4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Directory code for lustre client.
42 #include <linux/pagemap.h>
44 #include <asm/uaccess.h>
45 #include <linux/buffer_head.h> /* for wait_on_buffer */
46 #include <linux/pagevec.h>
47 #include <linux/prefetch.h>
49 #define DEBUG_SUBSYSTEM S_LLITE
51 #include "../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 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
188 /* Checked by mdc_readpage() */
189 LASSERT(body != NULL);
191 if (body->valid & OBD_MD_FLSIZE)
192 cl_isize_write(inode, body->size);
194 nrdpgs = (request->rq_bulk->bd_nob_transferred+PAGE_CACHE_SIZE-1)
196 SetPageUptodate(page0);
199 ptlrpc_req_finished(request);
201 CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
203 ll_pagevec_init(&lru_pvec, 0);
204 for (i = 1; i < npages; i++) {
205 unsigned long offset;
210 if (rc < 0 || i >= nrdpgs) {
211 page_cache_release(page);
215 SetPageUptodate(page);
218 hash = le64_to_cpu(dp->ldp_hash_start);
221 offset = hash_x_index(hash, hash64);
223 prefetchw(&page->flags);
224 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
228 if (ll_pagevec_add(&lru_pvec, page) == 0)
229 ll_pagevec_lru_add_file(&lru_pvec);
231 CDEBUG(D_VFSTRACE, "page %lu add to page cache failed: %d\n",
234 page_cache_release(page);
236 ll_pagevec_lru_add_file(&lru_pvec);
238 if (page_pool != &page0)
239 OBD_FREE(page_pool, sizeof(struct page *) * max_pages);
243 static void ll_check_page(struct inode *dir, struct page *page)
245 /* XXX: check page format later */
246 SetPageChecked(page);
249 void ll_release_page(struct page *page, int remove)
254 if (likely(page->mapping != NULL))
255 truncate_complete_page(page->mapping, page);
258 page_cache_release(page);
262 * Find, kmap and return page that contains given hash.
264 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
265 __u64 *start, __u64 *end)
267 int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
268 struct address_space *mapping = dir->i_mapping;
270 * Complement of hash is used as an index so that
271 * radix_tree_gang_lookup() can be used to find a page with starting
272 * hash _smaller_ than one we are looking for.
274 unsigned long offset = hash_x_index(*hash, hash64);
278 spin_lock_irq(&mapping->tree_lock);
279 found = radix_tree_gang_lookup(&mapping->page_tree,
280 (void **)&page, offset, 1);
282 struct lu_dirpage *dp;
284 page_cache_get(page);
285 spin_unlock_irq(&mapping->tree_lock);
287 * In contrast to find_lock_page() we are sure that directory
288 * page cannot be truncated (while DLM lock is held) and,
289 * hence, can avoid restart.
291 * In fact, page cannot be locked here at all, because
292 * ll_dir_filler() does synchronous io.
294 wait_on_page_locked(page);
295 if (PageUptodate(page)) {
297 if (BITS_PER_LONG == 32 && hash64) {
298 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
299 *end = le64_to_cpu(dp->ldp_hash_end) >> 32;
302 *start = le64_to_cpu(dp->ldp_hash_start);
303 *end = le64_to_cpu(dp->ldp_hash_end);
305 LASSERTF(*start <= *hash, "start = %#llx,end = %#llx,hash = %#llx\n",
306 *start, *end, *hash);
307 CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash %llu\n",
308 offset, *start, *end, *hash);
310 ll_release_page(page, 0);
312 } else if (*end != *start && *hash == *end) {
314 * upon hash collision, remove this page,
315 * otherwise put page reference, and
316 * ll_get_dir_page() will issue RPC to fetch
319 ll_release_page(page,
320 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
324 page_cache_release(page);
325 page = ERR_PTR(-EIO);
329 spin_unlock_irq(&mapping->tree_lock);
335 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
336 struct ll_dir_chain *chain)
338 ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
339 struct address_space *mapping = dir->i_mapping;
340 struct lustre_handle lockh;
341 struct lu_dirpage *dp;
348 struct ll_inode_info *lli = ll_i2info(dir);
349 int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
352 rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
353 ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
355 struct ldlm_enqueue_info einfo = {
356 .ei_type = LDLM_IBITS,
358 .ei_cb_bl = ll_md_blocking_ast,
359 .ei_cb_cp = ldlm_completion_ast,
361 struct lookup_intent it = { .it_op = IT_READDIR };
362 struct ptlrpc_request *request;
363 struct md_op_data *op_data;
365 op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
366 LUSTRE_OPC_ANY, NULL);
368 return (void *)op_data;
370 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
371 op_data, &lockh, NULL, 0, NULL, 0);
373 ll_finish_md_op_data(op_data);
375 request = (struct ptlrpc_request *)it.d.lustre.it_data;
377 ptlrpc_req_finished(request);
379 CERROR("lock enqueue: "DFID" at %llu: rc %d\n",
380 PFID(ll_inode2fid(dir)), hash, rc);
384 CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
385 dir, dir->i_ino, dir->i_generation);
386 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
387 &it.d.lustre.it_lock_handle, dir, NULL);
389 /* for cross-ref object, l_ast_data of the lock may not be set,
390 * we reset it here */
391 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
394 ldlm_lock_dump_handle(D_OTHER, &lockh);
396 mutex_lock(&lli->lli_readdir_mutex);
397 page = ll_dir_page_locate(dir, &lhash, &start, &end);
399 CERROR("dir page locate: "DFID" at %llu: rc %ld\n",
400 PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
402 } else if (page != NULL) {
404 * XXX nikita: not entirely correct handling of a corner case:
405 * suppose hash chain of entries with hash value HASH crosses
406 * border between pages P0 and P1. First both P0 and P1 are
407 * cached, seekdir() is called for some entry from the P0 part
408 * of the chain. Later P0 goes out of cache. telldir(HASH)
409 * happens and finds P1, as it starts with matching hash
410 * value. Remaining entries from P0 part of the chain are
411 * skipped. (Is that really a bug?)
413 * Possible solutions: 0. don't cache P1 is such case, handle
414 * it as an "overflow" page. 1. invalidate all pages at
415 * once. 2. use HASH|1 as an index for P1.
420 page = read_cache_page(mapping, hash_x_index(hash, hash64),
421 ll_dir_filler, &lhash);
423 CERROR("read cache page: "DFID" at %llu: rc %ld\n",
424 PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
428 wait_on_page_locked(page);
430 if (!PageUptodate(page)) {
431 CERROR("page not updated: "DFID" at %llu: rc %d\n",
432 PFID(ll_inode2fid(dir)), hash, -5);
435 if (!PageChecked(page))
436 ll_check_page(dir, page);
437 if (PageError(page)) {
438 CERROR("page error: "DFID" at %llu: rc %d\n",
439 PFID(ll_inode2fid(dir)), hash, -5);
443 dp = page_address(page);
444 if (BITS_PER_LONG == 32 && hash64) {
445 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
446 end = le64_to_cpu(dp->ldp_hash_end) >> 32;
449 start = le64_to_cpu(dp->ldp_hash_start);
450 end = le64_to_cpu(dp->ldp_hash_end);
454 LASSERT(start == lhash);
455 CWARN("Page-wide hash collision: %llu\n", end);
456 if (BITS_PER_LONG == 32 && hash64)
457 CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n",
458 le64_to_cpu(dp->ldp_hash_start),
459 le64_to_cpu(dp->ldp_hash_end), hash);
461 * Fetch whole overflow chain...
468 mutex_unlock(&lli->lli_readdir_mutex);
469 ldlm_lock_decref(&lockh, mode);
473 ll_release_page(page, 1);
474 page = ERR_PTR(-EIO);
478 int ll_dir_read(struct inode *inode, struct dir_context *ctx)
480 struct ll_inode_info *info = ll_i2info(inode);
481 struct ll_sb_info *sbi = ll_i2sbi(inode);
482 __u64 pos = ctx->pos;
483 int api32 = ll_need_32bit_api(sbi);
484 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
486 struct ll_dir_chain chain;
490 ll_dir_chain_init(&chain);
492 page = ll_get_dir_page(inode, pos, &chain);
494 while (rc == 0 && !done) {
495 struct lu_dirpage *dp;
496 struct lu_dirent *ent;
500 * If page is empty (end of directory is reached),
503 __u64 hash = MDS_DIR_END_OFF;
506 dp = page_address(page);
507 for (ent = lu_dirent_start(dp); ent != NULL && !done;
508 ent = lu_dirent_next(ent)) {
516 * XXX: implement correct swabbing here.
519 hash = le64_to_cpu(ent->lde_hash);
522 * Skip until we find target hash
527 namelen = le16_to_cpu(ent->lde_namelen);
538 fid_le_to_cpu(&fid, &ent->lde_fid);
539 ino = cl_fid_build_ino(&fid, api32);
540 type = ll_dirent_type_get(ent);
542 /* For 'll_nfs_get_name_filldir()', it will try
543 * to access the 'ent' through its 'lde_name',
544 * so the parameter 'name' for 'ctx->actor()'
545 * must be part of the 'ent'.
547 done = !dir_emit(ctx, ent->lde_name,
550 next = le64_to_cpu(dp->ldp_hash_end);
553 if (pos == MDS_DIR_END_OFF) {
555 * End of directory reached.
558 ll_release_page(page, 0);
559 } else if (1 /* chain is exhausted*/) {
561 * Normal case: continue to the next
564 ll_release_page(page,
565 le32_to_cpu(dp->ldp_flags) &
568 page = ll_get_dir_page(inode, pos,
572 * go into overflow page.
574 LASSERT(le32_to_cpu(dp->ldp_flags) &
576 ll_release_page(page, 1);
580 ll_release_page(page, 0);
584 CERROR("error reading dir "DFID" at %lu: rc %d\n",
585 PFID(&info->lli_fid), (unsigned long)pos, rc);
590 ll_dir_chain_fini(&chain);
594 static int ll_readdir(struct file *filp, struct dir_context *ctx)
596 struct inode *inode = file_inode(filp);
597 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
598 struct ll_sb_info *sbi = ll_i2sbi(inode);
599 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
600 int api32 = ll_need_32bit_api(sbi);
603 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
604 inode->i_ino, inode->i_generation,
605 inode, (unsigned long)lfd->lfd_pos, i_size_read(inode), api32);
607 if (lfd->lfd_pos == MDS_DIR_END_OFF) {
615 ctx->pos = lfd->lfd_pos;
616 rc = ll_dir_read(inode, ctx);
617 lfd->lfd_pos = ctx->pos;
618 if (ctx->pos == MDS_DIR_END_OFF) {
620 ctx->pos = LL_DIR_END_OFF_32BIT;
622 ctx->pos = LL_DIR_END_OFF;
627 filp->f_version = inode->i_version;
631 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
636 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
638 struct mgs_send_param *msp;
641 msp = kzalloc(sizeof(*msp), GFP_NOFS);
645 strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
646 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
647 sizeof(struct mgs_send_param), msp, NULL);
649 CERROR("Failed to set parameter: %d\n", rc);
655 int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
658 struct ptlrpc_request *request = NULL;
659 struct md_op_data *op_data;
660 struct ll_sb_info *sbi = ll_i2sbi(dir);
664 mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
665 op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
666 strlen(filename), mode, LUSTRE_OPC_MKDIR,
668 if (IS_ERR(op_data)) {
669 err = PTR_ERR(op_data);
673 op_data->op_cli_flags |= CLI_SET_MEA;
674 err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
675 from_kuid(&init_user_ns, current_fsuid()),
676 from_kgid(&init_user_ns, current_fsgid()),
677 cfs_curproc_cap_pack(), 0, &request);
678 ll_finish_md_op_data(op_data);
682 ptlrpc_req_finished(request);
686 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
689 struct ll_sb_info *sbi = ll_i2sbi(inode);
690 struct md_op_data *op_data;
691 struct ptlrpc_request *req = NULL;
693 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
694 struct obd_device *mgc = lsi->lsi_mgc;
699 * This is coming from userspace, so should be in
700 * local endian. But the MDS would like it in little
701 * endian, so we swab it before we send it.
703 switch (lump->lmm_magic) {
704 case LOV_USER_MAGIC_V1: {
705 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
706 lustre_swab_lov_user_md_v1(lump);
707 lum_size = sizeof(struct lov_user_md_v1);
710 case LOV_USER_MAGIC_V3: {
711 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
712 lustre_swab_lov_user_md_v3(
713 (struct lov_user_md_v3 *)lump);
714 lum_size = sizeof(struct lov_user_md_v3);
718 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
719 lump->lmm_magic, LOV_USER_MAGIC_V1,
725 lum_size = sizeof(struct lov_user_md_v1);
728 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
729 LUSTRE_OPC_ANY, NULL);
731 return PTR_ERR(op_data);
733 if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
734 op_data->op_cli_flags |= CLI_SET_MEA;
736 /* swabbing is done in lov_setstripe() on server side */
737 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
738 NULL, 0, &req, NULL);
739 ll_finish_md_op_data(op_data);
740 ptlrpc_req_finished(req);
742 if (rc != -EPERM && rc != -EACCES)
743 CERROR("mdc_setattr fails: rc = %d\n", rc);
746 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
747 LOV_USER_MAGIC_V3 have the same initial fields so we do not
748 need to make the distinction between the 2 versions */
749 if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
753 param = kzalloc(MGS_PARAM_MAXLEN, GFP_NOFS);
760 /* Get fsname and assume devname to be -MDT0000. */
761 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
762 strcat(buf, "-MDT0000.lov");
765 /* Set root stripesize */
766 sprintf(buf, ".stripesize=%u",
767 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
768 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
772 /* Set root stripecount */
773 sprintf(buf, ".stripecount=%hd",
774 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
775 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
779 /* Set root stripeoffset */
780 sprintf(buf, ".stripeoffset=%hd",
781 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
782 (typeof(lump->lmm_stripe_offset))(-1));
783 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
787 OBD_FREE(param, MGS_PARAM_MAXLEN);
792 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
793 int *lmm_size, struct ptlrpc_request **request)
795 struct ll_sb_info *sbi = ll_i2sbi(inode);
796 struct mdt_body *body;
797 struct lov_mds_md *lmm = NULL;
798 struct ptlrpc_request *req = NULL;
800 struct md_op_data *op_data;
802 rc = ll_get_default_mdsize(sbi, &lmmsize);
806 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
807 0, lmmsize, LUSTRE_OPC_ANY,
810 return PTR_ERR(op_data);
812 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
813 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
814 ll_finish_md_op_data(op_data);
816 CDEBUG(D_INFO, "md_getattr failed on inode %lu/%u: rc %d\n",
818 inode->i_generation, rc);
822 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
823 LASSERT(body != NULL);
825 lmmsize = body->eadatasize;
827 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
833 lmm = req_capsule_server_sized_get(&req->rq_pill,
834 &RMF_MDT_MD, lmmsize);
835 LASSERT(lmm != NULL);
838 * This is coming from the MDS, so is probably in
839 * little endian. We convert it to host endian before
840 * passing it to userspace.
842 /* We don't swab objects for directories */
843 switch (le32_to_cpu(lmm->lmm_magic)) {
845 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
846 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
849 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
850 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
853 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
864 * Get MDT index for the inode.
866 int ll_get_mdt_idx(struct inode *inode)
868 struct ll_sb_info *sbi = ll_i2sbi(inode);
869 struct md_op_data *op_data;
872 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
873 0, LUSTRE_OPC_ANY, NULL);
875 return PTR_ERR(op_data);
877 op_data->op_flags |= MF_GET_MDT_IDX;
878 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
879 mdtidx = op_data->op_mds;
880 ll_finish_md_op_data(op_data);
882 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
889 * Generic handler to do any pre-copy work.
891 * It send a first hsm_progress (with extent length == 0) to coordinator as a
892 * first information for it that real work has started.
894 * Moreover, for a ARCHIVE request, it will sample the file data version and
895 * store it in \a copy.
897 * \return 0 on success.
899 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
901 struct ll_sb_info *sbi = ll_s2sbi(sb);
902 struct hsm_progress_kernel hpk;
905 /* Forge a hsm_progress based on data from copy. */
906 hpk.hpk_fid = copy->hc_hai.hai_fid;
907 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
908 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
909 hpk.hpk_extent.length = 0;
912 hpk.hpk_data_version = 0;
915 /* For archive request, we need to read the current file version. */
916 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
918 __u64 data_version = 0;
920 /* Get inode for this fid */
921 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
923 hpk.hpk_flags |= HP_FLAG_RETRY;
924 /* hpk_errval is >= 0 */
925 hpk.hpk_errval = -PTR_ERR(inode);
930 /* Read current file data version */
931 rc = ll_data_version(inode, &data_version, 1);
934 CDEBUG(D_HSM, "Could not read file data version of "
935 DFID" (rc = %d). Archive request (%#llx) could not be done.\n",
936 PFID(©->hc_hai.hai_fid), rc,
937 copy->hc_hai.hai_cookie);
938 hpk.hpk_flags |= HP_FLAG_RETRY;
939 /* hpk_errval must be >= 0 */
940 hpk.hpk_errval = -rc;
944 /* Store it the hsm_copy for later copytool use.
945 * Always modified even if no lsm. */
946 copy->hc_data_version = data_version;
950 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
957 * Generic handler to do any post-copy work.
959 * It will send the last hsm_progress update to coordinator to inform it
960 * that copy is finished and whether it was successful or not.
963 * - for ARCHIVE request, it will sample the file data version and compare it
964 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
965 * will be considered as failed.
966 * - for RESTORE request, it will sample the file data version and send it to
967 * coordinator which is useful if the file was imported as 'released'.
969 * \return 0 on success.
971 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
973 struct ll_sb_info *sbi = ll_s2sbi(sb);
974 struct hsm_progress_kernel hpk;
977 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
978 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
979 * initialized if copy_end was called with copy == NULL.
982 /* Forge a hsm_progress based on data from copy. */
983 hpk.hpk_fid = copy->hc_hai.hai_fid;
984 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
985 hpk.hpk_extent = copy->hc_hai.hai_extent;
986 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
987 hpk.hpk_errval = copy->hc_errval;
988 hpk.hpk_data_version = 0;
990 /* For archive request, we need to check the file data was not changed.
992 * For restore request, we need to send the file data version, this is
993 * useful when the file was created using hsm_import.
995 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
996 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
997 (copy->hc_errval == 0)) {
999 __u64 data_version = 0;
1001 /* Get lsm for this fid */
1002 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
1003 if (IS_ERR(inode)) {
1004 hpk.hpk_flags |= HP_FLAG_RETRY;
1005 /* hpk_errval must be >= 0 */
1006 hpk.hpk_errval = -PTR_ERR(inode);
1007 rc = PTR_ERR(inode);
1011 rc = ll_data_version(inode, &data_version,
1012 copy->hc_hai.hai_action == HSMA_ARCHIVE);
1015 CDEBUG(D_HSM, "Could not read file data version. Request could not be confirmed.\n");
1016 if (hpk.hpk_errval == 0)
1017 hpk.hpk_errval = -rc;
1021 /* Store it the hsm_copy for later copytool use.
1022 * Always modified even if no lsm. */
1023 hpk.hpk_data_version = data_version;
1025 /* File could have been stripped during archiving, so we need
1026 * to check anyway. */
1027 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1028 (copy->hc_data_version != data_version)) {
1029 CDEBUG(D_HSM, "File data version mismatched. File content was changed during archiving. "
1030 DFID", start:%#llx current:%#llx\n",
1031 PFID(©->hc_hai.hai_fid),
1032 copy->hc_data_version, data_version);
1033 /* File was changed, send error to cdt. Do not ask for
1034 * retry because if a file is modified frequently,
1035 * the cdt will loop on retried archive requests.
1036 * The policy engine will ask for a new archive later
1037 * when the file will not be modified for some tunable
1039 /* we do not notify caller */
1040 hpk.hpk_flags &= ~HP_FLAG_RETRY;
1041 /* hpk_errval must be >= 0 */
1042 hpk.hpk_errval = EBUSY;
1048 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1055 static int copy_and_ioctl(int cmd, struct obd_export *exp,
1056 const void __user *data, size_t size)
1061 copy = kzalloc(size, GFP_NOFS);
1065 if (copy_from_user(copy, data, size)) {
1070 rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1072 OBD_FREE(copy, size);
1077 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1079 int cmd = qctl->qc_cmd;
1080 int type = qctl->qc_type;
1081 int id = qctl->qc_id;
1082 int valid = qctl->qc_valid;
1086 case LUSTRE_Q_INVALIDATE:
1087 case LUSTRE_Q_FINVALIDATE:
1092 if (!capable(CFS_CAP_SYS_ADMIN) ||
1093 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1097 if (((type == USRQUOTA &&
1098 !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
1099 (type == GRPQUOTA &&
1100 !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
1101 (!capable(CFS_CAP_SYS_ADMIN) ||
1102 sbi->ll_flags & LL_SBI_RMT_CLIENT))
1108 CERROR("unsupported quotactl op: %#x\n", cmd);
1112 if (valid != QC_GENERAL) {
1113 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1116 if (cmd == Q_GETINFO)
1117 qctl->qc_cmd = Q_GETOINFO;
1118 else if (cmd == Q_GETQUOTA)
1119 qctl->qc_cmd = Q_GETOQUOTA;
1125 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1126 sizeof(*qctl), qctl, NULL);
1129 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1130 sizeof(*qctl), qctl, NULL);
1133 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1134 sizeof(*qctl), qctl, NULL);
1136 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1138 sizeof(*qctl), qctl, NULL);
1150 struct obd_quotactl *oqctl;
1152 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1156 QCTL_COPY(oqctl, qctl);
1157 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1159 if (rc != -EALREADY && cmd == Q_QUOTAON) {
1160 oqctl->qc_cmd = Q_QUOTAOFF;
1161 obd_quotactl(sbi->ll_md_exp, oqctl);
1163 OBD_FREE_PTR(oqctl);
1166 /* If QIF_SPACE is not set, client should collect the
1167 * space usage from OSSs by itself */
1168 if (cmd == Q_GETQUOTA &&
1169 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1170 !oqctl->qc_dqblk.dqb_curspace) {
1171 struct obd_quotactl *oqctl_tmp;
1173 oqctl_tmp = kzalloc(sizeof(*oqctl_tmp), GFP_NOFS);
1179 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1180 oqctl_tmp->qc_id = oqctl->qc_id;
1181 oqctl_tmp->qc_type = oqctl->qc_type;
1183 /* collect space usage from OSTs */
1184 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1185 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1186 if (!rc || rc == -EREMOTEIO) {
1187 oqctl->qc_dqblk.dqb_curspace =
1188 oqctl_tmp->qc_dqblk.dqb_curspace;
1189 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1192 /* collect space & inode usage from MDTs */
1193 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1194 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1195 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1196 if (!rc || rc == -EREMOTEIO) {
1197 oqctl->qc_dqblk.dqb_curspace +=
1198 oqctl_tmp->qc_dqblk.dqb_curspace;
1199 oqctl->qc_dqblk.dqb_curinodes =
1200 oqctl_tmp->qc_dqblk.dqb_curinodes;
1201 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1203 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1206 OBD_FREE_PTR(oqctl_tmp);
1209 QCTL_COPY(qctl, oqctl);
1210 OBD_FREE_PTR(oqctl);
1217 ll_getname(const char __user *filename)
1220 char *tmp = __getname();
1223 return ERR_PTR(-ENOMEM);
1225 len = strncpy_from_user(tmp, filename, PATH_MAX);
1228 else if (len > PATH_MAX)
1229 ret = -ENAMETOOLONG;
1238 #define ll_putname(filename) __putname(filename)
1240 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1242 struct inode *inode = file_inode(file);
1243 struct ll_sb_info *sbi = ll_i2sbi(inode);
1244 struct obd_ioctl_data *data;
1247 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1248 inode->i_ino, inode->i_generation, inode, cmd);
1250 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1251 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1254 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1256 case FSFILT_IOC_GETFLAGS:
1257 case FSFILT_IOC_SETFLAGS:
1258 return ll_iocontrol(inode, file, cmd, arg);
1259 case FSFILT_IOC_GETVERSION_OLD:
1260 case FSFILT_IOC_GETVERSION:
1261 return put_user(inode->i_generation, (int *)arg);
1262 /* We need to special case any other ioctls we want to handle,
1263 * to send them to the MDS/OST as appropriate and to properly
1264 * network encode the arg field.
1265 case FSFILT_IOC_SETVERSION_OLD:
1266 case FSFILT_IOC_SETVERSION:
1268 case LL_IOC_GET_MDTIDX: {
1271 mdtidx = ll_get_mdt_idx(inode);
1275 if (put_user((int)mdtidx, (int *)arg))
1280 case IOC_MDC_LOOKUP: {
1281 struct ptlrpc_request *request = NULL;
1282 int namelen, len = 0;
1285 struct md_op_data *op_data;
1287 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1292 filename = data->ioc_inlbuf1;
1293 namelen = strlen(filename);
1296 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1301 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1302 0, LUSTRE_OPC_ANY, NULL);
1303 if (IS_ERR(op_data)) {
1304 rc = PTR_ERR(op_data);
1308 op_data->op_valid = OBD_MD_FLID;
1309 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1310 ll_finish_md_op_data(op_data);
1312 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1315 ptlrpc_req_finished(request);
1317 obd_ioctl_freedata(buf, len);
1320 case LL_IOC_LMV_SETSTRIPE: {
1321 struct lmv_user_md *lum;
1329 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1334 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1335 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0) {
1340 filename = data->ioc_inlbuf1;
1341 namelen = data->ioc_inllen1;
1344 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1348 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1349 lumlen = data->ioc_inllen2;
1351 if (lum->lum_magic != LMV_USER_MAGIC ||
1352 lumlen != sizeof(*lum)) {
1353 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1354 filename, lum->lum_magic, lumlen, -EFAULT);
1360 * ll_dir_setdirstripe will be used to set dir stripe
1361 * mdc_create--->mdt_reint_create (with dirstripe)
1363 rc = ll_dir_setdirstripe(inode, lum, filename);
1365 obd_ioctl_freedata(buf, len);
1369 case LL_IOC_LOV_SETSTRIPE: {
1370 struct lov_user_md_v3 lumv3;
1371 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1372 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1373 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1375 int set_default = 0;
1377 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1378 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1379 sizeof(lumv3p->lmm_objects[0]));
1380 /* first try with v1 which is smaller than v3 */
1381 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1384 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1385 if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1389 if (is_root_inode(inode))
1392 /* in v1 and v3 cases lumv1 points to data */
1393 rc = ll_dir_setstripe(inode, lumv1, set_default);
1397 case LL_IOC_LMV_GETSTRIPE: {
1398 struct lmv_user_md *lump = (struct lmv_user_md *)arg;
1399 struct lmv_user_md lum;
1400 struct lmv_user_md *tmp;
1405 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1408 if (lum.lum_magic != LMV_MAGIC_V1)
1411 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1412 tmp = kzalloc(lum_size, GFP_NOFS);
1419 tmp->lum_type = LMV_STRIPE_TYPE;
1420 tmp->lum_stripe_count = 1;
1421 mdtindex = ll_get_mdt_idx(inode);
1427 tmp->lum_stripe_offset = mdtindex;
1428 tmp->lum_objects[0].lum_mds = mdtindex;
1429 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1430 sizeof(struct lu_fid));
1431 if (copy_to_user((void *)arg, tmp, lum_size)) {
1437 OBD_FREE(tmp, lum_size);
1440 case LL_IOC_REMOVE_ENTRY: {
1441 char *filename = NULL;
1445 /* Here is a little hack to avoid sending REINT_RMENTRY to
1446 * unsupported server, which might crash the server(LU-2730),
1447 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1448 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1449 * server will support REINT_RMENTRY XXX*/
1450 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1453 filename = ll_getname((const char *)arg);
1454 if (IS_ERR(filename))
1455 return PTR_ERR(filename);
1457 namelen = strlen(filename);
1463 rc = ll_rmdir_entry(inode, filename, namelen);
1466 ll_putname(filename);
1469 case LL_IOC_LOV_SWAP_LAYOUTS:
1471 case LL_IOC_OBD_STATFS:
1472 return ll_obd_statfs(inode, (void *)arg);
1473 case LL_IOC_LOV_GETSTRIPE:
1474 case LL_IOC_MDC_GETINFO:
1475 case IOC_MDC_GETFILEINFO:
1476 case IOC_MDC_GETFILESTRIPE: {
1477 struct ptlrpc_request *request = NULL;
1478 struct lov_user_md *lump;
1479 struct lov_mds_md *lmm = NULL;
1480 struct mdt_body *body;
1481 char *filename = NULL;
1484 if (cmd == IOC_MDC_GETFILEINFO ||
1485 cmd == IOC_MDC_GETFILESTRIPE) {
1486 filename = ll_getname((const char *)arg);
1487 if (IS_ERR(filename))
1488 return PTR_ERR(filename);
1490 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1491 &lmmsize, &request);
1493 rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1497 body = req_capsule_server_get(&request->rq_pill,
1499 LASSERT(body != NULL);
1505 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1506 cmd == LL_IOC_MDC_GETINFO)) {
1513 if (cmd == IOC_MDC_GETFILESTRIPE ||
1514 cmd == LL_IOC_LOV_GETSTRIPE) {
1515 lump = (struct lov_user_md *)arg;
1517 struct lov_user_mds_data *lmdp;
1518 lmdp = (struct lov_user_mds_data *)arg;
1519 lump = &lmdp->lmd_lmm;
1521 if (copy_to_user(lump, lmm, lmmsize)) {
1522 if (copy_to_user(lump, lmm, sizeof(*lump))) {
1529 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1530 struct lov_user_mds_data *lmdp;
1533 st.st_dev = inode->i_sb->s_dev;
1534 st.st_mode = body->mode;
1535 st.st_nlink = body->nlink;
1536 st.st_uid = body->uid;
1537 st.st_gid = body->gid;
1538 st.st_rdev = body->rdev;
1539 st.st_size = body->size;
1540 st.st_blksize = PAGE_CACHE_SIZE;
1541 st.st_blocks = body->blocks;
1542 st.st_atime = body->atime;
1543 st.st_mtime = body->mtime;
1544 st.st_ctime = body->ctime;
1545 st.st_ino = inode->i_ino;
1547 lmdp = (struct lov_user_mds_data *)arg;
1548 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st))) {
1555 ptlrpc_req_finished(request);
1557 ll_putname(filename);
1560 case IOC_LOV_GETINFO: {
1561 struct lov_user_mds_data *lumd;
1562 struct lov_stripe_md *lsm;
1563 struct lov_user_md *lum;
1564 struct lov_mds_md *lmm;
1568 lumd = (struct lov_user_mds_data *)arg;
1569 lum = &lumd->lmd_lmm;
1571 rc = ll_get_max_mdsize(sbi, &lmmsize);
1575 OBD_ALLOC_LARGE(lmm, lmmsize);
1578 if (copy_from_user(lmm, lum, lmmsize)) {
1583 switch (lmm->lmm_magic) {
1584 case LOV_USER_MAGIC_V1:
1585 if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1587 /* swab objects first so that stripes num will be sane */
1588 lustre_swab_lov_user_md_objects(
1589 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1590 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1591 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1593 case LOV_USER_MAGIC_V3:
1594 if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1596 /* swab objects first so that stripes num will be sane */
1597 lustre_swab_lov_user_md_objects(
1598 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1599 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1600 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1607 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1613 /* Perform glimpse_size operation. */
1614 memset(&st, 0, sizeof(st));
1616 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1620 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st))) {
1626 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1628 OBD_FREE_LARGE(lmm, lmmsize);
1631 case OBD_IOC_LLOG_CATINFO: {
1634 case OBD_IOC_QUOTACHECK: {
1635 struct obd_quotactl *oqctl;
1638 if (!capable(CFS_CAP_SYS_ADMIN) ||
1639 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1642 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1645 oqctl->qc_type = arg;
1646 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1648 CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1652 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1654 CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1656 OBD_FREE_PTR(oqctl);
1659 case OBD_IOC_POLL_QUOTACHECK: {
1660 struct if_quotacheck *check;
1662 if (!capable(CFS_CAP_SYS_ADMIN) ||
1663 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1666 check = kzalloc(sizeof(*check), GFP_NOFS);
1670 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1673 CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1674 if (copy_to_user((void *)arg, check,
1676 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1680 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1683 CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1684 if (copy_to_user((void *)arg, check,
1686 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1690 OBD_FREE_PTR(check);
1693 case LL_IOC_QUOTACTL: {
1694 struct if_quotactl *qctl;
1696 qctl = kzalloc(sizeof(*qctl), GFP_NOFS);
1700 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl))) {
1705 rc = quotactl_ioctl(sbi, qctl);
1707 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1714 case OBD_IOC_GETDTNAME:
1715 case OBD_IOC_GETMDNAME:
1716 return ll_get_obd_name(inode, cmd, arg);
1717 case LL_IOC_FLUSHCTX:
1718 return ll_flush_ctx(inode);
1719 #ifdef CONFIG_FS_POSIX_ACL
1720 case LL_IOC_RMTACL: {
1721 if (sbi->ll_flags & LL_SBI_RMT_CLIENT && is_root_inode(inode)) {
1722 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1724 LASSERT(fd != NULL);
1725 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1727 fd->fd_flags |= LL_FILE_RMTACL;
1733 case LL_IOC_GETOBDCOUNT: {
1735 struct obd_export *exp;
1737 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1740 /* get ost count when count is zero, get mdt count otherwise */
1741 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1742 vallen = sizeof(count);
1743 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1744 KEY_TGT_COUNT, &vallen, &count, NULL);
1746 CERROR("get target count failed: %d\n", rc);
1750 if (copy_to_user((int *)arg, &count, sizeof(int)))
1755 case LL_IOC_PATH2FID:
1756 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1757 sizeof(struct lu_fid)))
1760 case LL_IOC_GET_CONNECT_FLAGS: {
1761 return obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void *)arg);
1763 case OBD_IOC_CHANGELOG_SEND:
1764 case OBD_IOC_CHANGELOG_CLEAR:
1765 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1766 sizeof(struct ioc_changelog));
1768 case OBD_IOC_FID2PATH:
1769 return ll_fid2path(inode, (void *)arg);
1770 case LL_IOC_HSM_REQUEST: {
1771 struct hsm_user_request *hur;
1774 hur = kzalloc(sizeof(*hur), GFP_NOFS);
1778 /* We don't know the true size yet; copy the fixed-size part */
1779 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1784 /* Compute the whole struct size */
1785 totalsize = hur_len(hur);
1790 /* Final size will be more than double totalsize */
1791 if (totalsize >= MDS_MAXREQSIZE / 3)
1794 OBD_ALLOC_LARGE(hur, totalsize);
1798 /* Copy the whole struct */
1799 if (copy_from_user(hur, (void *)arg, totalsize)) {
1800 OBD_FREE_LARGE(hur, totalsize);
1804 if (hur->hur_request.hr_action == HUA_RELEASE) {
1805 const struct lu_fid *fid;
1809 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1810 fid = &hur->hur_user_item[i].hui_fid;
1811 f = search_inode_for_lustre(inode->i_sb, fid);
1817 rc = ll_hsm_release(f);
1823 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1827 OBD_FREE_LARGE(hur, totalsize);
1831 case LL_IOC_HSM_PROGRESS: {
1832 struct hsm_progress_kernel hpk;
1833 struct hsm_progress hp;
1835 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1838 hpk.hpk_fid = hp.hp_fid;
1839 hpk.hpk_cookie = hp.hp_cookie;
1840 hpk.hpk_extent = hp.hp_extent;
1841 hpk.hpk_flags = hp.hp_flags;
1842 hpk.hpk_errval = hp.hp_errval;
1843 hpk.hpk_data_version = 0;
1845 /* File may not exist in Lustre; all progress
1846 * reported to Lustre root */
1847 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1851 case LL_IOC_HSM_CT_START:
1852 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1853 sizeof(struct lustre_kernelcomm));
1856 case LL_IOC_HSM_COPY_START: {
1857 struct hsm_copy *copy;
1860 copy = kzalloc(sizeof(*copy), GFP_NOFS);
1863 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1868 rc = ll_ioc_copy_start(inode->i_sb, copy);
1869 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1875 case LL_IOC_HSM_COPY_END: {
1876 struct hsm_copy *copy;
1879 copy = kzalloc(sizeof(*copy), GFP_NOFS);
1882 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1887 rc = ll_ioc_copy_end(inode->i_sb, copy);
1888 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1895 return obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL, (void *)arg);
1899 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1901 struct inode *inode = file->f_mapping->host;
1902 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1903 struct ll_sb_info *sbi = ll_i2sbi(inode);
1904 int api32 = ll_need_32bit_api(sbi);
1905 loff_t ret = -EINVAL;
1907 mutex_lock(&inode->i_mutex);
1912 offset += file->f_pos;
1918 offset += LL_DIR_END_OFF_32BIT;
1920 offset += LL_DIR_END_OFF;
1927 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1928 (!api32 && offset <= LL_DIR_END_OFF))) {
1929 if (offset != file->f_pos) {
1930 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1931 (!api32 && offset == LL_DIR_END_OFF))
1932 fd->lfd_pos = MDS_DIR_END_OFF;
1933 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1934 fd->lfd_pos = offset << 32;
1936 fd->lfd_pos = offset;
1937 file->f_pos = offset;
1938 file->f_version = 0;
1945 mutex_unlock(&inode->i_mutex);
1949 static int ll_dir_open(struct inode *inode, struct file *file)
1951 return ll_file_open(inode, file);
1954 static int ll_dir_release(struct inode *inode, struct file *file)
1956 return ll_file_release(inode, file);
1959 const struct file_operations ll_dir_operations = {
1960 .llseek = ll_dir_seek,
1961 .open = ll_dir_open,
1962 .release = ll_dir_release,
1963 .read = generic_read_dir,
1964 .iterate = ll_readdir,
1965 .unlocked_ioctl = ll_dir_ioctl,