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, 2013, 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 PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
138 * lu_dirpage for this integrated page will be adjusted. See
139 * lmv_adjust_dirpages().
143 /* returns the page unlocked, but with a reference */
144 static int ll_dir_filler(void *_hash, struct page *page0)
146 struct inode *inode = page0->mapping->host;
147 int hash64 = ll_i2sbi(inode)->ll_flags & LL_SBI_64BIT_HASH;
148 struct obd_export *exp = ll_i2sbi(inode)->ll_md_exp;
149 struct ptlrpc_request *request;
150 struct mdt_body *body;
151 struct md_op_data *op_data;
152 __u64 hash = *((__u64 *)_hash);
153 struct page **page_pool;
155 struct lu_dirpage *dp;
156 int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> PAGE_CACHE_SHIFT;
157 int nrdpgs = 0; /* number of pages read actually */
163 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) hash "LPU64"\n",
164 PFID(ll_inode2fid(inode)), inode, hash);
166 LASSERT(max_pages > 0 && max_pages <= MD_MAX_BRW_PAGES);
168 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
169 LUSTRE_OPC_ANY, NULL);
171 RETURN(PTR_ERR(op_data));
173 OBD_ALLOC(page_pool, sizeof(page) * max_pages);
174 if (page_pool != NULL) {
175 page_pool[0] = page0;
180 for (npages = 1; npages < max_pages; npages++) {
181 page = page_cache_alloc_cold(inode->i_mapping);
184 page_pool[npages] = page;
187 op_data->op_npages = npages;
188 op_data->op_offset = hash;
189 rc = md_readpage(exp, op_data, page_pool, &request);
190 ll_finish_md_op_data(op_data);
192 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
193 /* Checked by mdc_readpage() */
194 LASSERT(body != NULL);
196 if (body->valid & OBD_MD_FLSIZE)
197 cl_isize_write(inode, body->size);
199 nrdpgs = (request->rq_bulk->bd_nob_transferred +
200 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
201 SetPageUptodate(page0);
204 ptlrpc_req_finished(request);
206 CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
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,
233 CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
234 " %d\n", offset, ret);
235 page_cache_release(page);
238 if (page_pool != &page0)
239 OBD_FREE(page_pool, sizeof(struct page *) * max_pages);
244 static void ll_check_page(struct inode *dir, struct page *page)
246 /* XXX: check page format later */
247 SetPageChecked(page);
250 void ll_release_page(struct page *page, int remove)
255 if (likely(page->mapping != NULL))
256 truncate_complete_page(page->mapping, page);
259 page_cache_release(page);
263 * Find, kmap and return page that contains given hash.
265 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
266 __u64 *start, __u64 *end)
268 int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
269 struct address_space *mapping = dir->i_mapping;
271 * Complement of hash is used as an index so that
272 * radix_tree_gang_lookup() can be used to find a page with starting
273 * hash _smaller_ than one we are looking for.
275 unsigned long offset = hash_x_index(*hash, hash64);
279 spin_lock_irq(&mapping->tree_lock);
280 found = radix_tree_gang_lookup(&mapping->page_tree,
281 (void **)&page, offset, 1);
283 struct lu_dirpage *dp;
285 page_cache_get(page);
286 spin_unlock_irq(&mapping->tree_lock);
288 * In contrast to find_lock_page() we are sure that directory
289 * page cannot be truncated (while DLM lock is held) and,
290 * hence, can avoid restart.
292 * In fact, page cannot be locked here at all, because
293 * ll_dir_filler() does synchronous io.
295 wait_on_page_locked(page);
296 if (PageUptodate(page)) {
298 if (BITS_PER_LONG == 32 && hash64) {
299 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
300 *end = le64_to_cpu(dp->ldp_hash_end) >> 32;
303 *start = le64_to_cpu(dp->ldp_hash_start);
304 *end = le64_to_cpu(dp->ldp_hash_end);
306 LASSERTF(*start <= *hash, "start = "LPX64",end = "
307 LPX64",hash = "LPX64"\n", *start, *end, *hash);
308 CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash "LPU64"\n",
309 offset, *start, *end, *hash);
311 ll_release_page(page, 0);
313 } else if (*end != *start && *hash == *end) {
315 * upon hash collision, remove this page,
316 * otherwise put page reference, and
317 * ll_get_dir_page() will issue RPC to fetch
320 ll_release_page(page,
321 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
325 page_cache_release(page);
326 page = ERR_PTR(-EIO);
330 spin_unlock_irq(&mapping->tree_lock);
336 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
337 struct ll_dir_chain *chain)
339 ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
340 struct address_space *mapping = dir->i_mapping;
341 struct lustre_handle lockh;
342 struct lu_dirpage *dp;
349 struct ll_inode_info *lli = ll_i2info(dir);
350 int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
353 rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
354 ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
356 struct ldlm_enqueue_info einfo = {
357 .ei_type = LDLM_IBITS,
359 .ei_cb_bl = ll_md_blocking_ast,
360 .ei_cb_cp = ldlm_completion_ast,
362 struct lookup_intent it = { .it_op = IT_READDIR };
363 struct ptlrpc_request *request;
364 struct md_op_data *op_data;
366 op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
367 LUSTRE_OPC_ANY, NULL);
369 return (void *)op_data;
371 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
372 op_data, &lockh, NULL, 0, NULL, 0);
374 ll_finish_md_op_data(op_data);
376 request = (struct ptlrpc_request *)it.d.lustre.it_data;
378 ptlrpc_req_finished(request);
380 CERROR("lock enqueue: "DFID" at "LPU64": rc %d\n",
381 PFID(ll_inode2fid(dir)), hash, rc);
385 CDEBUG(D_INODE, "setting lr_lvb_inode to inode "DFID"(%p)\n",
386 PFID(ll_inode2fid(dir)), dir);
387 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
388 &it.d.lustre.it_lock_handle, dir, NULL);
390 /* for cross-ref object, l_ast_data of the lock may not be set,
391 * we reset it here */
392 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
395 ldlm_lock_dump_handle(D_OTHER, &lockh);
397 mutex_lock(&lli->lli_readdir_mutex);
398 page = ll_dir_page_locate(dir, &lhash, &start, &end);
400 CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
401 PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
402 GOTO(out_unlock, page);
403 } else if (page != NULL) {
405 * XXX nikita: not entirely correct handling of a corner case:
406 * suppose hash chain of entries with hash value HASH crosses
407 * border between pages P0 and P1. First both P0 and P1 are
408 * cached, seekdir() is called for some entry from the P0 part
409 * of the chain. Later P0 goes out of cache. telldir(HASH)
410 * happens and finds P1, as it starts with matching hash
411 * value. Remaining entries from P0 part of the chain are
412 * skipped. (Is that really a bug?)
414 * Possible solutions: 0. don't cache P1 is such case, handle
415 * it as an "overflow" page. 1. invalidate all pages at
416 * once. 2. use HASH|1 as an index for P1.
418 GOTO(hash_collision, page);
421 page = read_cache_page(mapping, hash_x_index(hash, hash64),
422 ll_dir_filler, &lhash);
424 CERROR("read cache page: "DFID" at "LPU64": rc %ld\n",
425 PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
426 GOTO(out_unlock, page);
429 wait_on_page_locked(page);
431 if (!PageUptodate(page)) {
432 CERROR("page not updated: "DFID" at "LPU64": rc %d\n",
433 PFID(ll_inode2fid(dir)), hash, -5);
436 if (!PageChecked(page))
437 ll_check_page(dir, page);
438 if (PageError(page)) {
439 CERROR("page error: "DFID" at "LPU64": rc %d\n",
440 PFID(ll_inode2fid(dir)), hash, -5);
444 dp = page_address(page);
445 if (BITS_PER_LONG == 32 && hash64) {
446 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
447 end = le64_to_cpu(dp->ldp_hash_end) >> 32;
450 start = le64_to_cpu(dp->ldp_hash_start);
451 end = le64_to_cpu(dp->ldp_hash_end);
455 LASSERT(start == lhash);
456 CWARN("Page-wide hash collision: "LPU64"\n", end);
457 if (BITS_PER_LONG == 32 && hash64)
458 CWARN("Real page-wide hash collision at ["LPU64" "LPU64
459 "] with hash "LPU64"\n",
460 le64_to_cpu(dp->ldp_hash_start),
461 le64_to_cpu(dp->ldp_hash_end), hash);
463 * Fetch whole overflow chain...
470 mutex_unlock(&lli->lli_readdir_mutex);
471 ldlm_lock_decref(&lockh, mode);
475 ll_release_page(page, 1);
476 page = ERR_PTR(-EIO);
480 int ll_dir_read(struct inode *inode, __u64 *_pos, void *cookie,
483 struct ll_inode_info *info = ll_i2info(inode);
484 struct ll_sb_info *sbi = ll_i2sbi(inode);
486 int api32 = ll_need_32bit_api(sbi);
487 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
489 struct ll_dir_chain chain;
494 ll_dir_chain_init(&chain);
496 page = ll_get_dir_page(inode, pos, &chain);
498 while (rc == 0 && !done) {
499 struct lu_dirpage *dp;
500 struct lu_dirent *ent;
504 * If page is empty (end of directory is reached),
507 __u64 hash = MDS_DIR_END_OFF;
510 dp = page_address(page);
511 for (ent = lu_dirent_start(dp); ent != NULL && !done;
512 ent = lu_dirent_next(ent)) {
520 * XXX: implement correct swabbing here.
523 hash = le64_to_cpu(ent->lde_hash);
526 * Skip until we find target hash
531 namelen = le16_to_cpu(ent->lde_namelen);
542 fid_le_to_cpu(&fid, &ent->lde_fid);
543 ino = cl_fid_build_ino(&fid, api32);
544 type = ll_dirent_type_get(ent);
545 /* For 'll_nfs_get_name_filldir()', it will try
546 * to access the 'ent' through its 'lde_name',
547 * so the parameter 'name' for 'filldir()' must
548 * be part of the 'ent'. */
549 done = filldir(cookie, ent->lde_name, namelen,
552 next = le64_to_cpu(dp->ldp_hash_end);
555 if (pos == MDS_DIR_END_OFF) {
557 * End of directory reached.
560 ll_release_page(page, 0);
561 } else if (1 /* chain is exhausted*/) {
563 * Normal case: continue to the next
566 ll_release_page(page,
567 le32_to_cpu(dp->ldp_flags) &
570 page = ll_get_dir_page(inode, pos,
574 * go into overflow page.
576 LASSERT(le32_to_cpu(dp->ldp_flags) &
578 ll_release_page(page, 1);
582 ll_release_page(page, 0);
586 CERROR("error reading dir "DFID" at %lu: rc %d\n",
587 PFID(&info->lli_fid), (unsigned long)pos, rc);
592 ll_dir_chain_fini(&chain);
596 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
598 struct inode *inode = filp->f_dentry->d_inode;
599 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
600 struct ll_sb_info *sbi = ll_i2sbi(inode);
602 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
603 int api32 = ll_need_32bit_api(sbi);
605 #ifdef HAVE_TOUCH_ATIME_1ARG
615 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos %lu/%llu "
616 " 32bit_api %d\n", PFID(ll_inode2fid(inode)),
617 inode, (unsigned long)pos, i_size_read(inode), api32);
619 if (pos == MDS_DIR_END_OFF)
625 rc = ll_dir_read(inode, &pos, cookie, filldir);
628 if (pos == MDS_DIR_END_OFF) {
630 filp->f_pos = LL_DIR_END_OFF_32BIT;
632 filp->f_pos = LL_DIR_END_OFF;
635 filp->f_pos = pos >> 32;
639 filp->f_version = inode->i_version;
640 #ifdef HAVE_TOUCH_ATIME_1ARG
641 #ifdef HAVE_F_PATH_MNT
642 path.mnt = filp->f_path.mnt;
644 path.mnt = filp->f_vfsmnt;
646 path.dentry = filp->f_dentry;
649 touch_atime(filp->f_vfsmnt, filp->f_dentry);
654 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
659 int ll_send_mgc_param(struct obd_export *mgc, char *string)
661 struct mgs_send_param *msp;
668 strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
669 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
670 sizeof(struct mgs_send_param), msp, NULL);
672 CERROR("Failed to set parameter: %d\n", rc);
678 int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
681 struct ptlrpc_request *request = NULL;
682 struct md_op_data *op_data;
683 struct ll_sb_info *sbi = ll_i2sbi(dir);
689 mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
690 op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
691 strlen(filename), mode, LUSTRE_OPC_MKDIR,
694 GOTO(err_exit, err = PTR_ERR(op_data));
696 op_data->op_cli_flags |= CLI_SET_MEA;
697 err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
698 current_fsuid(), current_fsgid(),
699 cfs_curproc_cap_pack(), 0, &request);
700 ll_finish_md_op_data(op_data);
704 ptlrpc_req_finished(request);
708 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
711 struct ll_sb_info *sbi = ll_i2sbi(inode);
712 struct md_op_data *op_data;
713 struct ptlrpc_request *req = NULL;
715 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
716 struct obd_device *mgc = lsi->lsi_mgc;
722 * This is coming from userspace, so should be in
723 * local endian. But the MDS would like it in little
724 * endian, so we swab it before we send it.
726 switch (lump->lmm_magic) {
727 case LOV_USER_MAGIC_V1: {
728 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
729 lustre_swab_lov_user_md_v1(lump);
730 lum_size = sizeof(struct lov_user_md_v1);
733 case LOV_USER_MAGIC_V3: {
734 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
735 lustre_swab_lov_user_md_v3(
736 (struct lov_user_md_v3 *)lump);
737 lum_size = sizeof(struct lov_user_md_v3);
741 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
742 " %#08x != %#08x nor %#08x\n",
743 lump->lmm_magic, LOV_USER_MAGIC_V1,
749 lum_size = sizeof(struct lov_user_md_v1);
752 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
753 LUSTRE_OPC_ANY, NULL);
755 RETURN(PTR_ERR(op_data));
757 if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
758 op_data->op_cli_flags |= CLI_SET_MEA;
760 /* swabbing is done in lov_setstripe() on server side */
761 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
762 NULL, 0, &req, NULL);
763 ll_finish_md_op_data(op_data);
764 ptlrpc_req_finished(req);
766 if (rc != -EPERM && rc != -EACCES)
767 CERROR("mdc_setattr fails: rc = %d\n", rc);
770 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
771 LOV_USER_MAGIC_V3 have the same initial fields so we do not
772 need the make the distiction between the 2 versions */
773 if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
777 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
779 GOTO(end, rc = -ENOMEM);
782 /* Get fsname and assume devname to be -MDT0000. */
783 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
784 strcat(buf, "-MDT0000.lov");
787 /* Set root stripesize */
788 sprintf(buf, ".stripesize=%u",
789 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
790 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
794 /* Set root stripecount */
795 sprintf(buf, ".stripecount=%hd",
796 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
797 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
801 /* Set root stripeoffset */
802 sprintf(buf, ".stripeoffset=%hd",
803 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
804 (typeof(lump->lmm_stripe_offset))(-1));
805 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
809 OBD_FREE(param, MGS_PARAM_MAXLEN);
814 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
815 int *lmm_size, struct ptlrpc_request **request)
817 struct ll_sb_info *sbi = ll_i2sbi(inode);
818 struct mdt_body *body;
819 struct lov_mds_md *lmm = NULL;
820 struct ptlrpc_request *req = NULL;
822 struct md_op_data *op_data;
824 rc = ll_get_max_mdsize(sbi, &lmmsize);
828 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
829 0, lmmsize, LUSTRE_OPC_ANY,
832 RETURN(PTR_ERR(op_data));
834 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
835 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
836 ll_finish_md_op_data(op_data);
838 CDEBUG(D_INFO, "md_getattr failed on inode "
839 DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
843 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
844 LASSERT(body != NULL);
846 lmmsize = body->eadatasize;
848 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
850 GOTO(out, rc = -ENODATA);
853 lmm = req_capsule_server_sized_get(&req->rq_pill,
854 &RMF_MDT_MD, lmmsize);
855 LASSERT(lmm != NULL);
858 * This is coming from the MDS, so is probably in
859 * little endian. We convert it to host endian before
860 * passing it to userspace.
862 /* We don't swab objects for directories */
863 switch (le32_to_cpu(lmm->lmm_magic)) {
865 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
866 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
869 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
870 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
873 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
884 * Get MDT index for the inode.
886 int ll_get_mdt_idx(struct inode *inode)
888 struct ll_sb_info *sbi = ll_i2sbi(inode);
889 struct md_op_data *op_data;
893 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
894 0, LUSTRE_OPC_ANY, NULL);
896 RETURN(PTR_ERR(op_data));
898 op_data->op_flags |= MF_GET_MDT_IDX;
899 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
900 mdtidx = op_data->op_mds;
901 ll_finish_md_op_data(op_data);
903 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
910 * Generic handler to do any pre-copy work.
912 * It send a first hsm_progress (with extent length == 0) to coordinator as a
913 * first information for it that real work has started.
915 * Moreover, for a ARCHIVE request, it will sample the file data version and
916 * store it in \a copy.
918 * \return 0 on success.
920 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
922 struct ll_sb_info *sbi = ll_s2sbi(sb);
923 struct hsm_progress_kernel hpk;
927 /* Forge a hsm_progress based on data from copy. */
928 hpk.hpk_fid = copy->hc_hai.hai_fid;
929 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
930 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
931 hpk.hpk_extent.length = 0;
934 hpk.hpk_data_version = 0;
937 /* For archive request, we need to read the current file version. */
938 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
940 __u64 data_version = 0;
942 /* Get inode for this fid */
943 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
945 hpk.hpk_flags |= HP_FLAG_RETRY;
946 /* hpk_errval is >= 0 */
947 hpk.hpk_errval = -PTR_ERR(inode);
948 GOTO(progress, rc = PTR_ERR(inode));
951 /* Read current file data version */
952 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
955 CDEBUG(D_HSM, "Could not read file data version of "
956 DFID" (rc = %d). Archive request ("
957 LPX64") could not be done.\n",
958 PFID(©->hc_hai.hai_fid), rc,
959 copy->hc_hai.hai_cookie);
960 hpk.hpk_flags |= HP_FLAG_RETRY;
961 /* hpk_errval must be >= 0 */
962 hpk.hpk_errval = -rc;
966 /* Store it the hsm_copy for later copytool use.
967 * Always modified even if no lsm. */
968 copy->hc_data_version = data_version;
972 /* On error, the request should be considered as completed */
973 if (hpk.hpk_errval > 0)
974 hpk.hpk_flags |= HP_FLAG_COMPLETED;
975 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
982 * Generic handler to do any post-copy work.
984 * It will send the last hsm_progress update to coordinator to inform it
985 * that copy is finished and whether it was successful or not.
988 * - for ARCHIVE request, it will sample the file data version and compare it
989 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
990 * will be considered as failed.
991 * - for RESTORE request, it will sample the file data version and send it to
992 * coordinator which is useful if the file was imported as 'released'.
994 * \return 0 on success.
996 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
998 struct ll_sb_info *sbi = ll_s2sbi(sb);
999 struct hsm_progress_kernel hpk;
1003 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
1004 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
1005 * initialized if copy_end was called with copy == NULL.
1008 /* Forge a hsm_progress based on data from copy. */
1009 hpk.hpk_fid = copy->hc_hai.hai_fid;
1010 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
1011 hpk.hpk_extent = copy->hc_hai.hai_extent;
1012 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
1013 hpk.hpk_errval = copy->hc_errval;
1014 hpk.hpk_data_version = 0;
1016 /* For archive request, we need to check the file data was not changed.
1018 * For restore request, we need to send the file data version, this is
1019 * useful when the file was created using hsm_import.
1021 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
1022 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
1023 (copy->hc_errval == 0)) {
1024 struct inode *inode;
1025 __u64 data_version = 0;
1027 /* Get lsm for this fid */
1028 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
1029 if (IS_ERR(inode)) {
1030 hpk.hpk_flags |= HP_FLAG_RETRY;
1031 /* hpk_errval must be >= 0 */
1032 hpk.hpk_errval = -PTR_ERR(inode);
1033 GOTO(progress, rc = PTR_ERR(inode));
1036 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
1039 CDEBUG(D_HSM, "Could not read file data version. "
1040 "Request could not be confirmed.\n");
1041 if (hpk.hpk_errval == 0)
1042 hpk.hpk_errval = -rc;
1046 /* Store it the hsm_copy for later copytool use.
1047 * Always modified even if no lsm. */
1048 hpk.hpk_data_version = data_version;
1050 /* File could have been stripped during archiving, so we need
1051 * to check anyway. */
1052 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1053 (copy->hc_data_version != data_version)) {
1054 CDEBUG(D_HSM, "File data version mismatched. "
1055 "File content was changed during archiving. "
1056 DFID", start:"LPX64" current:"LPX64"\n",
1057 PFID(©->hc_hai.hai_fid),
1058 copy->hc_data_version, data_version);
1059 /* File was changed, send error to cdt. Do not ask for
1060 * retry because if a file is modified frequently,
1061 * the cdt will loop on retried archive requests.
1062 * The policy engine will ask for a new archive later
1063 * when the file will not be modified for some tunable
1065 /* we do not notify caller */
1066 hpk.hpk_flags &= ~HP_FLAG_RETRY;
1067 /* hpk_errval must be >= 0 */
1068 hpk.hpk_errval = EBUSY;
1074 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1081 static int copy_and_ioctl(int cmd, struct obd_export *exp,
1082 const void __user *data, size_t size)
1087 OBD_ALLOC(copy, size);
1091 if (copy_from_user(copy, data, size)) {
1096 rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1098 OBD_FREE(copy, size);
1103 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1105 int cmd = qctl->qc_cmd;
1106 int type = qctl->qc_type;
1107 int id = qctl->qc_id;
1108 int valid = qctl->qc_valid;
1113 case LUSTRE_Q_INVALIDATE:
1114 case LUSTRE_Q_FINVALIDATE:
1119 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1120 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1124 if (((type == USRQUOTA && current_euid() != id) ||
1125 (type == GRPQUOTA && !in_egroup_p(id))) &&
1126 (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1127 sbi->ll_flags & LL_SBI_RMT_CLIENT))
1133 CERROR("unsupported quotactl op: %#x\n", cmd);
1137 if (valid != QC_GENERAL) {
1138 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1139 RETURN(-EOPNOTSUPP);
1141 if (cmd == Q_GETINFO)
1142 qctl->qc_cmd = Q_GETOINFO;
1143 else if (cmd == Q_GETQUOTA)
1144 qctl->qc_cmd = Q_GETOQUOTA;
1150 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1151 sizeof(*qctl), qctl, NULL);
1154 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1155 sizeof(*qctl), qctl, NULL);
1158 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1159 sizeof(*qctl), qctl, NULL);
1161 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1163 sizeof(*qctl), qctl, NULL);
1175 struct obd_quotactl *oqctl;
1177 OBD_ALLOC_PTR(oqctl);
1181 QCTL_COPY(oqctl, qctl);
1182 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1184 if (rc != -EALREADY && cmd == Q_QUOTAON) {
1185 oqctl->qc_cmd = Q_QUOTAOFF;
1186 obd_quotactl(sbi->ll_md_exp, oqctl);
1188 OBD_FREE_PTR(oqctl);
1191 /* If QIF_SPACE is not set, client should collect the
1192 * space usage from OSSs by itself */
1193 if (cmd == Q_GETQUOTA &&
1194 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1195 !oqctl->qc_dqblk.dqb_curspace) {
1196 struct obd_quotactl *oqctl_tmp;
1198 OBD_ALLOC_PTR(oqctl_tmp);
1199 if (oqctl_tmp == NULL)
1200 GOTO(out, rc = -ENOMEM);
1202 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1203 oqctl_tmp->qc_id = oqctl->qc_id;
1204 oqctl_tmp->qc_type = oqctl->qc_type;
1206 /* collect space usage from OSTs */
1207 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1208 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1209 if (!rc || rc == -EREMOTEIO) {
1210 oqctl->qc_dqblk.dqb_curspace =
1211 oqctl_tmp->qc_dqblk.dqb_curspace;
1212 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1215 /* collect space & inode usage from MDTs */
1216 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1217 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1218 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1219 if (!rc || rc == -EREMOTEIO) {
1220 oqctl->qc_dqblk.dqb_curspace +=
1221 oqctl_tmp->qc_dqblk.dqb_curspace;
1222 oqctl->qc_dqblk.dqb_curinodes =
1223 oqctl_tmp->qc_dqblk.dqb_curinodes;
1224 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1226 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1229 OBD_FREE_PTR(oqctl_tmp);
1232 QCTL_COPY(qctl, oqctl);
1233 OBD_FREE_PTR(oqctl);
1240 ll_getname(const char __user *filename)
1243 char *tmp = __getname();
1246 return ERR_PTR(-ENOMEM);
1248 len = strncpy_from_user(tmp, filename, PATH_MAX);
1251 else if (len > PATH_MAX)
1252 ret = -ENAMETOOLONG;
1261 #define ll_putname(filename) __putname(filename)
1263 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1265 struct inode *inode = file->f_dentry->d_inode;
1266 struct ll_sb_info *sbi = ll_i2sbi(inode);
1267 struct obd_ioctl_data *data;
1271 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1272 PFID(ll_inode2fid(inode)), inode, cmd);
1274 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1275 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1278 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1280 case FSFILT_IOC_GETFLAGS:
1281 case FSFILT_IOC_SETFLAGS:
1282 RETURN(ll_iocontrol(inode, file, cmd, arg));
1283 case FSFILT_IOC_GETVERSION_OLD:
1284 case FSFILT_IOC_GETVERSION:
1285 RETURN(put_user(inode->i_generation, (int *)arg));
1286 /* We need to special case any other ioctls we want to handle,
1287 * to send them to the MDS/OST as appropriate and to properly
1288 * network encode the arg field.
1289 case FSFILT_IOC_SETVERSION_OLD:
1290 case FSFILT_IOC_SETVERSION:
1292 case LL_IOC_GET_MDTIDX: {
1295 mdtidx = ll_get_mdt_idx(inode);
1299 if (put_user((int)mdtidx, (int*)arg))
1304 case IOC_MDC_LOOKUP: {
1305 struct ptlrpc_request *request = NULL;
1306 int namelen, len = 0;
1309 struct md_op_data *op_data;
1311 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1316 filename = data->ioc_inlbuf1;
1317 namelen = strlen(filename);
1320 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1321 GOTO(out_free, rc = -EINVAL);
1324 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1325 0, LUSTRE_OPC_ANY, NULL);
1326 if (IS_ERR(op_data))
1327 GOTO(out_free, rc = PTR_ERR(op_data));
1329 op_data->op_valid = OBD_MD_FLID;
1330 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1331 ll_finish_md_op_data(op_data);
1333 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1336 ptlrpc_req_finished(request);
1339 obd_ioctl_freedata(buf, len);
1342 case LL_IOC_LMV_SETSTRIPE: {
1343 struct lmv_user_md *lum;
1351 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1356 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1357 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1358 GOTO(lmv_out_free, rc = -EINVAL);
1360 filename = data->ioc_inlbuf1;
1361 namelen = data->ioc_inllen1;
1364 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1365 GOTO(lmv_out_free, rc = -EINVAL);
1367 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1368 lumlen = data->ioc_inllen2;
1370 if (lum->lum_magic != LMV_USER_MAGIC ||
1371 lumlen != sizeof(*lum)) {
1372 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1373 filename, lum->lum_magic, lumlen, -EFAULT);
1374 GOTO(lmv_out_free, rc = -EINVAL);
1378 * ll_dir_setdirstripe will be used to set dir stripe
1379 * mdc_create--->mdt_reint_create (with dirstripe)
1381 rc = ll_dir_setdirstripe(inode, lum, filename);
1383 obd_ioctl_freedata(buf, len);
1387 case LL_IOC_LOV_SETSTRIPE: {
1388 struct lov_user_md_v3 lumv3;
1389 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1390 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1391 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1393 int set_default = 0;
1395 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1396 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1397 sizeof(lumv3p->lmm_objects[0]));
1398 /* first try with v1 which is smaller than v3 */
1399 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1402 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1403 if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1407 if (inode->i_sb->s_root == file->f_dentry)
1410 /* in v1 and v3 cases lumv1 points to data */
1411 rc = ll_dir_setstripe(inode, lumv1, set_default);
1415 case LL_IOC_LMV_GETSTRIPE: {
1416 struct lmv_user_md *lump = (struct lmv_user_md *)arg;
1417 struct lmv_user_md lum;
1418 struct lmv_user_md *tmp;
1423 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1426 if (lum.lum_magic != LMV_MAGIC_V1)
1429 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1430 OBD_ALLOC(tmp, lum_size);
1432 GOTO(free_lmv, rc = -ENOMEM);
1434 memcpy(tmp, &lum, sizeof(lum));
1435 tmp->lum_type = LMV_STRIPE_TYPE;
1436 tmp->lum_stripe_count = 1;
1437 mdtindex = ll_get_mdt_idx(inode);
1439 GOTO(free_lmv, rc = -ENOMEM);
1441 tmp->lum_stripe_offset = mdtindex;
1442 tmp->lum_objects[0].lum_mds = mdtindex;
1443 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1444 sizeof(struct lu_fid));
1445 if (copy_to_user((void *)arg, tmp, lum_size))
1446 GOTO(free_lmv, rc = -EFAULT);
1449 OBD_FREE(tmp, lum_size);
1452 case LL_IOC_REMOVE_ENTRY: {
1453 char *filename = NULL;
1457 /* Here is a little hack to avoid sending REINT_RMENTRY to
1458 * unsupported server, which might crash the server(LU-2730),
1459 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1460 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1461 * server will support REINT_RMENTRY XXX*/
1462 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1465 filename = ll_getname((const char *)arg);
1466 if (IS_ERR(filename))
1467 RETURN(PTR_ERR(filename));
1469 namelen = strlen(filename);
1471 GOTO(out_rmdir, rc = -EINVAL);
1473 rc = ll_rmdir_entry(inode, filename, namelen);
1476 ll_putname(filename);
1479 case LL_IOC_LOV_SWAP_LAYOUTS:
1481 case LL_IOC_OBD_STATFS:
1482 RETURN(ll_obd_statfs(inode, (void *)arg));
1483 case LL_IOC_LOV_GETSTRIPE:
1484 case LL_IOC_MDC_GETINFO:
1485 case IOC_MDC_GETFILEINFO:
1486 case IOC_MDC_GETFILESTRIPE: {
1487 struct ptlrpc_request *request = NULL;
1488 struct lov_user_md *lump;
1489 struct lov_mds_md *lmm = NULL;
1490 struct mdt_body *body;
1491 char *filename = NULL;
1494 if (cmd == IOC_MDC_GETFILEINFO ||
1495 cmd == IOC_MDC_GETFILESTRIPE) {
1496 filename = ll_getname((const char *)arg);
1497 if (IS_ERR(filename))
1498 RETURN(PTR_ERR(filename));
1500 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1501 &lmmsize, &request);
1503 rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1507 body = req_capsule_server_get(&request->rq_pill,
1509 LASSERT(body != NULL);
1515 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1516 cmd == LL_IOC_MDC_GETINFO))
1517 GOTO(skip_lmm, rc = 0);
1522 if (cmd == IOC_MDC_GETFILESTRIPE ||
1523 cmd == LL_IOC_LOV_GETSTRIPE) {
1524 lump = (struct lov_user_md *)arg;
1526 struct lov_user_mds_data *lmdp;
1527 lmdp = (struct lov_user_mds_data *)arg;
1528 lump = &lmdp->lmd_lmm;
1530 if (copy_to_user(lump, lmm, lmmsize)) {
1531 if (copy_to_user(lump, lmm, sizeof(*lump)))
1532 GOTO(out_req, rc = -EFAULT);
1536 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1537 struct lov_user_mds_data *lmdp;
1540 st.st_dev = inode->i_sb->s_dev;
1541 st.st_mode = body->mode;
1542 st.st_nlink = body->nlink;
1543 st.st_uid = body->uid;
1544 st.st_gid = body->gid;
1545 st.st_rdev = body->rdev;
1546 st.st_size = body->size;
1547 st.st_blksize = PAGE_CACHE_SIZE;
1548 st.st_blocks = body->blocks;
1549 st.st_atime = body->atime;
1550 st.st_mtime = body->mtime;
1551 st.st_ctime = body->ctime;
1552 st.st_ino = inode->i_ino;
1554 lmdp = (struct lov_user_mds_data *)arg;
1555 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1556 GOTO(out_req, rc = -EFAULT);
1561 ptlrpc_req_finished(request);
1563 ll_putname(filename);
1566 case IOC_LOV_GETINFO: {
1567 struct lov_user_mds_data *lumd;
1568 struct lov_stripe_md *lsm;
1569 struct lov_user_md *lum;
1570 struct lov_mds_md *lmm;
1574 lumd = (struct lov_user_mds_data *)arg;
1575 lum = &lumd->lmd_lmm;
1577 rc = ll_get_max_mdsize(sbi, &lmmsize);
1581 OBD_ALLOC_LARGE(lmm, lmmsize);
1585 if (copy_from_user(lmm, lum, lmmsize))
1586 GOTO(free_lmm, rc = -EFAULT);
1588 switch (lmm->lmm_magic) {
1589 case LOV_USER_MAGIC_V1:
1590 if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1592 /* swab objects first so that stripes num will be sane */
1593 lustre_swab_lov_user_md_objects(
1594 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1595 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1596 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1598 case LOV_USER_MAGIC_V3:
1599 if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1601 /* swab objects first so that stripes num will be sane */
1602 lustre_swab_lov_user_md_objects(
1603 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1604 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1605 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1608 GOTO(free_lmm, rc = -EINVAL);
1611 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1613 GOTO(free_lmm, rc = -ENOMEM);
1615 /* Perform glimpse_size operation. */
1616 memset(&st, 0, sizeof(st));
1618 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1622 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1623 GOTO(free_lsm, rc = -EFAULT);
1627 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1629 OBD_FREE_LARGE(lmm, lmmsize);
1632 case OBD_IOC_LLOG_CATINFO: {
1633 RETURN(-EOPNOTSUPP);
1635 case OBD_IOC_QUOTACHECK: {
1636 struct obd_quotactl *oqctl;
1639 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1640 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1643 OBD_ALLOC_PTR(oqctl);
1646 oqctl->qc_type = arg;
1647 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1649 CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1653 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1655 CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1657 OBD_FREE_PTR(oqctl);
1660 case OBD_IOC_POLL_QUOTACHECK: {
1661 struct if_quotacheck *check;
1663 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1664 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1667 OBD_ALLOC_PTR(check);
1671 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1674 CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1675 if (copy_to_user((void *)arg, check,
1677 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1681 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1684 CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1685 if (copy_to_user((void *)arg, check,
1687 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1691 OBD_FREE_PTR(check);
1694 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1695 case LL_IOC_QUOTACTL_18: {
1696 /* copy the old 1.x quota struct for internal use, then copy
1697 * back into old format struct. For 1.8 compatibility. */
1698 struct if_quotactl_18 *qctl_18;
1699 struct if_quotactl *qctl_20;
1701 OBD_ALLOC_PTR(qctl_18);
1705 OBD_ALLOC_PTR(qctl_20);
1707 GOTO(out_quotactl_18, rc = -ENOMEM);
1709 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1710 GOTO(out_quotactl_20, rc = -ENOMEM);
1712 QCTL_COPY(qctl_20, qctl_18);
1713 qctl_20->qc_idx = 0;
1715 /* XXX: dqb_valid was borrowed as a flag to mark that
1716 * only mds quota is wanted */
1717 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1718 qctl_18->qc_dqblk.dqb_valid) {
1719 qctl_20->qc_valid = QC_MDTIDX;
1720 qctl_20->qc_dqblk.dqb_valid = 0;
1721 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1722 qctl_20->qc_valid = QC_UUID;
1723 qctl_20->obd_uuid = qctl_18->obd_uuid;
1725 qctl_20->qc_valid = QC_GENERAL;
1728 rc = quotactl_ioctl(sbi, qctl_20);
1731 QCTL_COPY(qctl_18, qctl_20);
1732 qctl_18->obd_uuid = qctl_20->obd_uuid;
1734 if (copy_to_user((void *)arg, qctl_18,
1740 OBD_FREE_PTR(qctl_20);
1742 OBD_FREE_PTR(qctl_18);
1746 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1747 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1748 case LL_IOC_QUOTACTL: {
1749 struct if_quotactl *qctl;
1751 OBD_ALLOC_PTR(qctl);
1755 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1756 GOTO(out_quotactl, rc = -EFAULT);
1758 rc = quotactl_ioctl(sbi, qctl);
1760 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1767 case OBD_IOC_GETDTNAME:
1768 case OBD_IOC_GETMDNAME:
1769 RETURN(ll_get_obd_name(inode, cmd, arg));
1770 case LL_IOC_FLUSHCTX:
1771 RETURN(ll_flush_ctx(inode));
1772 #ifdef CONFIG_FS_POSIX_ACL
1773 case LL_IOC_RMTACL: {
1774 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1775 inode == inode->i_sb->s_root->d_inode) {
1776 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1778 LASSERT(fd != NULL);
1779 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1781 fd->fd_flags |= LL_FILE_RMTACL;
1787 case LL_IOC_GETOBDCOUNT: {
1789 struct obd_export *exp;
1791 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1794 /* get ost count when count is zero, get mdt count otherwise */
1795 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1796 vallen = sizeof(count);
1797 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1798 KEY_TGT_COUNT, &vallen, &count, NULL);
1800 CERROR("get target count failed: %d\n", rc);
1804 if (copy_to_user((int *)arg, &count, sizeof(int)))
1809 case LL_IOC_PATH2FID:
1810 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1811 sizeof(struct lu_fid)))
1814 case LL_IOC_GET_CONNECT_FLAGS: {
1815 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1817 case OBD_IOC_CHANGELOG_SEND:
1818 case OBD_IOC_CHANGELOG_CLEAR:
1819 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1820 sizeof(struct ioc_changelog));
1822 case OBD_IOC_FID2PATH:
1823 RETURN(ll_fid2path(inode, (void *)arg));
1824 case LL_IOC_HSM_REQUEST: {
1825 struct hsm_user_request *hur;
1832 /* We don't know the true size yet; copy the fixed-size part */
1833 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1838 /* Compute the whole struct size */
1839 totalsize = hur_len(hur);
1842 /* Make sure the size is reasonable */
1843 if (totalsize >= MDS_MAXREQSIZE)
1846 OBD_ALLOC_LARGE(hur, totalsize);
1850 /* Copy the whole struct */
1851 if (copy_from_user(hur, (void *)arg, totalsize)) {
1852 OBD_FREE_LARGE(hur, totalsize);
1856 if (hur->hur_request.hr_action == HUA_RELEASE) {
1857 const struct lu_fid *fid;
1861 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1862 fid = &hur->hur_user_item[i].hui_fid;
1863 f = search_inode_for_lustre(inode->i_sb, fid);
1869 rc = ll_hsm_release(f);
1875 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1879 OBD_FREE_LARGE(hur, totalsize);
1883 case LL_IOC_HSM_PROGRESS: {
1884 struct hsm_progress_kernel hpk;
1885 struct hsm_progress hp;
1887 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1890 hpk.hpk_fid = hp.hp_fid;
1891 hpk.hpk_cookie = hp.hp_cookie;
1892 hpk.hpk_extent = hp.hp_extent;
1893 hpk.hpk_flags = hp.hp_flags;
1894 hpk.hpk_errval = hp.hp_errval;
1895 hpk.hpk_data_version = 0;
1897 /* File may not exist in Lustre; all progress
1898 * reported to Lustre root */
1899 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1903 case LL_IOC_HSM_CT_START:
1904 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1907 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1908 sizeof(struct lustre_kernelcomm));
1911 case LL_IOC_HSM_COPY_START: {
1912 struct hsm_copy *copy;
1915 OBD_ALLOC_PTR(copy);
1918 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1923 rc = ll_ioc_copy_start(inode->i_sb, copy);
1924 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1930 case LL_IOC_HSM_COPY_END: {
1931 struct hsm_copy *copy;
1934 OBD_ALLOC_PTR(copy);
1937 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1942 rc = ll_ioc_copy_end(inode->i_sb, copy);
1943 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1950 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1955 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1957 struct inode *inode = file->f_mapping->host;
1958 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1959 struct ll_sb_info *sbi = ll_i2sbi(inode);
1960 int api32 = ll_need_32bit_api(sbi);
1961 loff_t ret = -EINVAL;
1964 mutex_lock(&inode->i_mutex);
1969 offset += file->f_pos;
1975 offset += LL_DIR_END_OFF_32BIT;
1977 offset += LL_DIR_END_OFF;
1984 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1985 (!api32 && offset <= LL_DIR_END_OFF))) {
1986 if (offset != file->f_pos) {
1987 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1988 (!api32 && offset == LL_DIR_END_OFF))
1989 fd->lfd_pos = MDS_DIR_END_OFF;
1990 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1991 fd->lfd_pos = offset << 32;
1993 fd->lfd_pos = offset;
1994 file->f_pos = offset;
1995 file->f_version = 0;
2002 mutex_unlock(&inode->i_mutex);
2006 int ll_dir_open(struct inode *inode, struct file *file)
2009 RETURN(ll_file_open(inode, file));
2012 int ll_dir_release(struct inode *inode, struct file *file)
2015 RETURN(ll_file_release(inode, file));
2018 struct file_operations ll_dir_operations = {
2019 .llseek = ll_dir_seek,
2020 .open = ll_dir_open,
2021 .release = ll_dir_release,
2022 .read = generic_read_dir,
2023 .readdir = ll_readdir,
2024 .unlocked_ioctl = ll_dir_ioctl,