= (0755 & (S_IRWXUGO
|S_ISVTX
) & ~current
->fs
->umask
) | S_IFDIR
;
664 op_data
= ll_prep_md_op_data(NULL
, dir
, NULL
, filename
,
665 strlen(filename
), mode
, LUSTRE_OPC_MKDIR
,
668 GOTO(err_exit
, err
= PTR_ERR(op_data
));
670 op_data
->op_cli_flags
|= CLI_SET_MEA
;
671 err
= md_create(sbi
->ll_md_exp
, op_data
, lump
, sizeof(*lump
), mode
,
672 from_kuid(&init_user_ns
, current_fsuid()),
673 from_kgid(&init_user_ns
, current_fsgid()),
674 cfs_curproc_cap_pack(), 0, &request
);
675 ll_finish_md_op_data(op_data
);
679 ptlrpc_req_finished(request
);
683 int ll_dir_setstripe(struct inode
*inode
, struct lov_user_md
*lump
,
686 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
687 struct md_op_data
*op_data
;
688 struct ptlrpc_request
*req
= NULL
;
690 struct lustre_sb_info
*lsi
= s2lsi(inode
->i_sb
);
691 struct obd_device
*mgc
= lsi
->lsi_mgc
;
696 * This is coming from userspace, so should be in
697 * local endian. But the MDS would like it in little
698 * endian, so we swab it before we send it.
700 switch (lump
->lmm_magic
) {
701 case LOV_USER_MAGIC_V1
: {
702 if (lump
->lmm_magic
!= cpu_to_le32(LOV_USER_MAGIC_V1
))
703 lustre_swab_lov_user_md_v1(lump
);
704 lum_size
= sizeof(struct lov_user_md_v1
);
707 case LOV_USER_MAGIC_V3
: {
708 if (lump
->lmm_magic
!= cpu_to_le32(LOV_USER_MAGIC_V3
))
709 lustre_swab_lov_user_md_v3(
710 (struct lov_user_md_v3
*)lump
);
711 lum_size
= sizeof(struct lov_user_md_v3
);
715 CDEBUG(D_IOCTL
, "bad userland LOV MAGIC:"
716 " %#08x != %#08x nor %#08x\n",
717 lump
->lmm_magic
, LOV_USER_MAGIC_V1
,
723 lum_size
= sizeof(struct lov_user_md_v1
);
726 op_data
= ll_prep_md_op_data(NULL
, inode
, NULL
, NULL
, 0, 0,
727 LUSTRE_OPC_ANY
, NULL
);
729 return PTR_ERR(op_data
);
731 if (lump
!= NULL
&& lump
->lmm_magic
== cpu_to_le32(LMV_USER_MAGIC
))
732 op_data
->op_cli_flags
|= CLI_SET_MEA
;
734 /* swabbing is done in lov_setstripe() on server side */
735 rc
= md_setattr(sbi
->ll_md_exp
, op_data
, lump
, lum_size
,
736 NULL
, 0, &req
, NULL
);
737 ll_finish_md_op_data(op_data
);
738 ptlrpc_req_finished(req
);
740 if (rc
!= -EPERM
&& rc
!= -EACCES
)
741 CERROR("mdc_setattr fails: rc = %d\n", rc
);
744 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
745 LOV_USER_MAGIC_V3 have the same initial fields so we do not
746 need to make the distinction between the 2 versions */
747 if (set_default
&& mgc
->u
.cli
.cl_mgc_mgsexp
) {
751 OBD_ALLOC(param
, MGS_PARAM_MAXLEN
);
753 GOTO(end
, rc
= -ENOMEM
);
756 /* Get fsname and assume devname to be -MDT0000. */
757 ll_get_fsname(inode
->i_sb
, buf
, MTI_NAME_MAXLEN
);
758 strcat(buf
, "-MDT0000.lov");
761 /* Set root stripesize */
762 sprintf(buf
, ".stripesize=%u",
763 lump
? le32_to_cpu(lump
->lmm_stripe_size
) : 0);
764 rc
= ll_send_mgc_param(mgc
->u
.cli
.cl_mgc_mgsexp
, param
);
768 /* Set root stripecount */
769 sprintf(buf
, ".stripecount=%hd",
770 lump
? le16_to_cpu(lump
->lmm_stripe_count
) : 0);
771 rc
= ll_send_mgc_param(mgc
->u
.cli
.cl_mgc_mgsexp
, param
);
775 /* Set root stripeoffset */
776 sprintf(buf
, ".stripeoffset=%hd",
777 lump
? le16_to_cpu(lump
->lmm_stripe_offset
) :
778 (typeof(lump
->lmm_stripe_offset
))(-1));
779 rc
= ll_send_mgc_param(mgc
->u
.cli
.cl_mgc_mgsexp
, param
);
783 OBD_FREE(param
, MGS_PARAM_MAXLEN
);
788 int ll_dir_getstripe(struct inode
*inode
, struct lov_mds_md
**lmmp
,
789 int *lmm_size
, struct ptlrpc_request
**request
)
791 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
792 struct mdt_body
*body
;
793 struct lov_mds_md
*lmm
= NULL
;
794 struct ptlrpc_request
*req
= NULL
;
796 struct md_op_data
*op_data
;
798 rc
= ll_get_max_mdsize(sbi
, &lmmsize
);
802 op_data
= ll_prep_md_op_data(NULL
, inode
, NULL
, NULL
,
803 0, lmmsize
, LUSTRE_OPC_ANY
,
806 return PTR_ERR(op_data
);
808 op_data
->op_valid
= OBD_MD_FLEASIZE
| OBD_MD_FLDIREA
;
809 rc
= md_getattr(sbi
->ll_md_exp
, op_data
, &req
);
810 ll_finish_md_op_data(op_data
);
812 CDEBUG(D_INFO
, "md_getattr failed on inode "
813 "%lu/%u: rc %d\n", inode
->i_ino
,
814 inode
->i_generation
, rc
);
818 body
= req_capsule_server_get(&req
->rq_pill
, &RMF_MDT_BODY
);
819 LASSERT(body
!= NULL
);
821 lmmsize
= body
->eadatasize
;
823 if (!(body
->valid
& (OBD_MD_FLEASIZE
| OBD_MD_FLDIREA
)) ||
825 GOTO(out
, rc
= -ENODATA
);
828 lmm
= req_capsule_server_sized_get(&req
->rq_pill
,
829 &RMF_MDT_MD
, lmmsize
);
830 LASSERT(lmm
!= NULL
);
833 * This is coming from the MDS, so is probably in
834 * little endian. We convert it to host endian before
835 * passing it to userspace.
837 /* We don't swab objects for directories */
838 switch (le32_to_cpu(lmm
->lmm_magic
)) {
840 if (LOV_MAGIC
!= cpu_to_le32(LOV_MAGIC
))
841 lustre_swab_lov_user_md_v1((struct lov_user_md_v1
*)lmm
);
844 if (LOV_MAGIC
!= cpu_to_le32(LOV_MAGIC
))
845 lustre_swab_lov_user_md_v3((struct lov_user_md_v3
*)lmm
);
848 CERROR("unknown magic: %lX\n", (unsigned long)lmm
->lmm_magic
);
859 * Get MDT index for the inode.
861 int ll_get_mdt_idx(struct inode
*inode
)
863 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
864 struct md_op_data
*op_data
;
867 op_data
= ll_prep_md_op_data(NULL
, inode
, NULL
, NULL
, 0,
868 0, LUSTRE_OPC_ANY
, NULL
);
870 return PTR_ERR(op_data
);
872 op_data
->op_flags
|= MF_GET_MDT_IDX
;
873 rc
= md_getattr(sbi
->ll_md_exp
, op_data
, NULL
);
874 mdtidx
= op_data
->op_mds
;
875 ll_finish_md_op_data(op_data
);
877 CDEBUG(D_INFO
, "md_getattr_name: %d\n", rc
);
884 * Generic handler to do any pre-copy work.
886 * It send a first hsm_progress (with extent length == 0) to coordinator as a
887 * first information for it that real work has started.
889 * Moreover, for a ARCHIVE request, it will sample the file data version and
890 * store it in \a copy.
892 * \return 0 on success.
894 static int ll_ioc_copy_start(struct super_block
*sb
, struct hsm_copy
*copy
)
896 struct ll_sb_info
*sbi
= ll_s2sbi(sb
);
897 struct hsm_progress_kernel hpk
;
900 /* Forge a hsm_progress based on data from copy. */
901 hpk
.hpk_fid
= copy
->hc_hai
.hai_fid
;
902 hpk
.hpk_cookie
= copy
->hc_hai
.hai_cookie
;
903 hpk
.hpk_extent
.offset
= copy
->hc_hai
.hai_extent
.offset
;
904 hpk
.hpk_extent
.length
= 0;
907 hpk
.hpk_data_version
= 0;
910 /* For archive request, we need to read the current file version. */
911 if (copy
->hc_hai
.hai_action
== HSMA_ARCHIVE
) {
913 __u64 data_version
= 0;
915 /* Get inode for this fid */
916 inode
= search_inode_for_lustre(sb
, ©
->hc_hai
.hai_fid
);
918 hpk
.hpk_flags
|= HP_FLAG_RETRY
;
919 /* hpk_errval is >= 0 */
920 hpk
.hpk_errval
= -PTR_ERR(inode
);
921 GOTO(progress
, rc
= PTR_ERR(inode
));
924 /* Read current file data version */
925 rc
= ll_data_version(inode
, &data_version
, 1);
928 CDEBUG(D_HSM
, "Could not read file data version of "
929 DFID
" (rc = %d). Archive request ("
930 LPX64
") could not be done.\n",
931 PFID(©
->hc_hai
.hai_fid
), rc
,
932 copy
->hc_hai
.hai_cookie
);
933 hpk
.hpk_flags
|= HP_FLAG_RETRY
;
934 /* hpk_errval must be >= 0 */
935 hpk
.hpk_errval
= -rc
;
939 /* Store it the hsm_copy for later copytool use.
940 * Always modified even if no lsm. */
941 copy
->hc_data_version
= data_version
;
945 rc
= obd_iocontrol(LL_IOC_HSM_PROGRESS
, sbi
->ll_md_exp
, sizeof(hpk
),
952 * Generic handler to do any post-copy work.
954 * It will send the last hsm_progress update to coordinator to inform it
955 * that copy is finished and whether it was successful or not.
958 * - for ARCHIVE request, it will sample the file data version and compare it
959 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
960 * will be considered as failed.
961 * - for RESTORE request, it will sample the file data version and send it to
962 * coordinator which is useful if the file was imported as 'released'.
964 * \return 0 on success.
966 static int ll_ioc_copy_end(struct super_block
*sb
, struct hsm_copy
*copy
)
968 struct ll_sb_info
*sbi
= ll_s2sbi(sb
);
969 struct hsm_progress_kernel hpk
;
972 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
973 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
974 * initialized if copy_end was called with copy == NULL.
977 /* Forge a hsm_progress based on data from copy. */
978 hpk
.hpk_fid
= copy
->hc_hai
.hai_fid
;
979 hpk
.hpk_cookie
= copy
->hc_hai
.hai_cookie
;
980 hpk
.hpk_extent
= copy
->hc_hai
.hai_extent
;
981 hpk
.hpk_flags
= copy
->hc_flags
| HP_FLAG_COMPLETED
;
982 hpk
.hpk_errval
= copy
->hc_errval
;
983 hpk
.hpk_data_version
= 0;
985 /* For archive request, we need to check the file data was not changed.
987 * For restore request, we need to send the file data version, this is
988 * useful when the file was created using hsm_import.
990 if (((copy
->hc_hai
.hai_action
== HSMA_ARCHIVE
) ||
991 (copy
->hc_hai
.hai_action
== HSMA_RESTORE
)) &&
992 (copy
->hc_errval
== 0)) {
994 __u64 data_version
= 0;
996 /* Get lsm for this fid */
997 inode
= search_inode_for_lustre(sb
, ©
->hc_hai
.hai_fid
);
999 hpk
.hpk_flags
|= HP_FLAG_RETRY
;
1000 /* hpk_errval must be >= 0 */
1001 hpk
.hpk_errval
= -PTR_ERR(inode
);
1002 GOTO(progress
, rc
= PTR_ERR(inode
));
1005 rc
= ll_data_version(inode
, &data_version
,
1006 copy
->hc_hai
.hai_action
== HSMA_ARCHIVE
);
1009 CDEBUG(D_HSM
, "Could not read file data version. "
1010 "Request could not be confirmed.\n");
1011 if (hpk
.hpk_errval
== 0)
1012 hpk
.hpk_errval
= -rc
;
1016 /* Store it the hsm_copy for later copytool use.
1017 * Always modified even if no lsm. */
1018 hpk
.hpk_data_version
= data_version
;
1020 /* File could have been stripped during archiving, so we need
1021 * to check anyway. */
1022 if ((copy
->hc_hai
.hai_action
== HSMA_ARCHIVE
) &&
1023 (copy
->hc_data_version
!= data_version
)) {
1024 CDEBUG(D_HSM
, "File data version mismatched. "
1025 "File content was changed during archiving. "
1026 DFID
", start:"LPX64
" current:"LPX64
"\n",
1027 PFID(©
->hc_hai
.hai_fid
),
1028 copy
->hc_data_version
, data_version
);
1029 /* File was changed, send error to cdt. Do not ask for
1030 * retry because if a file is modified frequently,
1031 * the cdt will loop on retried archive requests.
1032 * The policy engine will ask for a new archive later
1033 * when the file will not be modified for some tunable
1035 /* we do not notify caller */
1036 hpk
.hpk_flags
&= ~HP_FLAG_RETRY
;
1037 /* hpk_errval must be >= 0 */
1038 hpk
.hpk_errval
= EBUSY
;
1044 rc
= obd_iocontrol(LL_IOC_HSM_PROGRESS
, sbi
->ll_md_exp
, sizeof(hpk
),
1051 static int copy_and_ioctl(int cmd
, struct obd_export
*exp
, void *data
, int len
)
1056 OBD_ALLOC(ptr
, len
);
1059 if (copy_from_user(ptr
, data
, len
)) {
1063 rc
= obd_iocontrol(cmd
, exp
, len
, data
, 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 (!cfs_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 (!cfs_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 OBD_ALLOC_PTR(oqctl
);
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
);
1154 OBD_FREE_PTR(oqctl
);
1157 /* If QIF_SPACE is not set, client should collect the
1158 * space usage from OSSs by itself */
1159 if (cmd
== Q_GETQUOTA
&&
1160 !(oqctl
->qc_dqblk
.dqb_valid
& QIF_SPACE
) &&
1161 !oqctl
->qc_dqblk
.dqb_curspace
) {
1162 struct obd_quotactl
*oqctl_tmp
;
1164 OBD_ALLOC_PTR(oqctl_tmp
);
1165 if (oqctl_tmp
== NULL
)
1166 GOTO(out
, rc
= -ENOMEM
);
1168 oqctl_tmp
->qc_cmd
= Q_GETOQUOTA
;
1169 oqctl_tmp
->qc_id
= oqctl
->qc_id
;
1170 oqctl_tmp
->qc_type
= oqctl
->qc_type
;
1172 /* collect space usage from OSTs */
1173 oqctl_tmp
->qc_dqblk
.dqb_curspace
= 0;
1174 rc
= obd_quotactl(sbi
->ll_dt_exp
, oqctl_tmp
);
1175 if (!rc
|| rc
== -EREMOTEIO
) {
1176 oqctl
->qc_dqblk
.dqb_curspace
=
1177 oqctl_tmp
->qc_dqblk
.dqb_curspace
;
1178 oqctl
->qc_dqblk
.dqb_valid
|= QIF_SPACE
;
1181 /* collect space & inode usage from MDTs */
1182 oqctl_tmp
->qc_dqblk
.dqb_curspace
= 0;
1183 oqctl_tmp
->qc_dqblk
.dqb_curinodes
= 0;
1184 rc
= obd_quotactl(sbi
->ll_md_exp
, oqctl_tmp
);
1185 if (!rc
|| rc
== -EREMOTEIO
) {
1186 oqctl
->qc_dqblk
.dqb_curspace
+=
1187 oqctl_tmp
->qc_dqblk
.dqb_curspace
;
1188 oqctl
->qc_dqblk
.dqb_curinodes
=
1189 oqctl_tmp
->qc_dqblk
.dqb_curinodes
;
1190 oqctl
->qc_dqblk
.dqb_valid
|= QIF_INODES
;
1192 oqctl
->qc_dqblk
.dqb_valid
&= ~QIF_SPACE
;
1195 OBD_FREE_PTR(oqctl_tmp
);
1198 QCTL_COPY(qctl
, oqctl
);
1199 OBD_FREE_PTR(oqctl
);
1206 ll_getname(const char __user
*filename
)
1209 char *tmp
= __getname();
1212 return ERR_PTR(-ENOMEM
);
1214 len
= strncpy_from_user(tmp
, filename
, PATH_MAX
);
1217 else if (len
> PATH_MAX
)
1218 ret
= -ENAMETOOLONG
;
1227 #define ll_putname(filename) __putname(filename)
1229 static long ll_dir_ioctl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
1231 struct inode
*inode
= file
->f_dentry
->d_inode
;
1232 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
1233 struct obd_ioctl_data
*data
;
1236 CDEBUG(D_VFSTRACE
, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1237 inode
->i_ino
, inode
->i_generation
, inode
, cmd
);
1239 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1240 if (_IOC_TYPE(cmd
) == 'T' || _IOC_TYPE(cmd
) == 't') /* tty ioctls */
1243 ll_stats_ops_tally(ll_i2sbi(inode
), LPROC_LL_IOCTL
, 1);
1245 case FSFILT_IOC_GETFLAGS
:
1246 case FSFILT_IOC_SETFLAGS
:
1247 return ll_iocontrol(inode
, file
, cmd
, arg
);
1248 case FSFILT_IOC_GETVERSION_OLD
:
1249 case FSFILT_IOC_GETVERSION
:
1250 return put_user(inode
->i_generation
, (int *)arg
);
1251 /* We need to special case any other ioctls we want to handle,
1252 * to send them to the MDS/OST as appropriate and to properly
1253 * network encode the arg field.
1254 case FSFILT_IOC_SETVERSION_OLD:
1255 case FSFILT_IOC_SETVERSION:
1257 case LL_IOC_GET_MDTIDX
: {
1260 mdtidx
= ll_get_mdt_idx(inode
);
1264 if (put_user((int)mdtidx
, (int*)arg
))
1269 case IOC_MDC_LOOKUP
: {
1270 struct ptlrpc_request
*request
= NULL
;
1271 int namelen
, len
= 0;
1274 struct md_op_data
*op_data
;
1276 rc
= obd_ioctl_getdata(&buf
, &len
, (void *)arg
);
1281 filename
= data
->ioc_inlbuf1
;
1282 namelen
= strlen(filename
);
1285 CDEBUG(D_INFO
, "IOC_MDC_LOOKUP missing filename\n");
1286 GOTO(out_free
, rc
= -EINVAL
);
1289 op_data
= ll_prep_md_op_data(NULL
, inode
, NULL
, filename
, namelen
,
1290 0, LUSTRE_OPC_ANY
, NULL
);
1291 if (IS_ERR(op_data
))
1292 GOTO(out_free
, rc
= PTR_ERR(op_data
));
1294 op_data
->op_valid
= OBD_MD_FLID
;
1295 rc
= md_getattr_name(sbi
->ll_md_exp
, op_data
, &request
);
1296 ll_finish_md_op_data(op_data
);
1298 CDEBUG(D_INFO
, "md_getattr_name: %d\n", rc
);
1301 ptlrpc_req_finished(request
);
1303 obd_ioctl_freedata(buf
, len
);
1306 case LL_IOC_LMV_SETSTRIPE
: {
1307 struct lmv_user_md
*lum
;
1315 rc
= obd_ioctl_getdata(&buf
, &len
, (void *)arg
);
1320 if (data
->ioc_inlbuf1
== NULL
|| data
->ioc_inlbuf2
== NULL
||
1321 data
->ioc_inllen1
== 0 || data
->ioc_inllen2
== 0)
1322 GOTO(lmv_out_free
, rc
= -EINVAL
);
1324 filename
= data
->ioc_inlbuf1
;
1325 namelen
= data
->ioc_inllen1
;
1328 CDEBUG(D_INFO
, "IOC_MDC_LOOKUP missing filename\n");
1329 GOTO(lmv_out_free
, rc
= -EINVAL
);
1331 lum
= (struct lmv_user_md
*)data
->ioc_inlbuf2
;
1332 lumlen
= data
->ioc_inllen2
;
1334 if (lum
->lum_magic
!= LMV_USER_MAGIC
||
1335 lumlen
!= sizeof(*lum
)) {
1336 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1337 filename
, lum
->lum_magic
, lumlen
, -EFAULT
);
1338 GOTO(lmv_out_free
, rc
= -EINVAL
);
1342 * ll_dir_setdirstripe will be used to set dir stripe
1343 * mdc_create--->mdt_reint_create (with dirstripe)
1345 rc
= ll_dir_setdirstripe(inode
, lum
, filename
);
1347 obd_ioctl_freedata(buf
, len
);
1351 case LL_IOC_LOV_SETSTRIPE
: {
1352 struct lov_user_md_v3 lumv3
;
1353 struct lov_user_md_v1
*lumv1
= (struct lov_user_md_v1
*)&lumv3
;
1354 struct lov_user_md_v1
*lumv1p
= (struct lov_user_md_v1
*)arg
;
1355 struct lov_user_md_v3
*lumv3p
= (struct lov_user_md_v3
*)arg
;
1357 int set_default
= 0;
1359 LASSERT(sizeof(lumv3
) == sizeof(*lumv3p
));
1360 LASSERT(sizeof(lumv3
.lmm_objects
[0]) ==
1361 sizeof(lumv3p
->lmm_objects
[0]));
1362 /* first try with v1 which is smaller than v3 */
1363 if (copy_from_user(lumv1
, lumv1p
, sizeof(*lumv1
)))
1366 if ((lumv1
->lmm_magic
== LOV_USER_MAGIC_V3
) ) {
1367 if (copy_from_user(&lumv3
, lumv3p
, sizeof(lumv3
)))
1371 if (inode
->i_sb
->s_root
== file
->f_dentry
)
1374 /* in v1 and v3 cases lumv1 points to data */
1375 rc
= ll_dir_setstripe(inode
, lumv1
, set_default
);
1379 case LL_IOC_LMV_GETSTRIPE
: {
1380 struct lmv_user_md
*lump
= (struct lmv_user_md
*)arg
;
1381 struct lmv_user_md lum
;
1382 struct lmv_user_md
*tmp
;
1387 if (copy_from_user(&lum
, lump
, sizeof(struct lmv_user_md
)))
1390 if (lum
.lum_magic
!= LMV_MAGIC_V1
)
1393 lum_size
= lmv_user_md_size(1, LMV_MAGIC_V1
);
1394 OBD_ALLOC(tmp
, lum_size
);
1396 GOTO(free_lmv
, rc
= -ENOMEM
);
1398 memcpy(tmp
, &lum
, sizeof(lum
));
1399 tmp
->lum_type
= LMV_STRIPE_TYPE
;
1400 tmp
->lum_stripe_count
= 1;
1401 mdtindex
= ll_get_mdt_idx(inode
);
1403 GOTO(free_lmv
, rc
= -ENOMEM
);
1405 tmp
->lum_stripe_offset
= mdtindex
;
1406 tmp
->lum_objects
[0].lum_mds
= mdtindex
;
1407 memcpy(&tmp
->lum_objects
[0].lum_fid
, ll_inode2fid(inode
),
1408 sizeof(struct lu_fid
));
1409 if (copy_to_user((void *)arg
, tmp
, lum_size
))
1410 GOTO(free_lmv
, rc
= -EFAULT
);
1413 OBD_FREE(tmp
, lum_size
);
1416 case LL_IOC_REMOVE_ENTRY
: {
1417 char *filename
= NULL
;
1421 /* Here is a little hack to avoid sending REINT_RMENTRY to
1422 * unsupported server, which might crash the server(LU-2730),
1423 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1424 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1425 * server will support REINT_RMENTRY XXX*/
1426 if (!(exp_connect_flags(sbi
->ll_md_exp
) & OBD_CONNECT_LVB_TYPE
))
1429 filename
= ll_getname((const char *)arg
);
1430 if (IS_ERR(filename
))
1431 return PTR_ERR(filename
);
1433 namelen
= strlen(filename
);
1435 GOTO(out_rmdir
, rc
= -EINVAL
);
1437 rc
= ll_rmdir_entry(inode
, filename
, namelen
);
1440 ll_putname(filename
);
1443 case LL_IOC_LOV_SWAP_LAYOUTS
:
1445 case LL_IOC_OBD_STATFS
:
1446 return ll_obd_statfs(inode
, (void *)arg
);
1447 case LL_IOC_LOV_GETSTRIPE
:
1448 case LL_IOC_MDC_GETINFO
:
1449 case IOC_MDC_GETFILEINFO
:
1450 case IOC_MDC_GETFILESTRIPE
: {
1451 struct ptlrpc_request
*request
= NULL
;
1452 struct lov_user_md
*lump
;
1453 struct lov_mds_md
*lmm
= NULL
;
1454 struct mdt_body
*body
;
1455 char *filename
= NULL
;
1458 if (cmd
== IOC_MDC_GETFILEINFO
||
1459 cmd
== IOC_MDC_GETFILESTRIPE
) {
1460 filename
= ll_getname((const char *)arg
);
1461 if (IS_ERR(filename
))
1462 return PTR_ERR(filename
);
1464 rc
= ll_lov_getstripe_ea_info(inode
, filename
, &lmm
,
1465 &lmmsize
, &request
);
1467 rc
= ll_dir_getstripe(inode
, &lmm
, &lmmsize
, &request
);
1471 body
= req_capsule_server_get(&request
->rq_pill
,
1473 LASSERT(body
!= NULL
);
1479 if (rc
== -ENODATA
&& (cmd
== IOC_MDC_GETFILEINFO
||
1480 cmd
== LL_IOC_MDC_GETINFO
))
1481 GOTO(skip_lmm
, rc
= 0);
1486 if (cmd
== IOC_MDC_GETFILESTRIPE
||
1487 cmd
== LL_IOC_LOV_GETSTRIPE
) {
1488 lump
= (struct lov_user_md
*)arg
;
1490 struct lov_user_mds_data
*lmdp
;
1491 lmdp
= (struct lov_user_mds_data
*)arg
;
1492 lump
= &lmdp
->lmd_lmm
;
1494 if (copy_to_user(lump
, lmm
, lmmsize
)) {
1495 if (copy_to_user(lump
, lmm
, sizeof(*lump
)))
1496 GOTO(out_req
, rc
= -EFAULT
);
1500 if (cmd
== IOC_MDC_GETFILEINFO
|| cmd
== LL_IOC_MDC_GETINFO
) {
1501 struct lov_user_mds_data
*lmdp
;
1504 st
.st_dev
= inode
->i_sb
->s_dev
;
1505 st
.st_mode
= body
->mode
;
1506 st
.st_nlink
= body
->nlink
;
1507 st
.st_uid
= body
->uid
;
1508 st
.st_gid
= body
->gid
;
1509 st
.st_rdev
= body
->rdev
;
1510 st
.st_size
= body
->size
;
1511 st
.st_blksize
= PAGE_CACHE_SIZE
;
1512 st
.st_blocks
= body
->blocks
;
1513 st
.st_atime
= body
->atime
;
1514 st
.st_mtime
= body
->mtime
;
1515 st
.st_ctime
= body
->ctime
;
1516 st
.st_ino
= inode
->i_ino
;
1518 lmdp
= (struct lov_user_mds_data
*)arg
;
1519 if (copy_to_user(&lmdp
->lmd_st
, &st
, sizeof(st
)))
1520 GOTO(out_req
, rc
= -EFAULT
);
1524 ptlrpc_req_finished(request
);
1526 ll_putname(filename
);
1529 case IOC_LOV_GETINFO
: {
1530 struct lov_user_mds_data
*lumd
;
1531 struct lov_stripe_md
*lsm
;
1532 struct lov_user_md
*lum
;
1533 struct lov_mds_md
*lmm
;
1537 lumd
= (struct lov_user_mds_data
*)arg
;
1538 lum
= &lumd
->lmd_lmm
;
1540 rc
= ll_get_max_mdsize(sbi
, &lmmsize
);
1544 OBD_ALLOC_LARGE(lmm
, lmmsize
);
1547 if (copy_from_user(lmm
, lum
, lmmsize
))
1548 GOTO(free_lmm
, rc
= -EFAULT
);
1550 switch (lmm
->lmm_magic
) {
1551 case LOV_USER_MAGIC_V1
:
1552 if (LOV_USER_MAGIC_V1
== cpu_to_le32(LOV_USER_MAGIC_V1
))
1554 /* swab objects first so that stripes num will be sane */
1555 lustre_swab_lov_user_md_objects(
1556 ((struct lov_user_md_v1
*)lmm
)->lmm_objects
,
1557 ((struct lov_user_md_v1
*)lmm
)->lmm_stripe_count
);
1558 lustre_swab_lov_user_md_v1((struct lov_user_md_v1
*)lmm
);
1560 case LOV_USER_MAGIC_V3
:
1561 if (LOV_USER_MAGIC_V3
== cpu_to_le32(LOV_USER_MAGIC_V3
))
1563 /* swab objects first so that stripes num will be sane */
1564 lustre_swab_lov_user_md_objects(
1565 ((struct lov_user_md_v3
*)lmm
)->lmm_objects
,
1566 ((struct lov_user_md_v3
*)lmm
)->lmm_stripe_count
);
1567 lustre_swab_lov_user_md_v3((struct lov_user_md_v3
*)lmm
);
1570 GOTO(free_lmm
, rc
= -EINVAL
);
1573 rc
= obd_unpackmd(sbi
->ll_dt_exp
, &lsm
, lmm
, lmmsize
);
1575 GOTO(free_lmm
, rc
= -ENOMEM
);
1577 /* Perform glimpse_size operation. */
1578 memset(&st
, 0, sizeof(st
));
1580 rc
= ll_glimpse_ioctl(sbi
, lsm
, &st
);
1584 if (copy_to_user(&lumd
->lmd_st
, &st
, sizeof(st
)))
1585 GOTO(free_lsm
, rc
= -EFAULT
);
1588 obd_free_memmd(sbi
->ll_dt_exp
, &lsm
);
1590 OBD_FREE_LARGE(lmm
, lmmsize
);
1593 case OBD_IOC_LLOG_CATINFO
: {
1596 case OBD_IOC_QUOTACHECK
: {
1597 struct obd_quotactl
*oqctl
;
1600 if (!cfs_capable(CFS_CAP_SYS_ADMIN
) ||
1601 sbi
->ll_flags
& LL_SBI_RMT_CLIENT
)
1604 OBD_ALLOC_PTR(oqctl
);
1607 oqctl
->qc_type
= arg
;
1608 rc
= obd_quotacheck(sbi
->ll_md_exp
, oqctl
);
1610 CDEBUG(D_INFO
, "md_quotacheck failed: rc %d\n", rc
);
1614 rc
= obd_quotacheck(sbi
->ll_dt_exp
, oqctl
);
1616 CDEBUG(D_INFO
, "obd_quotacheck failed: rc %d\n", rc
);
1618 OBD_FREE_PTR(oqctl
);
1621 case OBD_IOC_POLL_QUOTACHECK
: {
1622 struct if_quotacheck
*check
;
1624 if (!cfs_capable(CFS_CAP_SYS_ADMIN
) ||
1625 sbi
->ll_flags
& LL_SBI_RMT_CLIENT
)
1628 OBD_ALLOC_PTR(check
);
1632 rc
= obd_iocontrol(cmd
, sbi
->ll_md_exp
, 0, (void *)check
,
1635 CDEBUG(D_QUOTA
, "mdc ioctl %d failed: %d\n", cmd
, rc
);
1636 if (copy_to_user((void *)arg
, check
,
1638 CDEBUG(D_QUOTA
, "copy_to_user failed\n");
1642 rc
= obd_iocontrol(cmd
, sbi
->ll_dt_exp
, 0, (void *)check
,
1645 CDEBUG(D_QUOTA
, "osc ioctl %d failed: %d\n", cmd
, rc
);
1646 if (copy_to_user((void *)arg
, check
,
1648 CDEBUG(D_QUOTA
, "copy_to_user failed\n");
1652 OBD_FREE_PTR(check
);
1655 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1656 case LL_IOC_QUOTACTL_18
: {
1657 /* copy the old 1.x quota struct for internal use, then copy
1658 * back into old format struct. For 1.8 compatibility. */
1659 struct if_quotactl_18
*qctl_18
;
1660 struct if_quotactl
*qctl_20
;
1662 OBD_ALLOC_PTR(qctl_18
);
1666 OBD_ALLOC_PTR(qctl_20
);
1668 GOTO(out_quotactl_18
, rc
= -ENOMEM
);
1670 if (copy_from_user(qctl_18
, (void *)arg
, sizeof(*qctl_18
)))
1671 GOTO(out_quotactl_20
, rc
= -ENOMEM
);
1673 QCTL_COPY(qctl_20
, qctl_18
);
1674 qctl_20
->qc_idx
= 0;
1676 /* XXX: dqb_valid was borrowed as a flag to mark that
1677 * only mds quota is wanted */
1678 if (qctl_18
->qc_cmd
== Q_GETQUOTA
&&
1679 qctl_18
->qc_dqblk
.dqb_valid
) {
1680 qctl_20
->qc_valid
= QC_MDTIDX
;
1681 qctl_20
->qc_dqblk
.dqb_valid
= 0;
1682 } else if (qctl_18
->obd_uuid
.uuid
[0] != '\0') {
1683 qctl_20
->qc_valid
= QC_UUID
;
1684 qctl_20
->obd_uuid
= qctl_18
->obd_uuid
;
1686 qctl_20
->qc_valid
= QC_GENERAL
;
1689 rc
= quotactl_ioctl(sbi
, qctl_20
);
1692 QCTL_COPY(qctl_18
, qctl_20
);
1693 qctl_18
->obd_uuid
= qctl_20
->obd_uuid
;
1695 if (copy_to_user((void *)arg
, qctl_18
,
1701 OBD_FREE_PTR(qctl_20
);
1703 OBD_FREE_PTR(qctl_18
);
1707 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1708 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1709 case LL_IOC_QUOTACTL
: {
1710 struct if_quotactl
*qctl
;
1712 OBD_ALLOC_PTR(qctl
);
1716 if (copy_from_user(qctl
, (void *)arg
, sizeof(*qctl
)))
1717 GOTO(out_quotactl
, rc
= -EFAULT
);
1719 rc
= quotactl_ioctl(sbi
, qctl
);
1721 if (rc
== 0 && copy_to_user((void *)arg
,qctl
,sizeof(*qctl
)))
1728 case OBD_IOC_GETDTNAME
:
1729 case OBD_IOC_GETMDNAME
:
1730 return ll_get_obd_name(inode
, cmd
, arg
);
1731 case LL_IOC_FLUSHCTX
:
1732 return ll_flush_ctx(inode
);
1733 #ifdef CONFIG_FS_POSIX_ACL
1734 case LL_IOC_RMTACL
: {
1735 if (sbi
->ll_flags
& LL_SBI_RMT_CLIENT
&&
1736 inode
== inode
->i_sb
->s_root
->d_inode
) {
1737 struct ll_file_data
*fd
= LUSTRE_FPRIVATE(file
);
1739 LASSERT(fd
!= NULL
);
1740 rc
= rct_add(&sbi
->ll_rct
, current_pid(), arg
);
1742 fd
->fd_flags
|= LL_FILE_RMTACL
;
1748 case LL_IOC_GETOBDCOUNT
: {
1750 struct obd_export
*exp
;
1752 if (copy_from_user(&count
, (int *)arg
, sizeof(int)))
1755 /* get ost count when count is zero, get mdt count otherwise */
1756 exp
= count
? sbi
->ll_md_exp
: sbi
->ll_dt_exp
;
1757 vallen
= sizeof(count
);
1758 rc
= obd_get_info(NULL
, exp
, sizeof(KEY_TGT_COUNT
),
1759 KEY_TGT_COUNT
, &vallen
, &count
, NULL
);
1761 CERROR("get target count failed: %d\n", rc
);
1765 if (copy_to_user((int *)arg
, &count
, sizeof(int)))
1770 case LL_IOC_PATH2FID
:
1771 if (copy_to_user((void *)arg
, ll_inode2fid(inode
),
1772 sizeof(struct lu_fid
)))
1775 case LL_IOC_GET_CONNECT_FLAGS
: {
1776 return obd_iocontrol(cmd
, sbi
->ll_md_exp
, 0, NULL
, (void*)arg
);
1778 case OBD_IOC_CHANGELOG_SEND
:
1779 case OBD_IOC_CHANGELOG_CLEAR
:
1780 rc
= copy_and_ioctl(cmd
, sbi
->ll_md_exp
, (void *)arg
,
1781 sizeof(struct ioc_changelog
));
1783 case OBD_IOC_FID2PATH
:
1784 return ll_fid2path(inode
, (void *)arg
);
1785 case LL_IOC_HSM_REQUEST
: {
1786 struct hsm_user_request
*hur
;
1793 /* We don't know the true size yet; copy the fixed-size part */
1794 if (copy_from_user(hur
, (void *)arg
, sizeof(*hur
))) {
1799 /* Compute the whole struct size */
1800 totalsize
= hur_len(hur
);
1802 OBD_ALLOC_LARGE(hur
, totalsize
);
1806 /* Copy the whole struct */
1807 if (copy_from_user(hur
, (void *)arg
, totalsize
)) {
1808 OBD_FREE_LARGE(hur
, totalsize
);
1812 if (hur
->hur_request
.hr_action
== HUA_RELEASE
) {
1813 const struct lu_fid
*fid
;
1817 for (i
= 0; i
< hur
->hur_request
.hr_itemcount
; i
++) {
1818 fid
= &hur
->hur_user_item
[i
].hui_fid
;
1819 f
= search_inode_for_lustre(inode
->i_sb
, fid
);
1825 rc
= ll_hsm_release(f
);
1831 rc
= obd_iocontrol(cmd
, ll_i2mdexp(inode
), totalsize
,
1835 OBD_FREE_LARGE(hur
, totalsize
);
1839 case LL_IOC_HSM_PROGRESS
: {
1840 struct hsm_progress_kernel hpk
;
1841 struct hsm_progress hp
;
1843 if (copy_from_user(&hp
, (void *)arg
, sizeof(hp
)))
1846 hpk
.hpk_fid
= hp
.hp_fid
;
1847 hpk
.hpk_cookie
= hp
.hp_cookie
;
1848 hpk
.hpk_extent
= hp
.hp_extent
;
1849 hpk
.hpk_flags
= hp
.hp_flags
;
1850 hpk
.hpk_errval
= hp
.hp_errval
;
1851 hpk
.hpk_data_version
= 0;
1853 /* File may not exist in Lustre; all progress
1854 * reported to Lustre root */
1855 rc
= obd_iocontrol(cmd
, sbi
->ll_md_exp
, sizeof(hpk
), &hpk
,
1859 case LL_IOC_HSM_CT_START
:
1860 rc
= copy_and_ioctl(cmd
, sbi
->ll_md_exp
, (void *)arg
,
1861 sizeof(struct lustre_kernelcomm
));
1864 case LL_IOC_HSM_COPY_START
: {
1865 struct hsm_copy
*copy
;
1868 OBD_ALLOC_PTR(copy
);
1871 if (copy_from_user(copy
, (char *)arg
, sizeof(*copy
))) {
1876 rc
= ll_ioc_copy_start(inode
->i_sb
, copy
);
1877 if (copy_to_user((char *)arg
, copy
, sizeof(*copy
)))
1883 case LL_IOC_HSM_COPY_END
: {
1884 struct hsm_copy
*copy
;
1887 OBD_ALLOC_PTR(copy
);
1890 if (copy_from_user(copy
, (char *)arg
, sizeof(*copy
))) {
1895 rc
= ll_ioc_copy_end(inode
->i_sb
, copy
);
1896 if (copy_to_user((char *)arg
, copy
, sizeof(*copy
)))
1903 return obd_iocontrol(cmd
, sbi
->ll_dt_exp
, 0, NULL
, (void *)arg
);
1907 static loff_t
ll_dir_seek(struct file
*file
, loff_t offset
, int origin
)
1909 struct inode
*inode
= file
->f_mapping
->host
;
1910 struct ll_file_data
*fd
= LUSTRE_FPRIVATE(file
);
1911 struct ll_sb_info
*sbi
= ll_i2sbi(inode
);
1912 int api32
= ll_need_32bit_api(sbi
);
1913 loff_t ret
= -EINVAL
;
1915 mutex_lock(&inode
->i_mutex
);
1920 offset
+= file
->f_pos
;
1926 offset
+= LL_DIR_END_OFF_32BIT
;
1928 offset
+= LL_DIR_END_OFF
;
1935 ((api32
&& offset
<= LL_DIR_END_OFF_32BIT
) ||
1936 (!api32
&& offset
<= LL_DIR_END_OFF
))) {
1937 if (offset
!= file
->f_pos
) {
1938 if ((api32
&& offset
== LL_DIR_END_OFF_32BIT
) ||
1939 (!api32
&& offset
== LL_DIR_END_OFF
))
1940 fd
->lfd_pos
= MDS_DIR_END_OFF
;
1941 else if (api32
&& sbi
->ll_flags
& LL_SBI_64BIT_HASH
)
1942 fd
->lfd_pos
= offset
<< 32;
1944 fd
->lfd_pos
= offset
;
1945 file
->f_pos
= offset
;
1946 file
->f_version
= 0;
1953 mutex_unlock(&inode
->i_mutex
);
1957 int ll_dir_open(struct inode
*inode
, struct file
*file
)
1959 return ll_file_open(inode
, file
);
1962 int ll_dir_release(struct inode
*inode
, struct file
*file
)
1964 return ll_file_release(inode
, file
);
1967 struct file_operations ll_dir_operations
= {
1968 .llseek
= ll_dir_seek
,
1969 .open
= ll_dir_open
,
1970 .release
= ll_dir_release
,
1971 .read
= generic_read_dir
,
1972 .iterate
= ll_readdir
,
1973 .unlocked_ioctl
= ll_dir_ioctl
,