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/uaccess.h>
45 #include <linux/buffer_head.h> /* for wait_on_buffer */
46 #include <linux/pagevec.h>
47 #include <linux/prefetch.h>
49 #define DEBUG_SUBSYSTEM S_LLITE
51 #include "../include/obd_support.h"
52 #include "../include/obd_class.h"
53 #include "../include/lustre_lib.h"
54 #include "../include/lustre/lustre_idl.h"
55 #include "../include/lustre_lite.h"
56 #include "../include/lustre_dlm.h"
57 #include "../include/lustre_fid.h"
58 #include "llite_internal.h"
61 * (new) readdir implementation overview.
63 * Original lustre readdir implementation cached exact copy of raw directory
64 * pages on the client. These pages were indexed in client page cache by
65 * logical offset in the directory file. This design, while very simple and
66 * intuitive had some inherent problems:
68 * . it implies that byte offset to the directory entry serves as a
69 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
70 * ext3/htree directory entries may move due to splits, and more
73 * . it is incompatible with the design of split directories for cmd3,
74 * that assumes that names are distributed across nodes based on their
75 * hash, and so readdir should be done in hash order.
77 * New readdir implementation does readdir in hash order, and uses hash of a
78 * file name as a telldir/seekdir cookie. This led to number of complications:
80 * . hash is not unique, so it cannot be used to index cached directory
81 * pages on the client (note, that it requires a whole pageful of hash
82 * collided entries to cause two pages to have identical hashes);
84 * . hash is not unique, so it cannot, strictly speaking, be used as an
85 * entry cookie. ext3/htree has the same problem and lustre implementation
86 * mimics their solution: seekdir(hash) positions directory at the first
87 * entry with the given hash.
93 * Client caches directory pages using hash of the first entry as an index. As
94 * noted above hash is not unique, so this solution doesn't work as is:
95 * special processing is needed for "page hash chains" (i.e., sequences of
96 * pages filled with entries all having the same hash value).
98 * First, such chains have to be detected. To this end, server returns to the
99 * client the hash of the first entry on the page next to one returned. When
100 * client detects that this hash is the same as hash of the first entry on the
101 * returned page, page hash collision has to be handled. Pages in the
102 * hash chain, except first one, are termed "overflow pages".
104 * Solution to index uniqueness problem is to not cache overflow
105 * pages. Instead, when page hash collision is detected, all overflow pages
106 * from emerging chain are immediately requested from the server and placed in
107 * a special data structure (struct ll_dir_chain). This data structure is used
108 * by ll_readdir() to process entries from overflow pages. When readdir
109 * invocation finishes, overflow pages are discarded. If page hash collision
110 * chain weren't completely processed, next call to readdir will again detect
111 * page hash collision, again read overflow pages in, process next portion of
112 * entries and again discard the pages. This is not as wasteful as it looks,
113 * because, given reasonable hash, page hash collisions are extremely rare.
115 * 1. directory positioning
117 * When seekdir(hash) is called, original
128 * identification of and access to overflow pages
132 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
133 * a header lu_dirpage which describes the start/end hash, and whether this
134 * page is empty (contains no dir entry) or hash collide with next page.
135 * After client receives reply, several pages will be integrated into dir page
136 * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
137 * lu_dirpage for this integrated page will be adjusted. See
138 * lmv_adjust_dirpages().
142 /* returns the page unlocked, but with a reference */
143 static int ll_dir_filler(void *_hash, struct page *page0)
145 struct inode *inode = page0->mapping->host;
146 int hash64 = ll_i2sbi(inode)->ll_flags & LL_SBI_64BIT_HASH;
147 struct obd_export *exp = ll_i2sbi(inode)->ll_md_exp;
148 struct ptlrpc_request *request;
149 struct mdt_body *body;
150 struct md_op_data *op_data;
151 __u64 hash = *((__u64 *)_hash);
152 struct page **page_pool;
154 struct lu_dirpage *dp;
155 int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> PAGE_CACHE_SHIFT;
156 int nrdpgs = 0; /* number of pages read actually */
161 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash %llu\n",
162 inode->i_ino, inode->i_generation, inode, hash);
164 LASSERT(max_pages > 0 && max_pages <= MD_MAX_BRW_PAGES);
166 page_pool = kcalloc(max_pages, sizeof(page), GFP_NOFS);
168 page_pool[0] = page0;
173 for (npages = 1; npages < max_pages; npages++) {
174 page = page_cache_alloc_cold(inode->i_mapping);
177 page_pool[npages] = page;
180 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
181 LUSTRE_OPC_ANY, NULL);
182 op_data->op_npages = npages;
183 op_data->op_offset = hash;
184 rc = md_readpage(exp, op_data, page_pool, &request);
185 ll_finish_md_op_data(op_data);
187 /* page0 is special, which was added into page cache early */
188 delete_from_page_cache(page0);
189 } else if (rc == 0) {
190 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
191 /* Checked by mdc_readpage() */
192 LASSERT(body != NULL);
194 if (body->valid & OBD_MD_FLSIZE)
195 cl_isize_write(inode, body->size);
197 nrdpgs = (request->rq_bulk->bd_nob_transferred+PAGE_CACHE_SIZE-1)
199 SetPageUptodate(page0);
202 ptlrpc_req_finished(request);
204 CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
206 ll_pagevec_init(&lru_pvec, 0);
207 for (i = 1; i < npages; i++) {
208 unsigned long offset;
213 if (rc < 0 || i >= nrdpgs) {
214 page_cache_release(page);
218 SetPageUptodate(page);
221 hash = le64_to_cpu(dp->ldp_hash_start);
224 offset = hash_x_index(hash, hash64);
226 prefetchw(&page->flags);
227 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
231 if (ll_pagevec_add(&lru_pvec, page) == 0)
232 ll_pagevec_lru_add_file(&lru_pvec);
234 CDEBUG(D_VFSTRACE, "page %lu add to page cache failed: %d\n",
237 page_cache_release(page);
239 ll_pagevec_lru_add_file(&lru_pvec);
241 if (page_pool != &page0)
242 OBD_FREE(page_pool, sizeof(struct page *) * max_pages);
246 static void ll_check_page(struct inode *dir, struct page *page)
248 /* XXX: check page format later */
249 SetPageChecked(page);
252 void ll_release_page(struct page *page, int remove)
257 if (likely(page->mapping != NULL))
258 truncate_complete_page(page->mapping, page);
261 page_cache_release(page);
265 * Find, kmap and return page that contains given hash.
267 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
268 __u64 *start, __u64 *end)
270 int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
271 struct address_space *mapping = dir->i_mapping;
273 * Complement of hash is used as an index so that
274 * radix_tree_gang_lookup() can be used to find a page with starting
275 * hash _smaller_ than one we are looking for.
277 unsigned long offset = hash_x_index(*hash, hash64);
281 spin_lock_irq(&mapping->tree_lock);
282 found = radix_tree_gang_lookup(&mapping->page_tree,
283 (void **)&page, offset, 1);
284 if (found > 0 && !radix_tree_exceptional_entry(page)) {
285 struct lu_dirpage *dp;
287 page_cache_get(page);
288 spin_unlock_irq(&mapping->tree_lock);
290 * In contrast to find_lock_page() we are sure that directory
291 * page cannot be truncated (while DLM lock is held) and,
292 * hence, can avoid restart.
294 * In fact, page cannot be locked here at all, because
295 * ll_dir_filler() does synchronous io.
297 wait_on_page_locked(page);
298 if (PageUptodate(page)) {
300 if (BITS_PER_LONG == 32 && hash64) {
301 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
302 *end = le64_to_cpu(dp->ldp_hash_end) >> 32;
305 *start = le64_to_cpu(dp->ldp_hash_start);
306 *end = le64_to_cpu(dp->ldp_hash_end);
308 LASSERTF(*start <= *hash, "start = %#llx,end = %#llx,hash = %#llx\n",
309 *start, *end, *hash);
310 CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash %llu\n",
311 offset, *start, *end, *hash);
313 ll_release_page(page, 0);
315 } else if (*end != *start && *hash == *end) {
317 * upon hash collision, remove this page,
318 * otherwise put page reference, and
319 * ll_get_dir_page() will issue RPC to fetch
322 ll_release_page(page,
323 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
327 page_cache_release(page);
328 page = ERR_PTR(-EIO);
332 spin_unlock_irq(&mapping->tree_lock);
338 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
339 struct ll_dir_chain *chain)
341 ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
342 struct address_space *mapping = dir->i_mapping;
343 struct lustre_handle lockh;
344 struct lu_dirpage *dp;
351 struct ll_inode_info *lli = ll_i2info(dir);
352 int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
355 rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
356 ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
358 struct ldlm_enqueue_info einfo = {
359 .ei_type = LDLM_IBITS,
361 .ei_cb_bl = ll_md_blocking_ast,
362 .ei_cb_cp = ldlm_completion_ast,
364 struct lookup_intent it = { .it_op = IT_READDIR };
365 struct ptlrpc_request *request;
366 struct md_op_data *op_data;
368 op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
369 LUSTRE_OPC_ANY, NULL);
371 return (void *)op_data;
373 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
374 op_data, &lockh, NULL, 0, NULL, 0);
376 ll_finish_md_op_data(op_data);
378 request = (struct ptlrpc_request *)it.d.lustre.it_data;
380 ptlrpc_req_finished(request);
382 CERROR("lock enqueue: "DFID" at %llu: rc %d\n",
383 PFID(ll_inode2fid(dir)), hash, rc);
387 CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
388 dir, dir->i_ino, dir->i_generation);
389 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
390 &it.d.lustre.it_lock_handle, dir, NULL);
392 /* for cross-ref object, l_ast_data of the lock may not be set,
393 * we reset it here */
394 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
397 ldlm_lock_dump_handle(D_OTHER, &lockh);
399 mutex_lock(&lli->lli_readdir_mutex);
400 page = ll_dir_page_locate(dir, &lhash, &start, &end);
402 CERROR("dir page locate: "DFID" at %llu: rc %ld\n",
403 PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
405 } else if (page != NULL) {
407 * XXX nikita: not entirely correct handling of a corner case:
408 * suppose hash chain of entries with hash value HASH crosses
409 * border between pages P0 and P1. First both P0 and P1 are
410 * cached, seekdir() is called for some entry from the P0 part
411 * of the chain. Later P0 goes out of cache. telldir(HASH)
412 * happens and finds P1, as it starts with matching hash
413 * value. Remaining entries from P0 part of the chain are
414 * skipped. (Is that really a bug?)
416 * Possible solutions: 0. don't cache P1 is such case, handle
417 * it as an "overflow" page. 1. invalidate all pages at
418 * once. 2. use HASH|1 as an index for P1.
423 page = read_cache_page(mapping, hash_x_index(hash, hash64),
424 ll_dir_filler, &lhash);
426 CERROR("read cache page: "DFID" at %llu: rc %ld\n",
427 PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
431 wait_on_page_locked(page);
433 if (!PageUptodate(page)) {
434 CERROR("page not updated: "DFID" at %llu: rc %d\n",
435 PFID(ll_inode2fid(dir)), hash, -5);
438 if (!PageChecked(page))
439 ll_check_page(dir, page);
440 if (PageError(page)) {
441 CERROR("page error: "DFID" at %llu: rc %d\n",
442 PFID(ll_inode2fid(dir)), hash, -5);
446 dp = page_address(page);
447 if (BITS_PER_LONG == 32 && hash64) {
448 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
449 end = le64_to_cpu(dp->ldp_hash_end) >> 32;
452 start = le64_to_cpu(dp->ldp_hash_start);
453 end = le64_to_cpu(dp->ldp_hash_end);
457 LASSERT(start == lhash);
458 CWARN("Page-wide hash collision: %llu\n", end);
459 if (BITS_PER_LONG == 32 && hash64)
460 CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n",
461 le64_to_cpu(dp->ldp_hash_start),
462 le64_to_cpu(dp->ldp_hash_end), hash);
464 * Fetch whole overflow chain...
471 mutex_unlock(&lli->lli_readdir_mutex);
472 ldlm_lock_decref(&lockh, mode);
476 ll_release_page(page, 1);
477 page = ERR_PTR(-EIO);
481 int ll_dir_read(struct inode *inode, struct dir_context *ctx)
483 struct ll_inode_info *info = ll_i2info(inode);
484 struct ll_sb_info *sbi = ll_i2sbi(inode);
485 __u64 pos = ctx->pos;
486 int api32 = ll_need_32bit_api(sbi);
487 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
489 struct ll_dir_chain chain;
493 ll_dir_chain_init(&chain);
495 page = ll_get_dir_page(inode, pos, &chain);
497 while (rc == 0 && !done) {
498 struct lu_dirpage *dp;
499 struct lu_dirent *ent;
503 * If page is empty (end of directory is reached),
506 __u64 hash = MDS_DIR_END_OFF;
509 dp = page_address(page);
510 for (ent = lu_dirent_start(dp); ent != NULL && !done;
511 ent = lu_dirent_next(ent)) {
519 * XXX: implement correct swabbing here.
522 hash = le64_to_cpu(ent->lde_hash);
525 * Skip until we find target hash
530 namelen = le16_to_cpu(ent->lde_namelen);
541 fid_le_to_cpu(&fid, &ent->lde_fid);
542 ino = cl_fid_build_ino(&fid, api32);
543 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 'ctx->actor()'
548 * must be part of the 'ent'.
550 done = !dir_emit(ctx, ent->lde_name,
553 next = le64_to_cpu(dp->ldp_hash_end);
556 if (pos == MDS_DIR_END_OFF) {
558 * End of directory reached.
561 ll_release_page(page, 0);
562 } else if (1 /* chain is exhausted*/) {
564 * Normal case: continue to the next
567 ll_release_page(page,
568 le32_to_cpu(dp->ldp_flags) &
571 page = ll_get_dir_page(inode, pos,
575 * go into overflow page.
577 LASSERT(le32_to_cpu(dp->ldp_flags) &
579 ll_release_page(page, 1);
583 ll_release_page(page, 0);
587 CERROR("error reading dir "DFID" at %lu: rc %d\n",
588 PFID(&info->lli_fid), (unsigned long)pos, rc);
593 ll_dir_chain_fini(&chain);
597 static int ll_readdir(struct file *filp, struct dir_context *ctx)
599 struct inode *inode = file_inode(filp);
600 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
601 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);
606 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
607 inode->i_ino, inode->i_generation,
608 inode, (unsigned long)lfd->lfd_pos, i_size_read(inode), api32);
610 if (lfd->lfd_pos == MDS_DIR_END_OFF) {
618 ctx->pos = lfd->lfd_pos;
619 rc = ll_dir_read(inode, ctx);
620 lfd->lfd_pos = ctx->pos;
621 if (ctx->pos == MDS_DIR_END_OFF) {
623 ctx->pos = LL_DIR_END_OFF_32BIT;
625 ctx->pos = LL_DIR_END_OFF;
630 filp->f_version = inode->i_version;
634 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
639 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
641 struct mgs_send_param *msp;
644 msp = kzalloc(sizeof(*msp), GFP_NOFS);
648 strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
649 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
650 sizeof(struct mgs_send_param), msp, NULL);
652 CERROR("Failed to set parameter: %d\n", rc);
658 static int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
661 struct ptlrpc_request *request = NULL;
662 struct md_op_data *op_data;
663 struct ll_sb_info *sbi = ll_i2sbi(dir);
667 mode = (0755 & ~current_umask()) | S_IFDIR;
668 op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
669 strlen(filename), mode, LUSTRE_OPC_MKDIR,
671 if (IS_ERR(op_data)) {
672 err = PTR_ERR(op_data);
676 op_data->op_cli_flags |= CLI_SET_MEA;
677 err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
678 from_kuid(&init_user_ns, current_fsuid()),
679 from_kgid(&init_user_ns, current_fsgid()),
680 cfs_curproc_cap_pack(), 0, &request);
681 ll_finish_md_op_data(op_data);
685 ptlrpc_req_finished(request);
689 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
692 struct ll_sb_info *sbi = ll_i2sbi(inode);
693 struct md_op_data *op_data;
694 struct ptlrpc_request *req = NULL;
696 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
697 struct obd_device *mgc = lsi->lsi_mgc;
702 * This is coming from userspace, so should be in
703 * local endian. But the MDS would like it in little
704 * endian, so we swab it before we send it.
706 switch (lump->lmm_magic) {
707 case LOV_USER_MAGIC_V1: {
708 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
709 lustre_swab_lov_user_md_v1(lump);
710 lum_size = sizeof(struct lov_user_md_v1);
713 case LOV_USER_MAGIC_V3: {
714 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
715 lustre_swab_lov_user_md_v3(
716 (struct lov_user_md_v3 *)lump);
717 lum_size = sizeof(struct lov_user_md_v3);
721 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
722 lump->lmm_magic, LOV_USER_MAGIC_V1,
728 lum_size = sizeof(struct lov_user_md_v1);
731 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
732 LUSTRE_OPC_ANY, NULL);
734 return PTR_ERR(op_data);
736 if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
737 op_data->op_cli_flags |= CLI_SET_MEA;
739 /* swabbing is done in lov_setstripe() on server side */
740 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
741 NULL, 0, &req, NULL);
742 ll_finish_md_op_data(op_data);
743 ptlrpc_req_finished(req);
745 if (rc != -EPERM && rc != -EACCES)
746 CERROR("mdc_setattr fails: rc = %d\n", rc);
749 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
750 LOV_USER_MAGIC_V3 have the same initial fields so we do not
751 need to make the distinction between the 2 versions */
752 if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
756 param = kzalloc(MGS_PARAM_MAXLEN, GFP_NOFS);
763 /* Get fsname and assume devname to be -MDT0000. */
764 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
765 strcat(buf, "-MDT0000.lov");
768 /* Set root stripesize */
769 sprintf(buf, ".stripesize=%u",
770 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
771 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
775 /* Set root stripecount */
776 sprintf(buf, ".stripecount=%hd",
777 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
778 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
782 /* Set root stripeoffset */
783 sprintf(buf, ".stripeoffset=%hd",
784 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
785 (typeof(lump->lmm_stripe_offset))(-1));
786 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
790 OBD_FREE(param, MGS_PARAM_MAXLEN);
795 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
796 int *lmm_size, struct ptlrpc_request **request)
798 struct ll_sb_info *sbi = ll_i2sbi(inode);
799 struct mdt_body *body;
800 struct lov_mds_md *lmm = NULL;
801 struct ptlrpc_request *req = NULL;
803 struct md_op_data *op_data;
805 rc = ll_get_default_mdsize(sbi, &lmmsize);
809 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
810 0, lmmsize, LUSTRE_OPC_ANY,
813 return PTR_ERR(op_data);
815 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
816 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
817 ll_finish_md_op_data(op_data);
819 CDEBUG(D_INFO, "md_getattr failed on inode %lu/%u: rc %d\n",
821 inode->i_generation, rc);
825 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
826 LASSERT(body != NULL);
828 lmmsize = body->eadatasize;
830 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
836 lmm = req_capsule_server_sized_get(&req->rq_pill,
837 &RMF_MDT_MD, lmmsize);
838 LASSERT(lmm != NULL);
841 * This is coming from the MDS, so is probably in
842 * little endian. We convert it to host endian before
843 * passing it to userspace.
845 /* We don't swab objects for directories */
846 switch (le32_to_cpu(lmm->lmm_magic)) {
848 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
849 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
852 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
853 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
856 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
867 * Get MDT index for the inode.
869 int ll_get_mdt_idx(struct inode *inode)
871 struct ll_sb_info *sbi = ll_i2sbi(inode);
872 struct md_op_data *op_data;
875 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
876 0, LUSTRE_OPC_ANY, NULL);
878 return PTR_ERR(op_data);
880 op_data->op_flags |= MF_GET_MDT_IDX;
881 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
882 mdtidx = op_data->op_mds;
883 ll_finish_md_op_data(op_data);
885 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
892 * Generic handler to do any pre-copy work.
894 * It send a first hsm_progress (with extent length == 0) to coordinator as a
895 * first information for it that real work has started.
897 * Moreover, for a ARCHIVE request, it will sample the file data version and
898 * store it in \a copy.
900 * \return 0 on success.
902 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
904 struct ll_sb_info *sbi = ll_s2sbi(sb);
905 struct hsm_progress_kernel hpk;
908 /* Forge a hsm_progress based on data from copy. */
909 hpk.hpk_fid = copy->hc_hai.hai_fid;
910 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
911 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
912 hpk.hpk_extent.length = 0;
915 hpk.hpk_data_version = 0;
918 /* For archive request, we need to read the current file version. */
919 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
921 __u64 data_version = 0;
923 /* Get inode for this fid */
924 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
926 hpk.hpk_flags |= HP_FLAG_RETRY;
927 /* hpk_errval is >= 0 */
928 hpk.hpk_errval = -PTR_ERR(inode);
933 /* Read current file data version */
934 rc = ll_data_version(inode, &data_version, 1);
937 CDEBUG(D_HSM, "Could not read file data version of "
938 DFID" (rc = %d). Archive request (%#llx) could not be done.\n",
939 PFID(©->hc_hai.hai_fid), rc,
940 copy->hc_hai.hai_cookie);
941 hpk.hpk_flags |= HP_FLAG_RETRY;
942 /* hpk_errval must be >= 0 */
943 hpk.hpk_errval = -rc;
947 /* Store it the hsm_copy for later copytool use.
948 * Always modified even if no lsm. */
949 copy->hc_data_version = data_version;
953 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
960 * Generic handler to do any post-copy work.
962 * It will send the last hsm_progress update to coordinator to inform it
963 * that copy is finished and whether it was successful or not.
966 * - for ARCHIVE request, it will sample the file data version and compare it
967 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
968 * will be considered as failed.
969 * - for RESTORE request, it will sample the file data version and send it to
970 * coordinator which is useful if the file was imported as 'released'.
972 * \return 0 on success.
974 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
976 struct ll_sb_info *sbi = ll_s2sbi(sb);
977 struct hsm_progress_kernel hpk;
980 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
981 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
982 * initialized if copy_end was called with copy == NULL.
985 /* Forge a hsm_progress based on data from copy. */
986 hpk.hpk_fid = copy->hc_hai.hai_fid;
987 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
988 hpk.hpk_extent = copy->hc_hai.hai_extent;
989 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
990 hpk.hpk_errval = copy->hc_errval;
991 hpk.hpk_data_version = 0;
993 /* For archive request, we need to check the file data was not changed.
995 * For restore request, we need to send the file data version, this is
996 * useful when the file was created using hsm_import.
998 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
999 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
1000 (copy->hc_errval == 0)) {
1001 struct inode *inode;
1002 __u64 data_version = 0;
1004 /* Get lsm for this fid */
1005 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
1006 if (IS_ERR(inode)) {
1007 hpk.hpk_flags |= HP_FLAG_RETRY;
1008 /* hpk_errval must be >= 0 */
1009 hpk.hpk_errval = -PTR_ERR(inode);
1010 rc = PTR_ERR(inode);
1014 rc = ll_data_version(inode, &data_version,
1015 copy->hc_hai.hai_action == HSMA_ARCHIVE);
1018 CDEBUG(D_HSM, "Could not read file data version. Request could not be confirmed.\n");
1019 if (hpk.hpk_errval == 0)
1020 hpk.hpk_errval = -rc;
1024 /* Store it the hsm_copy for later copytool use.
1025 * Always modified even if no lsm. */
1026 hpk.hpk_data_version = data_version;
1028 /* File could have been stripped during archiving, so we need
1029 * to check anyway. */
1030 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1031 (copy->hc_data_version != data_version)) {
1032 CDEBUG(D_HSM, "File data version mismatched. File content was changed during archiving. "
1033 DFID", start:%#llx current:%#llx\n",
1034 PFID(©->hc_hai.hai_fid),
1035 copy->hc_data_version, data_version);
1036 /* File was changed, send error to cdt. Do not ask for
1037 * retry because if a file is modified frequently,
1038 * the cdt will loop on retried archive requests.
1039 * The policy engine will ask for a new archive later
1040 * when the file will not be modified for some tunable
1042 /* we do not notify caller */
1043 hpk.hpk_flags &= ~HP_FLAG_RETRY;
1044 /* hpk_errval must be >= 0 */
1045 hpk.hpk_errval = EBUSY;
1051 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1058 static int copy_and_ioctl(int cmd, struct obd_export *exp,
1059 const void __user *data, size_t size)
1064 copy = kzalloc(size, GFP_NOFS);
1068 if (copy_from_user(copy, data, size)) {
1073 rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1075 OBD_FREE(copy, size);
1080 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1082 int cmd = qctl->qc_cmd;
1083 int type = qctl->qc_type;
1084 int id = qctl->qc_id;
1085 int valid = qctl->qc_valid;
1089 case LUSTRE_Q_INVALIDATE:
1090 case LUSTRE_Q_FINVALIDATE:
1095 if (!capable(CFS_CAP_SYS_ADMIN) ||
1096 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1100 if (((type == USRQUOTA &&
1101 !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
1102 (type == GRPQUOTA &&
1103 !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
1104 (!capable(CFS_CAP_SYS_ADMIN) ||
1105 sbi->ll_flags & LL_SBI_RMT_CLIENT))
1111 CERROR("unsupported quotactl op: %#x\n", cmd);
1115 if (valid != QC_GENERAL) {
1116 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1119 if (cmd == Q_GETINFO)
1120 qctl->qc_cmd = Q_GETOINFO;
1121 else if (cmd == Q_GETQUOTA)
1122 qctl->qc_cmd = Q_GETOQUOTA;
1128 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1129 sizeof(*qctl), qctl, NULL);
1132 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1133 sizeof(*qctl), qctl, NULL);
1136 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1137 sizeof(*qctl), qctl, NULL);
1139 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1141 sizeof(*qctl), qctl, NULL);
1153 struct obd_quotactl *oqctl;
1155 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1159 QCTL_COPY(oqctl, qctl);
1160 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1162 if (rc != -EALREADY && cmd == Q_QUOTAON) {
1163 oqctl->qc_cmd = Q_QUOTAOFF;
1164 obd_quotactl(sbi->ll_md_exp, oqctl);
1166 OBD_FREE_PTR(oqctl);
1169 /* If QIF_SPACE is not set, client should collect the
1170 * space usage from OSSs by itself */
1171 if (cmd == Q_GETQUOTA &&
1172 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1173 !oqctl->qc_dqblk.dqb_curspace) {
1174 struct obd_quotactl *oqctl_tmp;
1176 oqctl_tmp = kzalloc(sizeof(*oqctl_tmp), GFP_NOFS);
1182 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1183 oqctl_tmp->qc_id = oqctl->qc_id;
1184 oqctl_tmp->qc_type = oqctl->qc_type;
1186 /* collect space usage from OSTs */
1187 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1188 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1189 if (!rc || rc == -EREMOTEIO) {
1190 oqctl->qc_dqblk.dqb_curspace =
1191 oqctl_tmp->qc_dqblk.dqb_curspace;
1192 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1195 /* collect space & inode usage from MDTs */
1196 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1197 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1198 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1199 if (!rc || rc == -EREMOTEIO) {
1200 oqctl->qc_dqblk.dqb_curspace +=
1201 oqctl_tmp->qc_dqblk.dqb_curspace;
1202 oqctl->qc_dqblk.dqb_curinodes =
1203 oqctl_tmp->qc_dqblk.dqb_curinodes;
1204 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1206 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1209 OBD_FREE_PTR(oqctl_tmp);
1212 QCTL_COPY(qctl, oqctl);
1213 OBD_FREE_PTR(oqctl);
1220 ll_getname(const char __user *filename)
1223 char *tmp = __getname();
1226 return ERR_PTR(-ENOMEM);
1228 len = strncpy_from_user(tmp, filename, PATH_MAX);
1231 else if (len > PATH_MAX)
1232 ret = -ENAMETOOLONG;
1241 #define ll_putname(filename) __putname(filename)
1243 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1245 struct inode *inode = file_inode(file);
1246 struct ll_sb_info *sbi = ll_i2sbi(inode);
1247 struct obd_ioctl_data *data;
1250 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1251 inode->i_ino, inode->i_generation, inode, cmd);
1253 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1254 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1257 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1259 case FSFILT_IOC_GETFLAGS:
1260 case FSFILT_IOC_SETFLAGS:
1261 return ll_iocontrol(inode, file, cmd, arg);
1262 case FSFILT_IOC_GETVERSION_OLD:
1263 case FSFILT_IOC_GETVERSION:
1264 return put_user(inode->i_generation, (int *)arg);
1265 /* We need to special case any other ioctls we want to handle,
1266 * to send them to the MDS/OST as appropriate and to properly
1267 * network encode the arg field.
1268 case FSFILT_IOC_SETVERSION_OLD:
1269 case FSFILT_IOC_SETVERSION:
1271 case LL_IOC_GET_MDTIDX: {
1274 mdtidx = ll_get_mdt_idx(inode);
1278 if (put_user((int)mdtidx, (int *)arg))
1283 case IOC_MDC_LOOKUP: {
1284 struct ptlrpc_request *request = NULL;
1285 int namelen, len = 0;
1288 struct md_op_data *op_data;
1290 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1295 filename = data->ioc_inlbuf1;
1296 namelen = strlen(filename);
1299 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1304 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1305 0, LUSTRE_OPC_ANY, NULL);
1306 if (IS_ERR(op_data)) {
1307 rc = PTR_ERR(op_data);
1311 op_data->op_valid = OBD_MD_FLID;
1312 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1313 ll_finish_md_op_data(op_data);
1315 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1318 ptlrpc_req_finished(request);
1320 obd_ioctl_freedata(buf, len);
1323 case LL_IOC_LMV_SETSTRIPE: {
1324 struct lmv_user_md *lum;
1332 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1337 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1338 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0) {
1343 filename = data->ioc_inlbuf1;
1344 namelen = data->ioc_inllen1;
1347 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1351 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1352 lumlen = data->ioc_inllen2;
1354 if (lum->lum_magic != LMV_USER_MAGIC ||
1355 lumlen != sizeof(*lum)) {
1356 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1357 filename, lum->lum_magic, lumlen, -EFAULT);
1363 * ll_dir_setdirstripe will be used to set dir stripe
1364 * mdc_create--->mdt_reint_create (with dirstripe)
1366 rc = ll_dir_setdirstripe(inode, lum, filename);
1368 obd_ioctl_freedata(buf, len);
1372 case LL_IOC_LOV_SETSTRIPE: {
1373 struct lov_user_md_v3 lumv3;
1374 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1375 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1376 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1378 int set_default = 0;
1380 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1381 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1382 sizeof(lumv3p->lmm_objects[0]));
1383 /* first try with v1 which is smaller than v3 */
1384 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1387 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1388 if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1392 if (is_root_inode(inode))
1395 /* in v1 and v3 cases lumv1 points to data */
1396 rc = ll_dir_setstripe(inode, lumv1, set_default);
1400 case LL_IOC_LMV_GETSTRIPE: {
1401 struct lmv_user_md *lump = (struct lmv_user_md *)arg;
1402 struct lmv_user_md lum;
1403 struct lmv_user_md *tmp;
1408 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1411 if (lum.lum_magic != LMV_MAGIC_V1)
1414 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1415 tmp = kzalloc(lum_size, GFP_NOFS);
1422 tmp->lum_type = LMV_STRIPE_TYPE;
1423 tmp->lum_stripe_count = 1;
1424 mdtindex = ll_get_mdt_idx(inode);
1430 tmp->lum_stripe_offset = mdtindex;
1431 tmp->lum_objects[0].lum_mds = mdtindex;
1432 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1433 sizeof(struct lu_fid));
1434 if (copy_to_user((void *)arg, tmp, lum_size)) {
1440 OBD_FREE(tmp, lum_size);
1443 case LL_IOC_REMOVE_ENTRY: {
1444 char *filename = NULL;
1448 /* Here is a little hack to avoid sending REINT_RMENTRY to
1449 * unsupported server, which might crash the server(LU-2730),
1450 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1451 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1452 * server will support REINT_RMENTRY XXX*/
1453 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1456 filename = ll_getname((const char *)arg);
1457 if (IS_ERR(filename))
1458 return PTR_ERR(filename);
1460 namelen = strlen(filename);
1466 rc = ll_rmdir_entry(inode, filename, namelen);
1469 ll_putname(filename);
1472 case LL_IOC_LOV_SWAP_LAYOUTS:
1474 case LL_IOC_OBD_STATFS:
1475 return ll_obd_statfs(inode, (void *)arg);
1476 case LL_IOC_LOV_GETSTRIPE:
1477 case LL_IOC_MDC_GETINFO:
1478 case IOC_MDC_GETFILEINFO:
1479 case IOC_MDC_GETFILESTRIPE: {
1480 struct ptlrpc_request *request = NULL;
1481 struct lov_user_md *lump;
1482 struct lov_mds_md *lmm = NULL;
1483 struct mdt_body *body;
1484 char *filename = NULL;
1487 if (cmd == IOC_MDC_GETFILEINFO ||
1488 cmd == IOC_MDC_GETFILESTRIPE) {
1489 filename = ll_getname((const char *)arg);
1490 if (IS_ERR(filename))
1491 return PTR_ERR(filename);
1493 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1494 &lmmsize, &request);
1496 rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1500 body = req_capsule_server_get(&request->rq_pill,
1502 LASSERT(body != NULL);
1508 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1509 cmd == LL_IOC_MDC_GETINFO)) {
1516 if (cmd == IOC_MDC_GETFILESTRIPE ||
1517 cmd == LL_IOC_LOV_GETSTRIPE) {
1518 lump = (struct lov_user_md *)arg;
1520 struct lov_user_mds_data *lmdp;
1522 lmdp = (struct lov_user_mds_data *)arg;
1523 lump = &lmdp->lmd_lmm;
1525 if (copy_to_user(lump, lmm, lmmsize)) {
1526 if (copy_to_user(lump, lmm, sizeof(*lump))) {
1533 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1534 struct lov_user_mds_data *lmdp;
1537 st.st_dev = inode->i_sb->s_dev;
1538 st.st_mode = body->mode;
1539 st.st_nlink = body->nlink;
1540 st.st_uid = body->uid;
1541 st.st_gid = body->gid;
1542 st.st_rdev = body->rdev;
1543 st.st_size = body->size;
1544 st.st_blksize = PAGE_CACHE_SIZE;
1545 st.st_blocks = body->blocks;
1546 st.st_atime = body->atime;
1547 st.st_mtime = body->mtime;
1548 st.st_ctime = body->ctime;
1549 st.st_ino = inode->i_ino;
1551 lmdp = (struct lov_user_mds_data *)arg;
1552 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st))) {
1559 ptlrpc_req_finished(request);
1561 ll_putname(filename);
1564 case IOC_LOV_GETINFO: {
1565 struct lov_user_mds_data *lumd;
1566 struct lov_stripe_md *lsm;
1567 struct lov_user_md *lum;
1568 struct lov_mds_md *lmm;
1572 lumd = (struct lov_user_mds_data *)arg;
1573 lum = &lumd->lmd_lmm;
1575 rc = ll_get_max_mdsize(sbi, &lmmsize);
1579 OBD_ALLOC_LARGE(lmm, lmmsize);
1582 if (copy_from_user(lmm, lum, lmmsize)) {
1587 switch (lmm->lmm_magic) {
1588 case LOV_USER_MAGIC_V1:
1589 if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1591 /* swab objects first so that stripes num will be sane */
1592 lustre_swab_lov_user_md_objects(
1593 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1594 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1595 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1597 case LOV_USER_MAGIC_V3:
1598 if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1600 /* swab objects first so that stripes num will be sane */
1601 lustre_swab_lov_user_md_objects(
1602 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1603 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1604 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1611 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1617 /* Perform glimpse_size operation. */
1618 memset(&st, 0, sizeof(st));
1620 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1624 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st))) {
1630 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1632 OBD_FREE_LARGE(lmm, lmmsize);
1635 case OBD_IOC_LLOG_CATINFO: {
1638 case OBD_IOC_QUOTACHECK: {
1639 struct obd_quotactl *oqctl;
1642 if (!capable(CFS_CAP_SYS_ADMIN) ||
1643 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1646 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1649 oqctl->qc_type = arg;
1650 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1652 CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1656 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1658 CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1660 OBD_FREE_PTR(oqctl);
1663 case OBD_IOC_POLL_QUOTACHECK: {
1664 struct if_quotacheck *check;
1666 if (!capable(CFS_CAP_SYS_ADMIN) ||
1667 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1670 check = kzalloc(sizeof(*check), GFP_NOFS);
1674 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1677 CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1678 if (copy_to_user((void *)arg, check,
1680 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1684 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1687 CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1688 if (copy_to_user((void *)arg, check,
1690 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1694 OBD_FREE_PTR(check);
1697 case LL_IOC_QUOTACTL: {
1698 struct if_quotactl *qctl;
1700 qctl = kzalloc(sizeof(*qctl), GFP_NOFS);
1704 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl))) {
1709 rc = quotactl_ioctl(sbi, qctl);
1711 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1718 case OBD_IOC_GETDTNAME:
1719 case OBD_IOC_GETMDNAME:
1720 return ll_get_obd_name(inode, cmd, arg);
1721 case LL_IOC_FLUSHCTX:
1722 return ll_flush_ctx(inode);
1723 #ifdef CONFIG_FS_POSIX_ACL
1724 case LL_IOC_RMTACL: {
1725 if (sbi->ll_flags & LL_SBI_RMT_CLIENT && is_root_inode(inode)) {
1726 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1728 LASSERT(fd != NULL);
1729 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1731 fd->fd_flags |= LL_FILE_RMTACL;
1737 case LL_IOC_GETOBDCOUNT: {
1739 struct obd_export *exp;
1741 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1744 /* get ost count when count is zero, get mdt count otherwise */
1745 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1746 vallen = sizeof(count);
1747 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1748 KEY_TGT_COUNT, &vallen, &count, NULL);
1750 CERROR("get target count failed: %d\n", rc);
1754 if (copy_to_user((int *)arg, &count, sizeof(int)))
1759 case LL_IOC_PATH2FID:
1760 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1761 sizeof(struct lu_fid)))
1764 case LL_IOC_GET_CONNECT_FLAGS: {
1765 return obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void *)arg);
1767 case OBD_IOC_CHANGELOG_SEND:
1768 case OBD_IOC_CHANGELOG_CLEAR:
1769 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1770 sizeof(struct ioc_changelog));
1772 case OBD_IOC_FID2PATH:
1773 return ll_fid2path(inode, (void *)arg);
1774 case LL_IOC_HSM_REQUEST: {
1775 struct hsm_user_request *hur;
1778 hur = kzalloc(sizeof(*hur), GFP_NOFS);
1782 /* We don't know the true size yet; copy the fixed-size part */
1783 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1788 /* Compute the whole struct size */
1789 totalsize = hur_len(hur);
1794 /* Final size will be more than double totalsize */
1795 if (totalsize >= MDS_MAXREQSIZE / 3)
1798 OBD_ALLOC_LARGE(hur, totalsize);
1802 /* Copy the whole struct */
1803 if (copy_from_user(hur, (void *)arg, totalsize)) {
1804 OBD_FREE_LARGE(hur, totalsize);
1808 if (hur->hur_request.hr_action == HUA_RELEASE) {
1809 const struct lu_fid *fid;
1813 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1814 fid = &hur->hur_user_item[i].hui_fid;
1815 f = search_inode_for_lustre(inode->i_sb, fid);
1821 rc = ll_hsm_release(f);
1827 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1831 OBD_FREE_LARGE(hur, totalsize);
1835 case LL_IOC_HSM_PROGRESS: {
1836 struct hsm_progress_kernel hpk;
1837 struct hsm_progress hp;
1839 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1842 hpk.hpk_fid = hp.hp_fid;
1843 hpk.hpk_cookie = hp.hp_cookie;
1844 hpk.hpk_extent = hp.hp_extent;
1845 hpk.hpk_flags = hp.hp_flags;
1846 hpk.hpk_errval = hp.hp_errval;
1847 hpk.hpk_data_version = 0;
1849 /* File may not exist in Lustre; all progress
1850 * reported to Lustre root */
1851 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1855 case LL_IOC_HSM_CT_START:
1856 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1857 sizeof(struct lustre_kernelcomm));
1860 case LL_IOC_HSM_COPY_START: {
1861 struct hsm_copy *copy;
1864 copy = kzalloc(sizeof(*copy), GFP_NOFS);
1867 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1872 rc = ll_ioc_copy_start(inode->i_sb, copy);
1873 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1879 case LL_IOC_HSM_COPY_END: {
1880 struct hsm_copy *copy;
1883 copy = kzalloc(sizeof(*copy), GFP_NOFS);
1886 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1891 rc = ll_ioc_copy_end(inode->i_sb, copy);
1892 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1899 return obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL, (void *)arg);
1903 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1905 struct inode *inode = file->f_mapping->host;
1906 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1907 struct ll_sb_info *sbi = ll_i2sbi(inode);
1908 int api32 = ll_need_32bit_api(sbi);
1909 loff_t ret = -EINVAL;
1911 mutex_lock(&inode->i_mutex);
1916 offset += file->f_pos;
1922 offset += LL_DIR_END_OFF_32BIT;
1924 offset += LL_DIR_END_OFF;
1931 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1932 (!api32 && offset <= LL_DIR_END_OFF))) {
1933 if (offset != file->f_pos) {
1934 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1935 (!api32 && offset == LL_DIR_END_OFF))
1936 fd->lfd_pos = MDS_DIR_END_OFF;
1937 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1938 fd->lfd_pos = offset << 32;
1940 fd->lfd_pos = offset;
1941 file->f_pos = offset;
1942 file->f_version = 0;
1949 mutex_unlock(&inode->i_mutex);
1953 static int ll_dir_open(struct inode *inode, struct file *file)
1955 return ll_file_open(inode, file);
1958 static int ll_dir_release(struct inode *inode, struct file *file)
1960 return ll_file_release(inode, file);
1963 const struct file_operations ll_dir_operations = {
1964 .llseek = ll_dir_seek,
1965 .open = ll_dir_open,
1966 .release = ll_dir_release,
1967 .read = generic_read_dir,
1968 .iterate = ll_readdir,
1969 .unlocked_ioctl = ll_dir_ioctl,