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 <linux/version.h>
45 #include <asm/uaccess.h>
46 #include <linux/buffer_head.h> // for wait_on_buffer
47 #include <linux/pagevec.h>
49 #define DEBUG_SUBSYSTEM S_LLITE
51 #include <lustre/lustre_idl.h>
52 #include <obd_support.h>
53 #include <obd_class.h>
54 #include <lustre_lib.h>
55 #include <lustre/lustre_idl.h>
56 #include <lustre_lite.h>
57 #include <lustre_dlm.h>
58 #include <lustre_fid.h>
59 #include "llite_internal.h"
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 CFS_PAGE_SIZE (if CFS_PAGE_SIZE greater than LU_PAGE_SIZE), and the
138 * lu_dirpage for this integrated page will be adjusted.
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 #ifndef HAVE_ADD_TO_PAGE_CACHE_LRU
155 struct pagevec lru_pvec;
157 struct lu_dirpage *dp;
158 int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> CFS_PAGE_SHIFT;
159 int nrdpgs = 0; /* number of pages read actually */
165 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash "LPU64"\n",
166 inode->i_ino, inode->i_generation, inode, hash);
168 LASSERT(max_pages > 0 && max_pages <= MD_MAX_BRW_PAGES);
170 OBD_ALLOC(page_pool, sizeof(page) * max_pages);
171 if (page_pool != NULL) {
172 page_pool[0] = page0;
177 for (npages = 1; npages < max_pages; npages++) {
178 page = page_cache_alloc_cold(inode->i_mapping);
181 page_pool[npages] = page;
184 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
185 LUSTRE_OPC_ANY, NULL);
186 op_data->op_npages = npages;
187 op_data->op_offset = hash;
188 rc = md_readpage(exp, op_data, page_pool, &request);
189 ll_finish_md_op_data(op_data);
191 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
192 /* Checked by mdc_readpage() */
193 LASSERT(body != NULL);
195 if (body->valid & OBD_MD_FLSIZE)
196 cl_isize_write(inode, body->size);
198 nrdpgs = (request->rq_bulk->bd_nob_transferred+CFS_PAGE_SIZE-1)
200 SetPageUptodate(page0);
203 ptlrpc_req_finished(request);
205 CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
207 ll_pagevec_init(&lru_pvec, 0);
208 for (i = 1; i < npages; i++) {
209 unsigned long offset;
214 if (rc < 0 || i >= nrdpgs) {
215 page_cache_release(page);
219 SetPageUptodate(page);
222 hash = le64_to_cpu(dp->ldp_hash_start);
225 offset = hash_x_index(hash, hash64);
227 prefetchw(&page->flags);
228 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
232 if (ll_pagevec_add(&lru_pvec, page) == 0)
233 ll_pagevec_lru_add_file(&lru_pvec);
235 CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
236 " %d\n", offset, ret);
238 page_cache_release(page);
240 ll_pagevec_lru_add_file(&lru_pvec);
242 if (page_pool != &page0)
243 OBD_FREE(page_pool, sizeof(struct page *) * max_pages);
248 static void ll_check_page(struct inode *dir, struct page *page)
250 /* XXX: check page format later */
251 SetPageChecked(page);
254 void ll_release_page(struct page *page, int remove)
259 if (likely(page->mapping != NULL))
260 truncate_complete_page(page->mapping, page);
263 page_cache_release(page);
267 * Find, kmap and return page that contains given hash.
269 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
270 __u64 *start, __u64 *end)
272 int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
273 struct address_space *mapping = dir->i_mapping;
275 * Complement of hash is used as an index so that
276 * radix_tree_gang_lookup() can be used to find a page with starting
277 * hash _smaller_ than one we are looking for.
279 unsigned long offset = hash_x_index(*hash, hash64);
283 TREE_READ_LOCK_IRQ(mapping);
284 found = radix_tree_gang_lookup(&mapping->page_tree,
285 (void **)&page, offset, 1);
287 struct lu_dirpage *dp;
289 page_cache_get(page);
290 TREE_READ_UNLOCK_IRQ(mapping);
292 * In contrast to find_lock_page() we are sure that directory
293 * page cannot be truncated (while DLM lock is held) and,
294 * hence, can avoid restart.
296 * In fact, page cannot be locked here at all, because
297 * ll_dir_filler() does synchronous io.
300 if (PageUptodate(page)) {
302 if (BITS_PER_LONG == 32 && hash64) {
303 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
304 *end = le64_to_cpu(dp->ldp_hash_end) >> 32;
307 *start = le64_to_cpu(dp->ldp_hash_start);
308 *end = le64_to_cpu(dp->ldp_hash_end);
310 LASSERTF(*start <= *hash, "start = "LPX64",end = "
311 LPX64",hash = "LPX64"\n", *start, *end, *hash);
312 CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash "LPU64"\n",
313 offset, *start, *end, *hash);
315 ll_release_page(page, 0);
317 } else if (*end != *start && *hash == *end) {
319 * upon hash collision, remove this page,
320 * otherwise put page reference, and
321 * ll_get_dir_page() will issue RPC to fetch
324 ll_release_page(page,
325 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
329 page_cache_release(page);
330 page = ERR_PTR(-EIO);
334 TREE_READ_UNLOCK_IRQ(mapping);
340 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
341 struct ll_dir_chain *chain)
343 ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
344 struct address_space *mapping = dir->i_mapping;
345 struct lustre_handle lockh;
346 struct lu_dirpage *dp;
353 struct ll_inode_info *lli = ll_i2info(dir);
354 int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
357 rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
358 ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
360 struct ldlm_enqueue_info einfo = {.ei_type = LDLM_IBITS,
369 struct lookup_intent it = { .it_op = IT_READDIR };
370 struct ptlrpc_request *request;
371 struct md_op_data *op_data;
373 op_data = ll_prep_md_op_data(NULL, dir, NULL, NULL, 0, 0,
374 LUSTRE_OPC_ANY, NULL);
376 return (void *)op_data;
378 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
379 op_data, &lockh, NULL, 0, NULL, 0);
381 ll_finish_md_op_data(op_data);
383 request = (struct ptlrpc_request *)it.d.lustre.it_data;
385 ptlrpc_req_finished(request);
387 CERROR("lock enqueue: "DFID" at "LPU64": rc %d\n",
388 PFID(ll_inode2fid(dir)), hash, rc);
392 CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
393 dir, dir->i_ino, dir->i_generation);
394 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
395 &it.d.lustre.it_lock_handle, dir, NULL);
397 /* for cross-ref object, l_ast_data of the lock may not be set,
398 * we reset it here */
399 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
402 ldlm_lock_dump_handle(D_OTHER, &lockh);
404 mutex_lock(&lli->lli_readdir_mutex);
405 page = ll_dir_page_locate(dir, &lhash, &start, &end);
407 CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
408 PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
409 GOTO(out_unlock, page);
410 } else if (page != NULL) {
412 * XXX nikita: not entirely correct handling of a corner case:
413 * suppose hash chain of entries with hash value HASH crosses
414 * border between pages P0 and P1. First both P0 and P1 are
415 * cached, seekdir() is called for some entry from the P0 part
416 * of the chain. Later P0 goes out of cache. telldir(HASH)
417 * happens and finds P1, as it starts with matching hash
418 * value. Remaining entries from P0 part of the chain are
419 * skipped. (Is that really a bug?)
421 * Possible solutions: 0. don't cache P1 is such case, handle
422 * it as an "overflow" page. 1. invalidate all pages at
423 * once. 2. use HASH|1 as an index for P1.
425 GOTO(hash_collision, page);
428 page = read_cache_page(mapping, hash_x_index(hash, hash64),
429 ll_dir_filler, &lhash);
431 CERROR("read cache page: "DFID" at "LPU64": rc %ld\n",
432 PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
433 GOTO(out_unlock, page);
438 if (!PageUptodate(page)) {
439 CERROR("page not updated: "DFID" at "LPU64": rc %d\n",
440 PFID(ll_inode2fid(dir)), hash, -5);
443 if (!PageChecked(page))
444 ll_check_page(dir, page);
445 if (PageError(page)) {
446 CERROR("page error: "DFID" at "LPU64": rc %d\n",
447 PFID(ll_inode2fid(dir)), hash, -5);
451 dp = page_address(page);
452 if (BITS_PER_LONG == 32 && hash64) {
453 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
454 end = le64_to_cpu(dp->ldp_hash_end) >> 32;
457 start = le64_to_cpu(dp->ldp_hash_start);
458 end = le64_to_cpu(dp->ldp_hash_end);
462 LASSERT(start == lhash);
463 CWARN("Page-wide hash collision: "LPU64"\n", end);
464 if (BITS_PER_LONG == 32 && hash64)
465 CWARN("Real page-wide hash collision at ["LPU64" "LPU64
466 "] with hash "LPU64"\n",
467 le64_to_cpu(dp->ldp_hash_start),
468 le64_to_cpu(dp->ldp_hash_end), hash);
470 * Fetch whole overflow chain...
477 mutex_unlock(&lli->lli_readdir_mutex);
478 ldlm_lock_decref(&lockh, mode);
482 ll_release_page(page, 1);
483 page = ERR_PTR(-EIO);
487 int ll_dir_read(struct inode *inode, __u64 *_pos, void *cookie,
490 struct ll_inode_info *info = ll_i2info(inode);
491 struct ll_sb_info *sbi = ll_i2sbi(inode);
493 int api32 = ll_need_32bit_api(sbi);
494 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
496 struct ll_dir_chain chain;
501 ll_dir_chain_init(&chain);
503 page = ll_get_dir_page(inode, pos, &chain);
505 while (rc == 0 && !done) {
506 struct lu_dirpage *dp;
507 struct lu_dirent *ent;
511 * If page is empty (end of directory is reached),
514 __u64 hash = MDS_DIR_END_OFF;
517 dp = page_address(page);
518 for (ent = lu_dirent_start(dp); ent != NULL && !done;
519 ent = lu_dirent_next(ent)) {
527 * XXX: implement correct swabbing here.
530 hash = le64_to_cpu(ent->lde_hash);
533 * Skip until we find target hash
538 namelen = le16_to_cpu(ent->lde_namelen);
549 fid_le_to_cpu(&fid, &ent->lde_fid);
550 ino = cl_fid_build_ino(&fid, api32);
551 type = ll_dirent_type_get(ent);
552 /* For 'll_nfs_get_name_filldir()', it will try
553 * to access the 'ent' through its 'lde_name',
554 * so the parameter 'name' for 'filldir()' must
555 * be part of the 'ent'. */
556 done = filldir(cookie, ent->lde_name, namelen,
559 next = le64_to_cpu(dp->ldp_hash_end);
562 if (pos == MDS_DIR_END_OFF) {
564 * End of directory reached.
567 ll_release_page(page, 0);
568 } else if (1 /* chain is exhausted*/) {
570 * Normal case: continue to the next
573 ll_release_page(page,
574 le32_to_cpu(dp->ldp_flags) &
577 page = ll_get_dir_page(inode, pos,
581 * go into overflow page.
583 LASSERT(le32_to_cpu(dp->ldp_flags) &
585 ll_release_page(page, 1);
589 ll_release_page(page, 0);
593 CERROR("error reading dir "DFID" at %lu: rc %d\n",
594 PFID(&info->lli_fid), (unsigned long)pos, rc);
599 ll_dir_chain_fini(&chain);
603 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
605 struct inode *inode = filp->f_dentry->d_inode;
606 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
607 struct ll_sb_info *sbi = ll_i2sbi(inode);
608 __u64 pos = lfd->lfd_pos;
609 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
610 int api32 = ll_need_32bit_api(sbi);
612 #ifdef HAVE_TOUCH_ATIME_1ARG
617 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu "
618 " 32bit_api %d\n", inode->i_ino, inode->i_generation,
619 inode, (unsigned long)pos, i_size_read(inode), api32);
621 if (pos == MDS_DIR_END_OFF)
627 rc = ll_dir_read(inode, &pos, cookie, filldir);
629 if (pos == MDS_DIR_END_OFF) {
631 filp->f_pos = LL_DIR_END_OFF_32BIT;
633 filp->f_pos = LL_DIR_END_OFF;
636 filp->f_pos = pos >> 32;
640 filp->f_version = inode->i_version;
641 #ifdef HAVE_TOUCH_ATIME_1ARG
642 path.mnt = filp->f_vfsmnt;
643 path.dentry = filp->f_dentry;
646 touch_atime(filp->f_vfsmnt, filp->f_dentry);
651 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
656 int ll_send_mgc_param(struct obd_export *mgc, char *string)
658 struct mgs_send_param *msp;
665 strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
666 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
667 sizeof(struct mgs_send_param), msp, NULL);
669 CERROR("Failed to set parameter: %d\n", rc);
675 int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
678 struct ptlrpc_request *request = NULL;
679 struct md_op_data *op_data;
680 struct ll_sb_info *sbi = ll_i2sbi(dir);
686 mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
687 op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
688 strlen(filename), mode, LUSTRE_OPC_MKDIR,
691 GOTO(err_exit, err = PTR_ERR(op_data));
693 op_data->op_cli_flags |= CLI_SET_MEA;
694 err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
695 cfs_curproc_fsuid(), cfs_curproc_fsgid(),
696 cfs_curproc_cap_pack(), 0, &request);
697 ll_finish_md_op_data(op_data);
701 ptlrpc_req_finished(request);
705 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
708 struct ll_sb_info *sbi = ll_i2sbi(inode);
709 struct md_op_data *op_data;
710 struct ptlrpc_request *req = NULL;
712 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
713 struct obd_device *mgc = lsi->lsi_mgc;
719 * This is coming from userspace, so should be in
720 * local endian. But the MDS would like it in little
721 * endian, so we swab it before we send it.
723 switch (lump->lmm_magic) {
724 case LOV_USER_MAGIC_V1: {
725 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
726 lustre_swab_lov_user_md_v1(lump);
727 lum_size = sizeof(struct lov_user_md_v1);
730 case LOV_USER_MAGIC_V3: {
731 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
732 lustre_swab_lov_user_md_v3(
733 (struct lov_user_md_v3 *)lump);
734 lum_size = sizeof(struct lov_user_md_v3);
738 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
739 " %#08x != %#08x nor %#08x\n",
740 lump->lmm_magic, LOV_USER_MAGIC_V1,
746 lum_size = sizeof(struct lov_user_md_v1);
749 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
750 LUSTRE_OPC_ANY, NULL);
752 RETURN(PTR_ERR(op_data));
754 if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
755 op_data->op_cli_flags |= CLI_SET_MEA;
757 /* swabbing is done in lov_setstripe() on server side */
758 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
759 NULL, 0, &req, NULL);
760 ll_finish_md_op_data(op_data);
761 ptlrpc_req_finished(req);
763 if (rc != -EPERM && rc != -EACCES)
764 CERROR("mdc_setattr fails: rc = %d\n", rc);
767 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
768 LOV_USER_MAGIC_V3 have the same initial fields so we do not
769 need the make the distiction between the 2 versions */
770 if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
774 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
776 GOTO(end, rc = -ENOMEM);
779 /* Get fsname and assume devname to be -MDT0000. */
780 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
781 strcat(buf, "-MDT0000.lov");
784 /* Set root stripesize */
785 sprintf(buf, ".stripesize=%u",
786 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
787 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
791 /* Set root stripecount */
792 sprintf(buf, ".stripecount=%hd",
793 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
794 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
798 /* Set root stripeoffset */
799 sprintf(buf, ".stripeoffset=%hd",
800 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
801 (typeof(lump->lmm_stripe_offset))(-1));
802 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
806 OBD_FREE(param, MGS_PARAM_MAXLEN);
811 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
812 int *lmm_size, struct ptlrpc_request **request)
814 struct ll_sb_info *sbi = ll_i2sbi(inode);
815 struct mdt_body *body;
816 struct lov_mds_md *lmm = NULL;
817 struct ptlrpc_request *req = NULL;
819 struct md_op_data *op_data;
821 rc = ll_get_max_mdsize(sbi, &lmmsize);
825 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
826 0, lmmsize, LUSTRE_OPC_ANY,
829 RETURN(PTR_ERR(op_data));
831 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
832 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
833 ll_finish_md_op_data(op_data);
835 CDEBUG(D_INFO, "md_getattr failed on inode "
836 "%lu/%u: rc %d\n", inode->i_ino,
837 inode->i_generation, rc);
841 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
842 LASSERT(body != NULL);
844 lmmsize = body->eadatasize;
846 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
848 GOTO(out, rc = -ENODATA);
851 lmm = req_capsule_server_sized_get(&req->rq_pill,
852 &RMF_MDT_MD, lmmsize);
853 LASSERT(lmm != NULL);
856 * This is coming from the MDS, so is probably in
857 * little endian. We convert it to host endian before
858 * passing it to userspace.
860 /* We don't swab objects for directories */
861 switch (le32_to_cpu(lmm->lmm_magic)) {
863 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
864 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
867 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
868 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
871 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
882 * Get MDT index for the inode.
884 int ll_get_mdt_idx(struct inode *inode)
886 struct ll_sb_info *sbi = ll_i2sbi(inode);
887 struct md_op_data *op_data;
891 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
892 0, LUSTRE_OPC_ANY, NULL);
894 RETURN(PTR_ERR(op_data));
896 op_data->op_flags |= MF_GET_MDT_IDX;
897 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
898 mdtidx = op_data->op_mds;
899 ll_finish_md_op_data(op_data);
901 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
908 * Generic handler to do any pre-copy work.
910 * It send a first hsm_progress (with extent length == 0) to coordinator as a
911 * first information for it that real work has started.
913 * Moreover, for a ARCHIVE request, it will sample the file data version and
914 * store it in \a copy.
916 * \return 0 on success.
918 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
920 struct ll_sb_info *sbi = ll_s2sbi(sb);
921 struct hsm_progress_kernel hpk;
925 /* Forge a hsm_progress based on data from copy. */
926 hpk.hpk_fid = copy->hc_hai.hai_fid;
927 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
928 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
929 hpk.hpk_extent.length = 0;
932 hpk.hpk_data_version = 0;
935 /* For archive request, we need to read the current file version. */
936 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
938 __u64 data_version = 0;
940 /* Get inode for this fid */
941 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
943 hpk.hpk_flags |= HP_FLAG_RETRY;
944 /* hpk_errval is >= 0 */
945 hpk.hpk_errval = -PTR_ERR(inode);
946 GOTO(progress, rc = PTR_ERR(inode));
949 /* Read current file data version */
950 rc = ll_data_version(inode, &data_version, 1);
953 CDEBUG(D_HSM, "Could not read file data version of "
954 DFID" (rc = %d). Archive request ("
955 LPX64") could not be done.\n",
956 PFID(©->hc_hai.hai_fid), rc,
957 copy->hc_hai.hai_cookie);
958 hpk.hpk_flags |= HP_FLAG_RETRY;
959 /* hpk_errval must be >= 0 */
960 hpk.hpk_errval = -rc;
964 /* Store it the hsm_copy for later copytool use.
965 * Always modified even if no lsm. */
966 copy->hc_data_version = data_version;
970 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
977 * Generic handler to do any post-copy work.
979 * It will send the last hsm_progress update to coordinator to inform it
980 * that copy is finished and whether it was successful or not.
983 * - for ARCHIVE request, it will sample the file data version and compare it
984 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
985 * will be considered as failed.
986 * - for RESTORE request, it will sample the file data version and send it to
987 * coordinator which is useful if the file was imported as 'released'.
989 * \return 0 on success.
991 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
993 struct ll_sb_info *sbi = ll_s2sbi(sb);
994 struct hsm_progress_kernel hpk;
998 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
999 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
1000 * initialized if copy_end was called with copy == NULL.
1003 /* Forge a hsm_progress based on data from copy. */
1004 hpk.hpk_fid = copy->hc_hai.hai_fid;
1005 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
1006 hpk.hpk_extent = copy->hc_hai.hai_extent;
1007 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
1008 hpk.hpk_errval = copy->hc_errval;
1009 hpk.hpk_data_version = 0;
1011 /* For archive request, we need to check the file data was not changed.
1013 * For restore request, we need to send the file data version, this is
1014 * useful when the file was created using hsm_import.
1016 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
1017 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
1018 (copy->hc_errval == 0)) {
1019 struct inode *inode;
1020 __u64 data_version = 0;
1022 /* Get lsm for this fid */
1023 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
1024 if (IS_ERR(inode)) {
1025 hpk.hpk_flags |= HP_FLAG_RETRY;
1026 /* hpk_errval must be >= 0 */
1027 hpk.hpk_errval = -PTR_ERR(inode);
1028 GOTO(progress, rc = PTR_ERR(inode));
1031 rc = ll_data_version(inode, &data_version,
1032 copy->hc_hai.hai_action == HSMA_ARCHIVE);
1035 CDEBUG(D_HSM, "Could not read file data version. "
1036 "Request could not be confirmed.\n");
1037 if (hpk.hpk_errval == 0)
1038 hpk.hpk_errval = -rc;
1042 /* Store it the hsm_copy for later copytool use.
1043 * Always modified even if no lsm. */
1044 hpk.hpk_data_version = data_version;
1046 /* File could have been stripped during archiving, so we need
1047 * to check anyway. */
1048 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1049 (copy->hc_data_version != data_version)) {
1050 CDEBUG(D_HSM, "File data version mismatched. "
1051 "File content was changed during archiving. "
1052 DFID", start:"LPX64" current:"LPX64"\n",
1053 PFID(©->hc_hai.hai_fid),
1054 copy->hc_data_version, data_version);
1055 /* File was changed, send error to cdt. Do not ask for
1056 * retry because if a file is modified frequently,
1057 * the cdt will loop on retried archive requests.
1058 * The policy engine will ask for a new archive later
1059 * when the file will not be modified for some tunable
1061 /* we do not notify caller */
1062 hpk.hpk_flags &= ~HP_FLAG_RETRY;
1063 /* hpk_errval must be >= 0 */
1064 hpk.hpk_errval = EBUSY;
1070 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1077 static int copy_and_ioctl(int cmd, struct obd_export *exp, void *data, int len)
1082 OBD_ALLOC(ptr, len);
1085 if (cfs_copy_from_user(ptr, data, len)) {
1089 rc = obd_iocontrol(cmd, exp, len, data, NULL);
1094 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1096 int cmd = qctl->qc_cmd;
1097 int type = qctl->qc_type;
1098 int id = qctl->qc_id;
1099 int valid = qctl->qc_valid;
1104 case LUSTRE_Q_INVALIDATE:
1105 case LUSTRE_Q_FINVALIDATE:
1110 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1111 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1115 if (((type == USRQUOTA && cfs_curproc_euid() != id) ||
1116 (type == GRPQUOTA && !in_egroup_p(id))) &&
1117 (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1118 sbi->ll_flags & LL_SBI_RMT_CLIENT))
1124 CERROR("unsupported quotactl op: %#x\n", cmd);
1128 if (valid != QC_GENERAL) {
1129 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1130 RETURN(-EOPNOTSUPP);
1132 if (cmd == Q_GETINFO)
1133 qctl->qc_cmd = Q_GETOINFO;
1134 else if (cmd == Q_GETQUOTA)
1135 qctl->qc_cmd = Q_GETOQUOTA;
1141 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1142 sizeof(*qctl), qctl, NULL);
1145 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1146 sizeof(*qctl), qctl, NULL);
1149 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1150 sizeof(*qctl), qctl, NULL);
1152 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1154 sizeof(*qctl), qctl, NULL);
1166 struct obd_quotactl *oqctl;
1168 OBD_ALLOC_PTR(oqctl);
1172 QCTL_COPY(oqctl, qctl);
1173 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1175 if (rc != -EALREADY && cmd == Q_QUOTAON) {
1176 oqctl->qc_cmd = Q_QUOTAOFF;
1177 obd_quotactl(sbi->ll_md_exp, oqctl);
1179 OBD_FREE_PTR(oqctl);
1182 /* If QIF_SPACE is not set, client should collect the
1183 * space usage from OSSs by itself */
1184 if (cmd == Q_GETQUOTA &&
1185 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1186 !oqctl->qc_dqblk.dqb_curspace) {
1187 struct obd_quotactl *oqctl_tmp;
1189 OBD_ALLOC_PTR(oqctl_tmp);
1190 if (oqctl_tmp == NULL)
1191 GOTO(out, rc = -ENOMEM);
1193 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1194 oqctl_tmp->qc_id = oqctl->qc_id;
1195 oqctl_tmp->qc_type = oqctl->qc_type;
1197 /* collect space usage from OSTs */
1198 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1199 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1200 if (!rc || rc == -EREMOTEIO) {
1201 oqctl->qc_dqblk.dqb_curspace =
1202 oqctl_tmp->qc_dqblk.dqb_curspace;
1203 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1206 /* collect space & inode usage from MDTs */
1207 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1208 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1209 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1210 if (!rc || rc == -EREMOTEIO) {
1211 oqctl->qc_dqblk.dqb_curspace +=
1212 oqctl_tmp->qc_dqblk.dqb_curspace;
1213 oqctl->qc_dqblk.dqb_curinodes =
1214 oqctl_tmp->qc_dqblk.dqb_curinodes;
1215 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1217 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1220 OBD_FREE_PTR(oqctl_tmp);
1223 QCTL_COPY(qctl, oqctl);
1224 OBD_FREE_PTR(oqctl);
1230 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1232 struct inode *inode = file->f_dentry->d_inode;
1233 struct ll_sb_info *sbi = ll_i2sbi(inode);
1234 struct obd_ioctl_data *data;
1238 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1239 inode->i_ino, inode->i_generation, inode, cmd);
1241 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1242 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1245 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1247 case FSFILT_IOC_GETFLAGS:
1248 case FSFILT_IOC_SETFLAGS:
1249 RETURN(ll_iocontrol(inode, file, cmd, arg));
1250 case FSFILT_IOC_GETVERSION_OLD:
1251 case FSFILT_IOC_GETVERSION:
1252 RETURN(put_user(inode->i_generation, (int *)arg));
1253 /* We need to special case any other ioctls we want to handle,
1254 * to send them to the MDS/OST as appropriate and to properly
1255 * network encode the arg field.
1256 case FSFILT_IOC_SETVERSION_OLD:
1257 case FSFILT_IOC_SETVERSION:
1259 case LL_IOC_GET_MDTIDX: {
1262 mdtidx = ll_get_mdt_idx(inode);
1266 if (put_user((int)mdtidx, (int*)arg))
1271 case IOC_MDC_LOOKUP: {
1272 struct ptlrpc_request *request = NULL;
1273 int namelen, len = 0;
1276 struct md_op_data *op_data;
1278 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1283 filename = data->ioc_inlbuf1;
1284 namelen = strlen(filename);
1287 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1288 GOTO(out_free, rc = -EINVAL);
1291 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1292 0, LUSTRE_OPC_ANY, NULL);
1293 if (IS_ERR(op_data))
1294 GOTO(out_free, rc = PTR_ERR(op_data));
1296 op_data->op_valid = OBD_MD_FLID;
1297 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1298 ll_finish_md_op_data(op_data);
1300 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1303 ptlrpc_req_finished(request);
1306 obd_ioctl_freedata(buf, len);
1309 case LL_IOC_LMV_SETSTRIPE: {
1310 struct lmv_user_md *lum;
1318 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1323 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1324 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1325 GOTO(lmv_out_free, rc = -EINVAL);
1327 filename = data->ioc_inlbuf1;
1328 namelen = data->ioc_inllen1;
1331 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1332 GOTO(lmv_out_free, rc = -EINVAL);
1334 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1335 lumlen = data->ioc_inllen2;
1337 if (lum->lum_magic != LMV_USER_MAGIC ||
1338 lumlen != sizeof(*lum)) {
1339 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1340 filename, lum->lum_magic, lumlen, -EFAULT);
1341 GOTO(lmv_out_free, rc = -EINVAL);
1345 * ll_dir_setdirstripe will be used to set dir stripe
1346 * mdc_create--->mdt_reint_create (with dirstripe)
1348 rc = ll_dir_setdirstripe(inode, lum, filename);
1350 obd_ioctl_freedata(buf, len);
1354 case LL_IOC_LOV_SETSTRIPE: {
1355 struct lov_user_md_v3 lumv3;
1356 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1357 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1358 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1360 int set_default = 0;
1362 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1363 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1364 sizeof(lumv3p->lmm_objects[0]));
1365 /* first try with v1 which is smaller than v3 */
1366 if (cfs_copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1369 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1370 if (cfs_copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1374 if (inode->i_sb->s_root == file->f_dentry)
1377 /* in v1 and v3 cases lumv1 points to data */
1378 rc = ll_dir_setstripe(inode, lumv1, set_default);
1382 case LL_IOC_LMV_GETSTRIPE: {
1383 struct lmv_user_md *lump = (struct lmv_user_md *)arg;
1384 struct lmv_user_md lum;
1385 struct lmv_user_md *tmp;
1390 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1393 if (lum.lum_magic != LMV_MAGIC_V1)
1396 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1397 OBD_ALLOC(tmp, lum_size);
1399 GOTO(free_lmv, rc = -ENOMEM);
1401 memcpy(tmp, &lum, sizeof(lum));
1402 tmp->lum_type = LMV_STRIPE_TYPE;
1403 tmp->lum_stripe_count = 1;
1404 mdtindex = ll_get_mdt_idx(inode);
1406 GOTO(free_lmv, rc = -ENOMEM);
1408 tmp->lum_stripe_offset = mdtindex;
1409 tmp->lum_objects[0].lum_mds = mdtindex;
1410 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1411 sizeof(struct lu_fid));
1412 if (copy_to_user((void *)arg, tmp, lum_size))
1413 GOTO(free_lmv, rc = -EFAULT);
1416 OBD_FREE(tmp, lum_size);
1419 case LL_IOC_REMOVE_ENTRY: {
1420 char *filename = NULL;
1424 /* Here is a little hack to avoid sending REINT_RMENTRY to
1425 * unsupported server, which might crash the server(LU-2730),
1426 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1427 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1428 * server will support REINT_RMENTRY XXX*/
1429 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1432 filename = getname((const char *)arg);
1433 if (IS_ERR(filename))
1434 RETURN(PTR_ERR(filename));
1436 namelen = strlen(filename);
1438 GOTO(out_rmdir, rc = -EINVAL);
1440 rc = ll_rmdir_entry(inode, filename, namelen);
1446 case LL_IOC_LOV_SWAP_LAYOUTS:
1448 case LL_IOC_OBD_STATFS:
1449 RETURN(ll_obd_statfs(inode, (void *)arg));
1450 case LL_IOC_LOV_GETSTRIPE:
1451 case LL_IOC_MDC_GETINFO:
1452 case IOC_MDC_GETFILEINFO:
1453 case IOC_MDC_GETFILESTRIPE: {
1454 struct ptlrpc_request *request = NULL;
1455 struct lov_user_md *lump;
1456 struct lov_mds_md *lmm = NULL;
1457 struct mdt_body *body;
1458 char *filename = NULL;
1461 if (cmd == IOC_MDC_GETFILEINFO ||
1462 cmd == IOC_MDC_GETFILESTRIPE) {
1463 filename = getname((const char *)arg);
1464 if (IS_ERR(filename))
1465 RETURN(PTR_ERR(filename));
1467 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1468 &lmmsize, &request);
1470 rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1474 body = req_capsule_server_get(&request->rq_pill,
1476 LASSERT(body != NULL);
1482 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1483 cmd == LL_IOC_MDC_GETINFO))
1484 GOTO(skip_lmm, rc = 0);
1489 if (cmd == IOC_MDC_GETFILESTRIPE ||
1490 cmd == LL_IOC_LOV_GETSTRIPE) {
1491 lump = (struct lov_user_md *)arg;
1493 struct lov_user_mds_data *lmdp;
1494 lmdp = (struct lov_user_mds_data *)arg;
1495 lump = &lmdp->lmd_lmm;
1497 if (cfs_copy_to_user(lump, lmm, lmmsize)) {
1498 if (cfs_copy_to_user(lump, lmm, sizeof(*lump)))
1499 GOTO(out_req, rc = -EFAULT);
1503 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1504 struct lov_user_mds_data *lmdp;
1507 st.st_dev = inode->i_sb->s_dev;
1508 st.st_mode = body->mode;
1509 st.st_nlink = body->nlink;
1510 st.st_uid = body->uid;
1511 st.st_gid = body->gid;
1512 st.st_rdev = body->rdev;
1513 st.st_size = body->size;
1514 st.st_blksize = CFS_PAGE_SIZE;
1515 st.st_blocks = body->blocks;
1516 st.st_atime = body->atime;
1517 st.st_mtime = body->mtime;
1518 st.st_ctime = body->ctime;
1519 st.st_ino = inode->i_ino;
1521 lmdp = (struct lov_user_mds_data *)arg;
1522 if (cfs_copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1523 GOTO(out_req, rc = -EFAULT);
1528 ptlrpc_req_finished(request);
1533 case IOC_LOV_GETINFO: {
1534 struct lov_user_mds_data *lumd;
1535 struct lov_stripe_md *lsm;
1536 struct lov_user_md *lum;
1537 struct lov_mds_md *lmm;
1541 lumd = (struct lov_user_mds_data *)arg;
1542 lum = &lumd->lmd_lmm;
1544 rc = ll_get_max_mdsize(sbi, &lmmsize);
1548 OBD_ALLOC_LARGE(lmm, lmmsize);
1549 if (cfs_copy_from_user(lmm, lum, lmmsize))
1550 GOTO(free_lmm, rc = -EFAULT);
1552 switch (lmm->lmm_magic) {
1553 case LOV_USER_MAGIC_V1:
1554 if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1556 /* swab objects first so that stripes num will be sane */
1557 lustre_swab_lov_user_md_objects(
1558 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1559 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1560 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1562 case LOV_USER_MAGIC_V3:
1563 if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1565 /* swab objects first so that stripes num will be sane */
1566 lustre_swab_lov_user_md_objects(
1567 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1568 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1569 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1572 GOTO(free_lmm, rc = -EINVAL);
1575 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1577 GOTO(free_lmm, rc = -ENOMEM);
1579 /* Perform glimpse_size operation. */
1580 memset(&st, 0, sizeof(st));
1582 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1586 if (cfs_copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1587 GOTO(free_lsm, rc = -EFAULT);
1591 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1593 OBD_FREE_LARGE(lmm, lmmsize);
1596 case OBD_IOC_LLOG_CATINFO: {
1597 RETURN(-EOPNOTSUPP);
1599 case OBD_IOC_QUOTACHECK: {
1600 struct obd_quotactl *oqctl;
1603 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1604 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1607 OBD_ALLOC_PTR(oqctl);
1610 oqctl->qc_type = arg;
1611 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1613 CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1617 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1619 CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1621 OBD_FREE_PTR(oqctl);
1624 case OBD_IOC_POLL_QUOTACHECK: {
1625 struct if_quotacheck *check;
1627 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1628 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1631 OBD_ALLOC_PTR(check);
1635 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1638 CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1639 if (cfs_copy_to_user((void *)arg, check,
1641 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1645 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1648 CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1649 if (cfs_copy_to_user((void *)arg, check,
1651 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1655 OBD_FREE_PTR(check);
1658 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1659 case LL_IOC_QUOTACTL_18: {
1660 /* copy the old 1.x quota struct for internal use, then copy
1661 * back into old format struct. For 1.8 compatibility. */
1662 struct if_quotactl_18 *qctl_18;
1663 struct if_quotactl *qctl_20;
1665 OBD_ALLOC_PTR(qctl_18);
1669 OBD_ALLOC_PTR(qctl_20);
1671 GOTO(out_quotactl_18, rc = -ENOMEM);
1673 if (cfs_copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1674 GOTO(out_quotactl_20, rc = -ENOMEM);
1676 QCTL_COPY(qctl_20, qctl_18);
1677 qctl_20->qc_idx = 0;
1679 /* XXX: dqb_valid was borrowed as a flag to mark that
1680 * only mds quota is wanted */
1681 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1682 qctl_18->qc_dqblk.dqb_valid) {
1683 qctl_20->qc_valid = QC_MDTIDX;
1684 qctl_20->qc_dqblk.dqb_valid = 0;
1685 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1686 qctl_20->qc_valid = QC_UUID;
1687 qctl_20->obd_uuid = qctl_18->obd_uuid;
1689 qctl_20->qc_valid = QC_GENERAL;
1692 rc = quotactl_ioctl(sbi, qctl_20);
1695 QCTL_COPY(qctl_18, qctl_20);
1696 qctl_18->obd_uuid = qctl_20->obd_uuid;
1698 if (cfs_copy_to_user((void *)arg, qctl_18,
1704 OBD_FREE_PTR(qctl_20);
1706 OBD_FREE_PTR(qctl_18);
1710 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1711 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1712 case LL_IOC_QUOTACTL: {
1713 struct if_quotactl *qctl;
1715 OBD_ALLOC_PTR(qctl);
1719 if (cfs_copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1720 GOTO(out_quotactl, rc = -EFAULT);
1722 rc = quotactl_ioctl(sbi, qctl);
1724 if (rc == 0 && cfs_copy_to_user((void *)arg,qctl,sizeof(*qctl)))
1731 case OBD_IOC_GETDTNAME:
1732 case OBD_IOC_GETMDNAME:
1733 RETURN(ll_get_obd_name(inode, cmd, arg));
1734 case LL_IOC_FLUSHCTX:
1735 RETURN(ll_flush_ctx(inode));
1736 #ifdef CONFIG_FS_POSIX_ACL
1737 case LL_IOC_RMTACL: {
1738 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1739 inode == inode->i_sb->s_root->d_inode) {
1740 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1742 LASSERT(fd != NULL);
1743 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1745 fd->fd_flags |= LL_FILE_RMTACL;
1751 case LL_IOC_GETOBDCOUNT: {
1753 struct obd_export *exp;
1755 if (cfs_copy_from_user(&count, (int *)arg, sizeof(int)))
1758 /* get ost count when count is zero, get mdt count otherwise */
1759 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1760 vallen = sizeof(count);
1761 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1762 KEY_TGT_COUNT, &vallen, &count, NULL);
1764 CERROR("get target count failed: %d\n", rc);
1768 if (cfs_copy_to_user((int *)arg, &count, sizeof(int)))
1773 case LL_IOC_PATH2FID:
1774 if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
1775 sizeof(struct lu_fid)))
1778 case LL_IOC_GET_CONNECT_FLAGS: {
1779 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1781 case OBD_IOC_CHANGELOG_SEND:
1782 case OBD_IOC_CHANGELOG_CLEAR:
1783 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1784 sizeof(struct ioc_changelog));
1786 case OBD_IOC_FID2PATH:
1787 RETURN(ll_fid2path(inode, (void *)arg));
1788 case LL_IOC_HSM_REQUEST: {
1789 struct hsm_user_request *hur;
1796 /* We don't know the true size yet; copy the fixed-size part */
1797 if (cfs_copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1802 /* Compute the whole struct size */
1803 totalsize = hur_len(hur);
1805 OBD_ALLOC_LARGE(hur, totalsize);
1809 /* Copy the whole struct */
1810 if (cfs_copy_from_user(hur, (void *)arg, totalsize)) {
1811 OBD_FREE_LARGE(hur, totalsize);
1815 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1818 OBD_FREE_LARGE(hur, totalsize);
1822 case LL_IOC_HSM_PROGRESS: {
1823 struct hsm_progress_kernel hpk;
1824 struct hsm_progress hp;
1826 if (cfs_copy_from_user(&hp, (void *)arg, sizeof(hp)))
1829 hpk.hpk_fid = hp.hp_fid;
1830 hpk.hpk_cookie = hp.hp_cookie;
1831 hpk.hpk_extent = hp.hp_extent;
1832 hpk.hpk_flags = hp.hp_flags;
1833 hpk.hpk_errval = hp.hp_errval;
1834 hpk.hpk_data_version = 0;
1836 /* File may not exist in Lustre; all progress
1837 * reported to Lustre root */
1838 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1842 case LL_IOC_HSM_CT_START:
1843 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1844 sizeof(struct lustre_kernelcomm));
1847 case LL_IOC_HSM_COPY_START: {
1848 struct hsm_copy *copy;
1851 OBD_ALLOC_PTR(copy);
1854 if (cfs_copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1859 rc = ll_ioc_copy_start(inode->i_sb, copy);
1860 if (cfs_copy_to_user((char *)arg, copy, sizeof(*copy)))
1866 case LL_IOC_HSM_COPY_END: {
1867 struct hsm_copy *copy;
1870 OBD_ALLOC_PTR(copy);
1873 if (cfs_copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1878 rc = ll_ioc_copy_end(inode->i_sb, copy);
1879 if (cfs_copy_to_user((char *)arg, copy, sizeof(*copy)))
1886 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1891 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1893 struct inode *inode = file->f_mapping->host;
1894 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1895 struct ll_sb_info *sbi = ll_i2sbi(inode);
1896 int api32 = ll_need_32bit_api(sbi);
1897 loff_t ret = -EINVAL;
1900 mutex_lock(&inode->i_mutex);
1905 offset += file->f_pos;
1911 offset += LL_DIR_END_OFF_32BIT;
1913 offset += LL_DIR_END_OFF;
1920 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1921 (!api32 && offset <= LL_DIR_END_OFF))) {
1922 if (offset != file->f_pos) {
1923 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1924 (!api32 && offset == LL_DIR_END_OFF))
1925 fd->lfd_pos = MDS_DIR_END_OFF;
1926 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1927 fd->lfd_pos = offset << 32;
1929 fd->lfd_pos = offset;
1930 file->f_pos = offset;
1931 file->f_version = 0;
1938 mutex_unlock(&inode->i_mutex);
1942 int ll_dir_open(struct inode *inode, struct file *file)
1945 RETURN(ll_file_open(inode, file));
1948 int ll_dir_release(struct inode *inode, struct file *file)
1951 RETURN(ll_file_release(inode, file));
1954 struct file_operations ll_dir_operations = {
1955 .llseek = ll_dir_seek,
1956 .open = ll_dir_open,
1957 .release = ll_dir_release,
1958 .read = generic_read_dir,
1959 .readdir = ll_readdir,
1960 .unlocked_ioctl = ll_dir_ioctl,