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 for (i = 1; i < npages; i++) {
207 unsigned long offset;
212 if (rc < 0 || i >= nrdpgs) {
213 page_cache_release(page);
217 SetPageUptodate(page);
220 hash = le64_to_cpu(dp->ldp_hash_start);
223 offset = hash_x_index(hash, hash64);
225 prefetchw(&page->flags);
226 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
231 CDEBUG(D_VFSTRACE, "page %lu add to page cache failed: %d\n",
234 page_cache_release(page);
237 if (page_pool != &page0)
242 static void ll_check_page(struct inode *dir, struct page *page)
244 /* XXX: check page format later */
245 SetPageChecked(page);
248 void ll_release_page(struct page *page, int remove)
253 if (likely(page->mapping != NULL))
254 truncate_complete_page(page->mapping, page);
257 page_cache_release(page);
261 * Find, kmap and return page that contains given hash.
263 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
264 __u64 *start, __u64 *end)
266 int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
267 struct address_space *mapping = dir->i_mapping;
269 * Complement of hash is used as an index so that
270 * radix_tree_gang_lookup() can be used to find a page with starting
271 * hash _smaller_ than one we are looking for.
273 unsigned long offset = hash_x_index(*hash, hash64);
277 spin_lock_irq(&mapping->tree_lock);
278 found = radix_tree_gang_lookup(&mapping->page_tree,
279 (void **)&page, offset, 1);
280 if (found > 0 && !radix_tree_exceptional_entry(page)) {
281 struct lu_dirpage *dp;
283 page_cache_get(page);
284 spin_unlock_irq(&mapping->tree_lock);
286 * In contrast to find_lock_page() we are sure that directory
287 * page cannot be truncated (while DLM lock is held) and,
288 * hence, can avoid restart.
290 * In fact, page cannot be locked here at all, because
291 * ll_dir_filler() does synchronous io.
293 wait_on_page_locked(page);
294 if (PageUptodate(page)) {
296 if (BITS_PER_LONG == 32 && hash64) {
297 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
298 *end = le64_to_cpu(dp->ldp_hash_end) >> 32;
301 *start = le64_to_cpu(dp->ldp_hash_start);
302 *end = le64_to_cpu(dp->ldp_hash_end);
304 LASSERTF(*start <= *hash, "start = %#llx,end = %#llx,hash = %#llx\n",
305 *start, *end, *hash);
306 CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash %llu\n",
307 offset, *start, *end, *hash);
309 ll_release_page(page, 0);
311 } else if (*end != *start && *hash == *end) {
313 * upon hash collision, remove this page,
314 * otherwise put page reference, and
315 * ll_get_dir_page() will issue RPC to fetch
318 ll_release_page(page,
319 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
323 page_cache_release(page);
324 page = ERR_PTR(-EIO);
328 spin_unlock_irq(&mapping->tree_lock);
334 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
335 struct ll_dir_chain *chain)
337 ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
338 struct address_space *mapping = dir->i_mapping;
339 struct lustre_handle lockh;
340 struct lu_dirpage *dp;
347 struct ll_inode_info *lli = ll_i2info(dir);
348 int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
351 rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
352 ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
354 struct ldlm_enqueue_info einfo = {
355 .ei_type = LDLM_IBITS,
357 .ei_cb_bl = ll_md_blocking_ast,
358 .ei_cb_cp = ldlm_completion_ast,
360 struct lookup_intent it = { .it_op = IT_READDIR };
361 struct ptlrpc_request *request;
362 struct md_op_data *op_data;
364 op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
365 LUSTRE_OPC_ANY, NULL);
367 return (void *)op_data;
369 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
370 op_data, &lockh, NULL, 0, NULL, 0);
372 ll_finish_md_op_data(op_data);
374 request = (struct ptlrpc_request *)it.d.lustre.it_data;
376 ptlrpc_req_finished(request);
378 CERROR("lock enqueue: "DFID" at %llu: rc %d\n",
379 PFID(ll_inode2fid(dir)), hash, rc);
383 CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
384 dir, dir->i_ino, dir->i_generation);
385 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
386 &it.d.lustre.it_lock_handle, dir, NULL);
388 /* for cross-ref object, l_ast_data of the lock may not be set,
389 * we reset it here */
390 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
393 ldlm_lock_dump_handle(D_OTHER, &lockh);
395 mutex_lock(&lli->lli_readdir_mutex);
396 page = ll_dir_page_locate(dir, &lhash, &start, &end);
398 CERROR("dir page locate: "DFID" at %llu: rc %ld\n",
399 PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
401 } else if (page != NULL) {
403 * XXX nikita: not entirely correct handling of a corner case:
404 * suppose hash chain of entries with hash value HASH crosses
405 * border between pages P0 and P1. First both P0 and P1 are
406 * cached, seekdir() is called for some entry from the P0 part
407 * of the chain. Later P0 goes out of cache. telldir(HASH)
408 * happens and finds P1, as it starts with matching hash
409 * value. Remaining entries from P0 part of the chain are
410 * skipped. (Is that really a bug?)
412 * Possible solutions: 0. don't cache P1 is such case, handle
413 * it as an "overflow" page. 1. invalidate all pages at
414 * once. 2. use HASH|1 as an index for P1.
419 page = read_cache_page(mapping, hash_x_index(hash, hash64),
420 ll_dir_filler, &lhash);
422 CERROR("read cache page: "DFID" at %llu: rc %ld\n",
423 PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
427 wait_on_page_locked(page);
429 if (!PageUptodate(page)) {
430 CERROR("page not updated: "DFID" at %llu: rc %d\n",
431 PFID(ll_inode2fid(dir)), hash, -5);
434 if (!PageChecked(page))
435 ll_check_page(dir, page);
436 if (PageError(page)) {
437 CERROR("page error: "DFID" at %llu: rc %d\n",
438 PFID(ll_inode2fid(dir)), hash, -5);
442 dp = page_address(page);
443 if (BITS_PER_LONG == 32 && hash64) {
444 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
445 end = le64_to_cpu(dp->ldp_hash_end) >> 32;
448 start = le64_to_cpu(dp->ldp_hash_start);
449 end = le64_to_cpu(dp->ldp_hash_end);
453 LASSERT(start == lhash);
454 CWARN("Page-wide hash collision: %llu\n", end);
455 if (BITS_PER_LONG == 32 && hash64)
456 CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n",
457 le64_to_cpu(dp->ldp_hash_start),
458 le64_to_cpu(dp->ldp_hash_end), hash);
460 * Fetch whole overflow chain...
467 mutex_unlock(&lli->lli_readdir_mutex);
468 ldlm_lock_decref(&lockh, mode);
472 ll_release_page(page, 1);
473 page = ERR_PTR(-EIO);
477 int ll_dir_read(struct inode *inode, struct dir_context *ctx)
479 struct ll_inode_info *info = ll_i2info(inode);
480 struct ll_sb_info *sbi = ll_i2sbi(inode);
481 __u64 pos = ctx->pos;
482 int api32 = ll_need_32bit_api(sbi);
483 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
485 struct ll_dir_chain chain;
489 ll_dir_chain_init(&chain);
491 page = ll_get_dir_page(inode, pos, &chain);
493 while (rc == 0 && !done) {
494 struct lu_dirpage *dp;
495 struct lu_dirent *ent;
499 * If page is empty (end of directory is reached),
502 __u64 hash = MDS_DIR_END_OFF;
505 dp = page_address(page);
506 for (ent = lu_dirent_start(dp); ent != NULL && !done;
507 ent = lu_dirent_next(ent)) {
515 * XXX: implement correct swabbing here.
518 hash = le64_to_cpu(ent->lde_hash);
521 * Skip until we find target hash
526 namelen = le16_to_cpu(ent->lde_namelen);
537 fid_le_to_cpu(&fid, &ent->lde_fid);
538 ino = cl_fid_build_ino(&fid, api32);
539 type = ll_dirent_type_get(ent);
541 /* For 'll_nfs_get_name_filldir()', it will try
542 * to access the 'ent' through its 'lde_name',
543 * so the parameter 'name' for 'ctx->actor()'
544 * must be part of the 'ent'.
546 done = !dir_emit(ctx, ent->lde_name,
549 next = le64_to_cpu(dp->ldp_hash_end);
552 if (pos == MDS_DIR_END_OFF) {
554 * End of directory reached.
557 ll_release_page(page, 0);
558 } else if (1 /* chain is exhausted*/) {
560 * Normal case: continue to the next
563 ll_release_page(page,
564 le32_to_cpu(dp->ldp_flags) &
567 page = ll_get_dir_page(inode, pos,
571 * go into overflow page.
573 LASSERT(le32_to_cpu(dp->ldp_flags) &
575 ll_release_page(page, 1);
579 ll_release_page(page, 0);
583 CERROR("error reading dir "DFID" at %lu: rc %d\n",
584 PFID(&info->lli_fid), (unsigned long)pos, rc);
589 ll_dir_chain_fini(&chain);
593 static int ll_readdir(struct file *filp, struct dir_context *ctx)
595 struct inode *inode = file_inode(filp);
596 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
597 struct ll_sb_info *sbi = ll_i2sbi(inode);
598 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
599 int api32 = ll_need_32bit_api(sbi);
602 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
603 inode->i_ino, inode->i_generation,
604 inode, (unsigned long)lfd->lfd_pos, i_size_read(inode), api32);
606 if (lfd->lfd_pos == MDS_DIR_END_OFF) {
614 ctx->pos = lfd->lfd_pos;
615 rc = ll_dir_read(inode, ctx);
616 lfd->lfd_pos = ctx->pos;
617 if (ctx->pos == MDS_DIR_END_OFF) {
619 ctx->pos = LL_DIR_END_OFF_32BIT;
621 ctx->pos = LL_DIR_END_OFF;
626 filp->f_version = inode->i_version;
630 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
635 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
637 struct mgs_send_param *msp;
640 msp = kzalloc(sizeof(*msp), GFP_NOFS);
644 strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
645 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
646 sizeof(struct mgs_send_param), msp, NULL);
648 CERROR("Failed to set parameter: %d\n", rc);
654 static int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
657 struct ptlrpc_request *request = NULL;
658 struct md_op_data *op_data;
659 struct ll_sb_info *sbi = ll_i2sbi(dir);
663 mode = (0755 & ~current_umask()) | S_IFDIR;
664 op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
665 strlen(filename), mode, LUSTRE_OPC_MKDIR,
667 if (IS_ERR(op_data)) {
668 err = PTR_ERR(op_data);
672 op_data->op_cli_flags |= CLI_SET_MEA;
673 err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
674 from_kuid(&init_user_ns, current_fsuid()),
675 from_kgid(&init_user_ns, current_fsgid()),
676 cfs_curproc_cap_pack(), 0, &request);
677 ll_finish_md_op_data(op_data);
681 ptlrpc_req_finished(request);
685 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
688 struct ll_sb_info *sbi = ll_i2sbi(inode);
689 struct md_op_data *op_data;
690 struct ptlrpc_request *req = NULL;
692 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
693 struct obd_device *mgc = lsi->lsi_mgc;
698 * This is coming from userspace, so should be in
699 * local endian. But the MDS would like it in little
700 * endian, so we swab it before we send it.
702 switch (lump->lmm_magic) {
703 case LOV_USER_MAGIC_V1: {
704 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
705 lustre_swab_lov_user_md_v1(lump);
706 lum_size = sizeof(struct lov_user_md_v1);
709 case LOV_USER_MAGIC_V3: {
710 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
711 lustre_swab_lov_user_md_v3(
712 (struct lov_user_md_v3 *)lump);
713 lum_size = sizeof(struct lov_user_md_v3);
717 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
718 lump->lmm_magic, LOV_USER_MAGIC_V1,
724 lum_size = sizeof(struct lov_user_md_v1);
727 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
728 LUSTRE_OPC_ANY, NULL);
730 return PTR_ERR(op_data);
732 if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
733 op_data->op_cli_flags |= CLI_SET_MEA;
735 /* swabbing is done in lov_setstripe() on server side */
736 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
737 NULL, 0, &req, NULL);
738 ll_finish_md_op_data(op_data);
739 ptlrpc_req_finished(req);
741 if (rc != -EPERM && rc != -EACCES)
742 CERROR("mdc_setattr fails: rc = %d\n", rc);
745 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
746 LOV_USER_MAGIC_V3 have the same initial fields so we do not
747 need to make the distinction between the 2 versions */
748 if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
752 param = kzalloc(MGS_PARAM_MAXLEN, GFP_NOFS);
757 /* Get fsname and assume devname to be -MDT0000. */
758 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
759 strcat(buf, "-MDT0000.lov");
762 /* Set root stripesize */
763 sprintf(buf, ".stripesize=%u",
764 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
765 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
769 /* Set root stripecount */
770 sprintf(buf, ".stripecount=%hd",
771 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
772 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
776 /* Set root stripeoffset */
777 sprintf(buf, ".stripeoffset=%hd",
778 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
779 (typeof(lump->lmm_stripe_offset))(-1));
780 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
788 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
789 int *lmm_size, struct ptlrpc_request **request)
791 struct ll_sb_info *sbi = ll_i2sbi(inode);
792 struct mdt_body *body;
793 struct lov_mds_md *lmm = NULL;
794 struct ptlrpc_request *req = NULL;
796 struct md_op_data *op_data;
798 rc = ll_get_default_mdsize(sbi, &lmmsize);
802 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
803 0, lmmsize, LUSTRE_OPC_ANY,
806 return PTR_ERR(op_data);
808 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
809 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
810 ll_finish_md_op_data(op_data);
812 CDEBUG(D_INFO, "md_getattr failed on inode %lu/%u: rc %d\n",
814 inode->i_generation, rc);
818 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
819 LASSERT(body != NULL);
821 lmmsize = body->eadatasize;
823 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
829 lmm = req_capsule_server_sized_get(&req->rq_pill,
830 &RMF_MDT_MD, lmmsize);
831 LASSERT(lmm != NULL);
834 * This is coming from the MDS, so is probably in
835 * little endian. We convert it to host endian before
836 * passing it to userspace.
838 /* We don't swab objects for directories */
839 switch (le32_to_cpu(lmm->lmm_magic)) {
841 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
842 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
845 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
846 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
849 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
860 * Get MDT index for the inode.
862 int ll_get_mdt_idx(struct inode *inode)
864 struct ll_sb_info *sbi = ll_i2sbi(inode);
865 struct md_op_data *op_data;
868 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
869 0, LUSTRE_OPC_ANY, NULL);
871 return PTR_ERR(op_data);
873 op_data->op_flags |= MF_GET_MDT_IDX;
874 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
875 mdtidx = op_data->op_mds;
876 ll_finish_md_op_data(op_data);
878 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
885 * Generic handler to do any pre-copy work.
887 * It send a first hsm_progress (with extent length == 0) to coordinator as a
888 * first information for it that real work has started.
890 * Moreover, for a ARCHIVE request, it will sample the file data version and
891 * store it in \a copy.
893 * \return 0 on success.
895 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
897 struct ll_sb_info *sbi = ll_s2sbi(sb);
898 struct hsm_progress_kernel hpk;
901 /* Forge a hsm_progress based on data from copy. */
902 hpk.hpk_fid = copy->hc_hai.hai_fid;
903 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
904 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
905 hpk.hpk_extent.length = 0;
908 hpk.hpk_data_version = 0;
911 /* For archive request, we need to read the current file version. */
912 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
914 __u64 data_version = 0;
916 /* Get inode for this fid */
917 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
919 hpk.hpk_flags |= HP_FLAG_RETRY;
920 /* hpk_errval is >= 0 */
921 hpk.hpk_errval = -PTR_ERR(inode);
926 /* Read current file data version */
927 rc = ll_data_version(inode, &data_version, 1);
930 CDEBUG(D_HSM, "Could not read file data version of "
931 DFID" (rc = %d). Archive request (%#llx) could not be done.\n",
932 PFID(©->hc_hai.hai_fid), rc,
933 copy->hc_hai.hai_cookie);
934 hpk.hpk_flags |= HP_FLAG_RETRY;
935 /* hpk_errval must be >= 0 */
936 hpk.hpk_errval = -rc;
940 /* Store it the hsm_copy for later copytool use.
941 * Always modified even if no lsm. */
942 copy->hc_data_version = data_version;
946 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
953 * Generic handler to do any post-copy work.
955 * It will send the last hsm_progress update to coordinator to inform it
956 * that copy is finished and whether it was successful or not.
959 * - for ARCHIVE request, it will sample the file data version and compare it
960 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
961 * will be considered as failed.
962 * - for RESTORE request, it will sample the file data version and send it to
963 * coordinator which is useful if the file was imported as 'released'.
965 * \return 0 on success.
967 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
969 struct ll_sb_info *sbi = ll_s2sbi(sb);
970 struct hsm_progress_kernel hpk;
973 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
974 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
975 * initialized if copy_end was called with copy == NULL.
978 /* Forge a hsm_progress based on data from copy. */
979 hpk.hpk_fid = copy->hc_hai.hai_fid;
980 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
981 hpk.hpk_extent = copy->hc_hai.hai_extent;
982 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
983 hpk.hpk_errval = copy->hc_errval;
984 hpk.hpk_data_version = 0;
986 /* For archive request, we need to check the file data was not changed.
988 * For restore request, we need to send the file data version, this is
989 * useful when the file was created using hsm_import.
991 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
992 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
993 (copy->hc_errval == 0)) {
995 __u64 data_version = 0;
997 /* Get lsm for this fid */
998 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
1000 hpk.hpk_flags |= HP_FLAG_RETRY;
1001 /* hpk_errval must be >= 0 */
1002 hpk.hpk_errval = -PTR_ERR(inode);
1003 rc = PTR_ERR(inode);
1007 rc = ll_data_version(inode, &data_version,
1008 copy->hc_hai.hai_action == HSMA_ARCHIVE);
1011 CDEBUG(D_HSM, "Could not read file data version. Request could not be confirmed.\n");
1012 if (hpk.hpk_errval == 0)
1013 hpk.hpk_errval = -rc;
1017 /* Store it the hsm_copy for later copytool use.
1018 * Always modified even if no lsm. */
1019 hpk.hpk_data_version = data_version;
1021 /* File could have been stripped during archiving, so we need
1022 * to check anyway. */
1023 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1024 (copy->hc_data_version != data_version)) {
1025 CDEBUG(D_HSM, "File data version mismatched. File content was changed during archiving. "
1026 DFID", start:%#llx current:%#llx\n",
1027 PFID(©->hc_hai.hai_fid),
1028 copy->hc_data_version, data_version);
1029 /* File was changed, send error to cdt. Do not ask for
1030 * retry because if a file is modified frequently,
1031 * the cdt will loop on retried archive requests.
1032 * The policy engine will ask for a new archive later
1033 * when the file will not be modified for some tunable
1035 /* we do not notify caller */
1036 hpk.hpk_flags &= ~HP_FLAG_RETRY;
1037 /* hpk_errval must be >= 0 */
1038 hpk.hpk_errval = EBUSY;
1044 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1051 static int copy_and_ioctl(int cmd, struct obd_export *exp,
1052 const void __user *data, size_t size)
1057 copy = kzalloc(size, GFP_NOFS);
1061 if (copy_from_user(copy, data, size)) {
1066 rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1073 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1075 int cmd = qctl->qc_cmd;
1076 int type = qctl->qc_type;
1077 int id = qctl->qc_id;
1078 int valid = qctl->qc_valid;
1082 case LUSTRE_Q_INVALIDATE:
1083 case LUSTRE_Q_FINVALIDATE:
1088 if (!capable(CFS_CAP_SYS_ADMIN) ||
1089 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1093 if (((type == USRQUOTA &&
1094 !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
1095 (type == GRPQUOTA &&
1096 !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
1097 (!capable(CFS_CAP_SYS_ADMIN) ||
1098 sbi->ll_flags & LL_SBI_RMT_CLIENT))
1104 CERROR("unsupported quotactl op: %#x\n", cmd);
1108 if (valid != QC_GENERAL) {
1109 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1112 if (cmd == Q_GETINFO)
1113 qctl->qc_cmd = Q_GETOINFO;
1114 else if (cmd == Q_GETQUOTA)
1115 qctl->qc_cmd = Q_GETOQUOTA;
1121 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1122 sizeof(*qctl), qctl, NULL);
1125 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1126 sizeof(*qctl), qctl, NULL);
1129 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1130 sizeof(*qctl), qctl, NULL);
1132 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1134 sizeof(*qctl), qctl, NULL);
1146 struct obd_quotactl *oqctl;
1148 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1152 QCTL_COPY(oqctl, qctl);
1153 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1155 if (rc != -EALREADY && cmd == Q_QUOTAON) {
1156 oqctl->qc_cmd = Q_QUOTAOFF;
1157 obd_quotactl(sbi->ll_md_exp, oqctl);
1162 /* If QIF_SPACE is not set, client should collect the
1163 * space usage from OSSs by itself */
1164 if (cmd == Q_GETQUOTA &&
1165 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1166 !oqctl->qc_dqblk.dqb_curspace) {
1167 struct obd_quotactl *oqctl_tmp;
1169 oqctl_tmp = kzalloc(sizeof(*oqctl_tmp), GFP_NOFS);
1175 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1176 oqctl_tmp->qc_id = oqctl->qc_id;
1177 oqctl_tmp->qc_type = oqctl->qc_type;
1179 /* collect space usage from OSTs */
1180 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1181 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1182 if (!rc || rc == -EREMOTEIO) {
1183 oqctl->qc_dqblk.dqb_curspace =
1184 oqctl_tmp->qc_dqblk.dqb_curspace;
1185 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1188 /* collect space & inode usage from MDTs */
1189 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1190 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1191 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1192 if (!rc || rc == -EREMOTEIO) {
1193 oqctl->qc_dqblk.dqb_curspace +=
1194 oqctl_tmp->qc_dqblk.dqb_curspace;
1195 oqctl->qc_dqblk.dqb_curinodes =
1196 oqctl_tmp->qc_dqblk.dqb_curinodes;
1197 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1199 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1205 QCTL_COPY(qctl, oqctl);
1212 /* This function tries to get a single name component,
1213 * to send to the server. No actual path traversal involved,
1214 * so we limit to NAME_MAX */
1215 static char *ll_getname(const char __user *filename)
1220 tmp = kzalloc(NAME_MAX + 1, GFP_KERNEL);
1222 return ERR_PTR(-ENOMEM);
1224 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1229 else if (len > NAME_MAX && tmp[NAME_MAX] != 0)
1230 ret = -ENAMETOOLONG;
1239 #define ll_putname(filename) kfree(filename)
1241 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1243 struct inode *inode = file_inode(file);
1244 struct ll_sb_info *sbi = ll_i2sbi(inode);
1245 struct obd_ioctl_data *data;
1248 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1249 inode->i_ino, inode->i_generation, inode, cmd);
1251 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1252 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1255 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1257 case FSFILT_IOC_GETFLAGS:
1258 case FSFILT_IOC_SETFLAGS:
1259 return ll_iocontrol(inode, file, cmd, arg);
1260 case FSFILT_IOC_GETVERSION_OLD:
1261 case FSFILT_IOC_GETVERSION:
1262 return put_user(inode->i_generation, (int *)arg);
1263 /* We need to special case any other ioctls we want to handle,
1264 * to send them to the MDS/OST as appropriate and to properly
1265 * network encode the arg field.
1266 case FSFILT_IOC_SETVERSION_OLD:
1267 case FSFILT_IOC_SETVERSION:
1269 case LL_IOC_GET_MDTIDX: {
1272 mdtidx = ll_get_mdt_idx(inode);
1276 if (put_user((int)mdtidx, (int *)arg))
1281 case IOC_MDC_LOOKUP: {
1282 struct ptlrpc_request *request = NULL;
1283 int namelen, len = 0;
1286 struct md_op_data *op_data;
1288 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1293 filename = data->ioc_inlbuf1;
1294 namelen = strlen(filename);
1297 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1302 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1303 0, LUSTRE_OPC_ANY, NULL);
1304 if (IS_ERR(op_data)) {
1305 rc = PTR_ERR(op_data);
1309 op_data->op_valid = OBD_MD_FLID;
1310 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1311 ll_finish_md_op_data(op_data);
1313 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1316 ptlrpc_req_finished(request);
1318 obd_ioctl_freedata(buf, len);
1321 case LL_IOC_LMV_SETSTRIPE: {
1322 struct lmv_user_md *lum;
1330 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1335 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1336 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0) {
1341 filename = data->ioc_inlbuf1;
1342 namelen = data->ioc_inllen1;
1345 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1349 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1350 lumlen = data->ioc_inllen2;
1352 if (lum->lum_magic != LMV_USER_MAGIC ||
1353 lumlen != sizeof(*lum)) {
1354 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1355 filename, lum->lum_magic, lumlen, -EFAULT);
1361 * ll_dir_setdirstripe will be used to set dir stripe
1362 * mdc_create--->mdt_reint_create (with dirstripe)
1364 rc = ll_dir_setdirstripe(inode, lum, filename);
1366 obd_ioctl_freedata(buf, len);
1370 case LL_IOC_LOV_SETSTRIPE: {
1371 struct lov_user_md_v3 lumv3;
1372 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1373 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1374 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1376 int set_default = 0;
1378 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1379 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1380 sizeof(lumv3p->lmm_objects[0]));
1381 /* first try with v1 which is smaller than v3 */
1382 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1385 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1386 if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1390 if (is_root_inode(inode))
1393 /* in v1 and v3 cases lumv1 points to data */
1394 rc = ll_dir_setstripe(inode, lumv1, set_default);
1398 case LL_IOC_LMV_GETSTRIPE: {
1399 struct lmv_user_md *lump = (struct lmv_user_md *)arg;
1400 struct lmv_user_md lum;
1401 struct lmv_user_md *tmp;
1406 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1409 if (lum.lum_magic != LMV_MAGIC_V1)
1412 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1413 tmp = kzalloc(lum_size, GFP_NOFS);
1420 tmp->lum_type = LMV_STRIPE_TYPE;
1421 tmp->lum_stripe_count = 1;
1422 mdtindex = ll_get_mdt_idx(inode);
1428 tmp->lum_stripe_offset = mdtindex;
1429 tmp->lum_objects[0].lum_mds = mdtindex;
1430 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1431 sizeof(struct lu_fid));
1432 if (copy_to_user((void *)arg, tmp, lum_size)) {
1440 case LL_IOC_LOV_SWAP_LAYOUTS:
1442 case LL_IOC_OBD_STATFS:
1443 return ll_obd_statfs(inode, (void *)arg);
1444 case LL_IOC_LOV_GETSTRIPE:
1445 case LL_IOC_MDC_GETINFO:
1446 case IOC_MDC_GETFILEINFO:
1447 case IOC_MDC_GETFILESTRIPE: {
1448 struct ptlrpc_request *request = NULL;
1449 struct lov_user_md *lump;
1450 struct lov_mds_md *lmm = NULL;
1451 struct mdt_body *body;
1452 char *filename = NULL;
1455 if (cmd == IOC_MDC_GETFILEINFO ||
1456 cmd == IOC_MDC_GETFILESTRIPE) {
1457 filename = ll_getname((const char *)arg);
1458 if (IS_ERR(filename))
1459 return PTR_ERR(filename);
1461 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1462 &lmmsize, &request);
1464 rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1468 body = req_capsule_server_get(&request->rq_pill,
1470 LASSERT(body != NULL);
1476 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1477 cmd == LL_IOC_MDC_GETINFO)) {
1484 if (cmd == IOC_MDC_GETFILESTRIPE ||
1485 cmd == LL_IOC_LOV_GETSTRIPE) {
1486 lump = (struct lov_user_md *)arg;
1488 struct lov_user_mds_data *lmdp;
1490 lmdp = (struct lov_user_mds_data *)arg;
1491 lump = &lmdp->lmd_lmm;
1493 if (copy_to_user(lump, lmm, lmmsize)) {
1494 if (copy_to_user(lump, lmm, sizeof(*lump))) {
1501 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1502 struct lov_user_mds_data *lmdp;
1505 st.st_dev = inode->i_sb->s_dev;
1506 st.st_mode = body->mode;
1507 st.st_nlink = body->nlink;
1508 st.st_uid = body->uid;
1509 st.st_gid = body->gid;
1510 st.st_rdev = body->rdev;
1511 st.st_size = body->size;
1512 st.st_blksize = PAGE_CACHE_SIZE;
1513 st.st_blocks = body->blocks;
1514 st.st_atime = body->atime;
1515 st.st_mtime = body->mtime;
1516 st.st_ctime = body->ctime;
1517 st.st_ino = inode->i_ino;
1519 lmdp = (struct lov_user_mds_data *)arg;
1520 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st))) {
1527 ptlrpc_req_finished(request);
1529 ll_putname(filename);
1532 case IOC_LOV_GETINFO: {
1533 struct lov_user_mds_data *lumd;
1534 struct lov_stripe_md *lsm;
1535 struct lov_user_md *lum;
1536 struct lov_mds_md *lmm;
1540 lumd = (struct lov_user_mds_data *)arg;
1541 lum = &lumd->lmd_lmm;
1543 rc = ll_get_max_mdsize(sbi, &lmmsize);
1547 lmm = libcfs_kvzalloc(lmmsize, GFP_NOFS);
1550 if (copy_from_user(lmm, lum, lmmsize)) {
1555 switch (lmm->lmm_magic) {
1556 case LOV_USER_MAGIC_V1:
1557 if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1559 /* swab objects first so that stripes num will be sane */
1560 lustre_swab_lov_user_md_objects(
1561 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1562 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1563 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1565 case LOV_USER_MAGIC_V3:
1566 if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1568 /* swab objects first so that stripes num will be sane */
1569 lustre_swab_lov_user_md_objects(
1570 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1571 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1572 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1579 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1585 /* Perform glimpse_size operation. */
1586 memset(&st, 0, sizeof(st));
1588 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1592 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st))) {
1598 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1603 case OBD_IOC_LLOG_CATINFO: {
1606 case OBD_IOC_QUOTACHECK: {
1607 struct obd_quotactl *oqctl;
1610 if (!capable(CFS_CAP_SYS_ADMIN) ||
1611 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1614 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1617 oqctl->qc_type = arg;
1618 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1620 CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1624 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1626 CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1631 case OBD_IOC_POLL_QUOTACHECK: {
1632 struct if_quotacheck *check;
1634 if (!capable(CFS_CAP_SYS_ADMIN) ||
1635 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1638 check = kzalloc(sizeof(*check), GFP_NOFS);
1642 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1645 CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1646 if (copy_to_user((void *)arg, check,
1648 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1652 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1655 CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1656 if (copy_to_user((void *)arg, check,
1658 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1665 case LL_IOC_QUOTACTL: {
1666 struct if_quotactl *qctl;
1668 qctl = kzalloc(sizeof(*qctl), GFP_NOFS);
1672 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl))) {
1677 rc = quotactl_ioctl(sbi, qctl);
1679 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1686 case OBD_IOC_GETDTNAME:
1687 case OBD_IOC_GETMDNAME:
1688 return ll_get_obd_name(inode, cmd, arg);
1689 case LL_IOC_FLUSHCTX:
1690 return ll_flush_ctx(inode);
1691 #ifdef CONFIG_FS_POSIX_ACL
1692 case LL_IOC_RMTACL: {
1693 if (sbi->ll_flags & LL_SBI_RMT_CLIENT && is_root_inode(inode)) {
1694 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1696 LASSERT(fd != NULL);
1697 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1699 fd->fd_flags |= LL_FILE_RMTACL;
1705 case LL_IOC_GETOBDCOUNT: {
1707 struct obd_export *exp;
1709 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1712 /* get ost count when count is zero, get mdt count otherwise */
1713 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1714 vallen = sizeof(count);
1715 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1716 KEY_TGT_COUNT, &vallen, &count, NULL);
1718 CERROR("get target count failed: %d\n", rc);
1722 if (copy_to_user((int *)arg, &count, sizeof(int)))
1727 case LL_IOC_PATH2FID:
1728 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1729 sizeof(struct lu_fid)))
1732 case LL_IOC_GET_CONNECT_FLAGS: {
1733 return obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void *)arg);
1735 case OBD_IOC_CHANGELOG_SEND:
1736 case OBD_IOC_CHANGELOG_CLEAR:
1737 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1738 sizeof(struct ioc_changelog));
1740 case OBD_IOC_FID2PATH:
1741 return ll_fid2path(inode, (void *)arg);
1742 case LL_IOC_HSM_REQUEST: {
1743 struct hsm_user_request *hur;
1746 hur = memdup_user((void *)arg, sizeof(*hur));
1748 return PTR_ERR(hur);
1750 /* Compute the whole struct size */
1751 totalsize = hur_len(hur);
1756 /* Final size will be more than double totalsize */
1757 if (totalsize >= MDS_MAXREQSIZE / 3)
1760 hur = libcfs_kvzalloc(totalsize, GFP_NOFS);
1764 /* Copy the whole struct */
1765 if (copy_from_user(hur, (void *)arg, totalsize)) {
1770 if (hur->hur_request.hr_action == HUA_RELEASE) {
1771 const struct lu_fid *fid;
1775 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1776 fid = &hur->hur_user_item[i].hui_fid;
1777 f = search_inode_for_lustre(inode->i_sb, fid);
1783 rc = ll_hsm_release(f);
1789 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1797 case LL_IOC_HSM_PROGRESS: {
1798 struct hsm_progress_kernel hpk;
1799 struct hsm_progress hp;
1801 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1804 hpk.hpk_fid = hp.hp_fid;
1805 hpk.hpk_cookie = hp.hp_cookie;
1806 hpk.hpk_extent = hp.hp_extent;
1807 hpk.hpk_flags = hp.hp_flags;
1808 hpk.hpk_errval = hp.hp_errval;
1809 hpk.hpk_data_version = 0;
1811 /* File may not exist in Lustre; all progress
1812 * reported to Lustre root */
1813 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1817 case LL_IOC_HSM_CT_START:
1818 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1819 sizeof(struct lustre_kernelcomm));
1822 case LL_IOC_HSM_COPY_START: {
1823 struct hsm_copy *copy;
1826 copy = memdup_user((char *)arg, sizeof(*copy));
1828 return PTR_ERR(copy);
1830 rc = ll_ioc_copy_start(inode->i_sb, copy);
1831 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1837 case LL_IOC_HSM_COPY_END: {
1838 struct hsm_copy *copy;
1841 copy = memdup_user((char *)arg, sizeof(*copy));
1843 return PTR_ERR(copy);
1845 rc = ll_ioc_copy_end(inode->i_sb, copy);
1846 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1853 return obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL, (void *)arg);
1857 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1859 struct inode *inode = file->f_mapping->host;
1860 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1861 struct ll_sb_info *sbi = ll_i2sbi(inode);
1862 int api32 = ll_need_32bit_api(sbi);
1863 loff_t ret = -EINVAL;
1865 mutex_lock(&inode->i_mutex);
1870 offset += file->f_pos;
1876 offset += LL_DIR_END_OFF_32BIT;
1878 offset += LL_DIR_END_OFF;
1885 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1886 (!api32 && offset <= LL_DIR_END_OFF))) {
1887 if (offset != file->f_pos) {
1888 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1889 (!api32 && offset == LL_DIR_END_OFF))
1890 fd->lfd_pos = MDS_DIR_END_OFF;
1891 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1892 fd->lfd_pos = offset << 32;
1894 fd->lfd_pos = offset;
1895 file->f_pos = offset;
1896 file->f_version = 0;
1903 mutex_unlock(&inode->i_mutex);
1907 static int ll_dir_open(struct inode *inode, struct file *file)
1909 return ll_file_open(inode, file);
1912 static int ll_dir_release(struct inode *inode, struct file *file)
1914 return ll_file_release(inode, file);
1917 const struct file_operations ll_dir_operations = {
1918 .llseek = ll_dir_seek,
1919 .open = ll_dir_open,
1920 .release = ll_dir_release,
1921 .read = generic_read_dir,
1922 .iterate = ll_readdir,
1923 .unlocked_ioctl = ll_dir_ioctl,