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:"
232 " %d\n", offset, ret);
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 = filp->f_dentry->d_inode;
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 "
604 " 32bit_api %d\n", 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:"
719 " %#08x != %#08x nor %#08x\n",
720 lump->lmm_magic, LOV_USER_MAGIC_V1,
726 lum_size = sizeof(struct lov_user_md_v1);
729 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
730 LUSTRE_OPC_ANY, NULL);
732 return PTR_ERR(op_data);
734 if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
735 op_data->op_cli_flags |= CLI_SET_MEA;
737 /* swabbing is done in lov_setstripe() on server side */
738 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
739 NULL, 0, &req, NULL);
740 ll_finish_md_op_data(op_data);
741 ptlrpc_req_finished(req);
743 if (rc != -EPERM && rc != -EACCES)
744 CERROR("mdc_setattr fails: rc = %d\n", rc);
747 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
748 LOV_USER_MAGIC_V3 have the same initial fields so we do not
749 need to make the distinction between the 2 versions */
750 if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
754 param = kzalloc(MGS_PARAM_MAXLEN, GFP_NOFS);
761 /* Get fsname and assume devname to be -MDT0000. */
762 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
763 strcat(buf, "-MDT0000.lov");
766 /* Set root stripesize */
767 sprintf(buf, ".stripesize=%u",
768 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
769 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
773 /* Set root stripecount */
774 sprintf(buf, ".stripecount=%hd",
775 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
776 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
780 /* Set root stripeoffset */
781 sprintf(buf, ".stripeoffset=%hd",
782 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
783 (typeof(lump->lmm_stripe_offset))(-1));
784 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
788 OBD_FREE(param, MGS_PARAM_MAXLEN);
793 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
794 int *lmm_size, struct ptlrpc_request **request)
796 struct ll_sb_info *sbi = ll_i2sbi(inode);
797 struct mdt_body *body;
798 struct lov_mds_md *lmm = NULL;
799 struct ptlrpc_request *req = NULL;
801 struct md_op_data *op_data;
803 rc = ll_get_default_mdsize(sbi, &lmmsize);
807 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
808 0, lmmsize, LUSTRE_OPC_ANY,
811 return PTR_ERR(op_data);
813 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
814 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
815 ll_finish_md_op_data(op_data);
817 CDEBUG(D_INFO, "md_getattr failed on inode "
818 "%lu/%u: rc %d\n", inode->i_ino,
819 inode->i_generation, rc);
823 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
824 LASSERT(body != NULL);
826 lmmsize = body->eadatasize;
828 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
834 lmm = req_capsule_server_sized_get(&req->rq_pill,
835 &RMF_MDT_MD, lmmsize);
836 LASSERT(lmm != NULL);
839 * This is coming from the MDS, so is probably in
840 * little endian. We convert it to host endian before
841 * passing it to userspace.
843 /* We don't swab objects for directories */
844 switch (le32_to_cpu(lmm->lmm_magic)) {
846 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
847 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
850 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
851 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
854 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
865 * Get MDT index for the inode.
867 int ll_get_mdt_idx(struct inode *inode)
869 struct ll_sb_info *sbi = ll_i2sbi(inode);
870 struct md_op_data *op_data;
873 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
874 0, LUSTRE_OPC_ANY, NULL);
876 return PTR_ERR(op_data);
878 op_data->op_flags |= MF_GET_MDT_IDX;
879 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
880 mdtidx = op_data->op_mds;
881 ll_finish_md_op_data(op_data);
883 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
890 * Generic handler to do any pre-copy work.
892 * It send a first hsm_progress (with extent length == 0) to coordinator as a
893 * first information for it that real work has started.
895 * Moreover, for a ARCHIVE request, it will sample the file data version and
896 * store it in \a copy.
898 * \return 0 on success.
900 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
902 struct ll_sb_info *sbi = ll_s2sbi(sb);
903 struct hsm_progress_kernel hpk;
906 /* Forge a hsm_progress based on data from copy. */
907 hpk.hpk_fid = copy->hc_hai.hai_fid;
908 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
909 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
910 hpk.hpk_extent.length = 0;
913 hpk.hpk_data_version = 0;
916 /* For archive request, we need to read the current file version. */
917 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
919 __u64 data_version = 0;
921 /* Get inode for this fid */
922 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
924 hpk.hpk_flags |= HP_FLAG_RETRY;
925 /* hpk_errval is >= 0 */
926 hpk.hpk_errval = -PTR_ERR(inode);
931 /* Read current file data version */
932 rc = ll_data_version(inode, &data_version, 1);
935 CDEBUG(D_HSM, "Could not read file data version of "
936 DFID" (rc = %d). Archive request (%#llx) could not be done.\n",
937 PFID(©->hc_hai.hai_fid), rc,
938 copy->hc_hai.hai_cookie);
939 hpk.hpk_flags |= HP_FLAG_RETRY;
940 /* hpk_errval must be >= 0 */
941 hpk.hpk_errval = -rc;
945 /* Store it the hsm_copy for later copytool use.
946 * Always modified even if no lsm. */
947 copy->hc_data_version = data_version;
951 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
958 * Generic handler to do any post-copy work.
960 * It will send the last hsm_progress update to coordinator to inform it
961 * that copy is finished and whether it was successful or not.
964 * - for ARCHIVE request, it will sample the file data version and compare it
965 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
966 * will be considered as failed.
967 * - for RESTORE request, it will sample the file data version and send it to
968 * coordinator which is useful if the file was imported as 'released'.
970 * \return 0 on success.
972 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
974 struct ll_sb_info *sbi = ll_s2sbi(sb);
975 struct hsm_progress_kernel hpk;
978 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
979 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
980 * initialized if copy_end was called with copy == NULL.
983 /* Forge a hsm_progress based on data from copy. */
984 hpk.hpk_fid = copy->hc_hai.hai_fid;
985 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
986 hpk.hpk_extent = copy->hc_hai.hai_extent;
987 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
988 hpk.hpk_errval = copy->hc_errval;
989 hpk.hpk_data_version = 0;
991 /* For archive request, we need to check the file data was not changed.
993 * For restore request, we need to send the file data version, this is
994 * useful when the file was created using hsm_import.
996 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
997 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
998 (copy->hc_errval == 0)) {
1000 __u64 data_version = 0;
1002 /* Get lsm for this fid */
1003 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
1004 if (IS_ERR(inode)) {
1005 hpk.hpk_flags |= HP_FLAG_RETRY;
1006 /* hpk_errval must be >= 0 */
1007 hpk.hpk_errval = -PTR_ERR(inode);
1008 rc = PTR_ERR(inode);
1012 rc = ll_data_version(inode, &data_version,
1013 copy->hc_hai.hai_action == HSMA_ARCHIVE);
1016 CDEBUG(D_HSM, "Could not read file data version. "
1017 "Request could not be confirmed.\n");
1018 if (hpk.hpk_errval == 0)
1019 hpk.hpk_errval = -rc;
1023 /* Store it the hsm_copy for later copytool use.
1024 * Always modified even if no lsm. */
1025 hpk.hpk_data_version = data_version;
1027 /* File could have been stripped during archiving, so we need
1028 * to check anyway. */
1029 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1030 (copy->hc_data_version != data_version)) {
1031 CDEBUG(D_HSM, "File data version mismatched. "
1032 "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->f_dentry->d_inode;
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 (inode->i_sb->s_root == file->f_dentry)
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;
1521 lmdp = (struct lov_user_mds_data *)arg;
1522 lump = &lmdp->lmd_lmm;
1524 if (copy_to_user(lump, lmm, lmmsize)) {
1525 if (copy_to_user(lump, lmm, sizeof(*lump))) {
1532 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1533 struct lov_user_mds_data *lmdp;
1536 st.st_dev = inode->i_sb->s_dev;
1537 st.st_mode = body->mode;
1538 st.st_nlink = body->nlink;
1539 st.st_uid = body->uid;
1540 st.st_gid = body->gid;
1541 st.st_rdev = body->rdev;
1542 st.st_size = body->size;
1543 st.st_blksize = PAGE_CACHE_SIZE;
1544 st.st_blocks = body->blocks;
1545 st.st_atime = body->atime;
1546 st.st_mtime = body->mtime;
1547 st.st_ctime = body->ctime;
1548 st.st_ino = inode->i_ino;
1550 lmdp = (struct lov_user_mds_data *)arg;
1551 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st))) {
1558 ptlrpc_req_finished(request);
1560 ll_putname(filename);
1563 case IOC_LOV_GETINFO: {
1564 struct lov_user_mds_data *lumd;
1565 struct lov_stripe_md *lsm;
1566 struct lov_user_md *lum;
1567 struct lov_mds_md *lmm;
1571 lumd = (struct lov_user_mds_data *)arg;
1572 lum = &lumd->lmd_lmm;
1574 rc = ll_get_max_mdsize(sbi, &lmmsize);
1578 OBD_ALLOC_LARGE(lmm, lmmsize);
1581 if (copy_from_user(lmm, lum, lmmsize)) {
1586 switch (lmm->lmm_magic) {
1587 case LOV_USER_MAGIC_V1:
1588 if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1590 /* swab objects first so that stripes num will be sane */
1591 lustre_swab_lov_user_md_objects(
1592 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1593 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1594 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1596 case LOV_USER_MAGIC_V3:
1597 if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1599 /* swab objects first so that stripes num will be sane */
1600 lustre_swab_lov_user_md_objects(
1601 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1602 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1603 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1610 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1616 /* Perform glimpse_size operation. */
1617 memset(&st, 0, sizeof(st));
1619 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1623 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st))) {
1629 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1631 OBD_FREE_LARGE(lmm, lmmsize);
1634 case OBD_IOC_LLOG_CATINFO: {
1637 case OBD_IOC_QUOTACHECK: {
1638 struct obd_quotactl *oqctl;
1641 if (!capable(CFS_CAP_SYS_ADMIN) ||
1642 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1645 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1648 oqctl->qc_type = arg;
1649 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1651 CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1655 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1657 CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1659 OBD_FREE_PTR(oqctl);
1662 case OBD_IOC_POLL_QUOTACHECK: {
1663 struct if_quotacheck *check;
1665 if (!capable(CFS_CAP_SYS_ADMIN) ||
1666 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1669 check = kzalloc(sizeof(*check), GFP_NOFS);
1673 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1676 CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1677 if (copy_to_user((void *)arg, check,
1679 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1683 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1686 CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1687 if (copy_to_user((void *)arg, check,
1689 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1693 OBD_FREE_PTR(check);
1696 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1697 case LL_IOC_QUOTACTL_18: {
1698 /* copy the old 1.x quota struct for internal use, then copy
1699 * back into old format struct. For 1.8 compatibility. */
1700 struct if_quotactl_18 *qctl_18;
1701 struct if_quotactl *qctl_20;
1703 qctl_18 = kzalloc(sizeof(*qctl_18), GFP_NOFS);
1707 qctl_20 = kzalloc(sizeof(*qctl_20), GFP_NOFS);
1710 goto out_quotactl_18;
1713 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18))) {
1715 goto out_quotactl_20;
1718 QCTL_COPY(qctl_20, qctl_18);
1719 qctl_20->qc_idx = 0;
1721 /* XXX: dqb_valid was borrowed as a flag to mark that
1722 * only mds quota is wanted */
1723 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1724 qctl_18->qc_dqblk.dqb_valid) {
1725 qctl_20->qc_valid = QC_MDTIDX;
1726 qctl_20->qc_dqblk.dqb_valid = 0;
1727 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1728 qctl_20->qc_valid = QC_UUID;
1729 qctl_20->obd_uuid = qctl_18->obd_uuid;
1731 qctl_20->qc_valid = QC_GENERAL;
1734 rc = quotactl_ioctl(sbi, qctl_20);
1737 QCTL_COPY(qctl_18, qctl_20);
1738 qctl_18->obd_uuid = qctl_20->obd_uuid;
1740 if (copy_to_user((void *)arg, qctl_18,
1746 OBD_FREE_PTR(qctl_20);
1748 OBD_FREE_PTR(qctl_18);
1752 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1753 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1754 case LL_IOC_QUOTACTL: {
1755 struct if_quotactl *qctl;
1757 qctl = kzalloc(sizeof(*qctl), GFP_NOFS);
1761 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl))) {
1766 rc = quotactl_ioctl(sbi, qctl);
1768 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1775 case OBD_IOC_GETDTNAME:
1776 case OBD_IOC_GETMDNAME:
1777 return ll_get_obd_name(inode, cmd, arg);
1778 case LL_IOC_FLUSHCTX:
1779 return ll_flush_ctx(inode);
1780 #ifdef CONFIG_FS_POSIX_ACL
1781 case LL_IOC_RMTACL: {
1782 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1783 inode == inode->i_sb->s_root->d_inode) {
1784 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1786 LASSERT(fd != NULL);
1787 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1789 fd->fd_flags |= LL_FILE_RMTACL;
1795 case LL_IOC_GETOBDCOUNT: {
1797 struct obd_export *exp;
1799 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1802 /* get ost count when count is zero, get mdt count otherwise */
1803 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1804 vallen = sizeof(count);
1805 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1806 KEY_TGT_COUNT, &vallen, &count, NULL);
1808 CERROR("get target count failed: %d\n", rc);
1812 if (copy_to_user((int *)arg, &count, sizeof(int)))
1817 case LL_IOC_PATH2FID:
1818 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1819 sizeof(struct lu_fid)))
1822 case LL_IOC_GET_CONNECT_FLAGS: {
1823 return obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void *)arg);
1825 case OBD_IOC_CHANGELOG_SEND:
1826 case OBD_IOC_CHANGELOG_CLEAR:
1827 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1828 sizeof(struct ioc_changelog));
1830 case OBD_IOC_FID2PATH:
1831 return ll_fid2path(inode, (void *)arg);
1832 case LL_IOC_HSM_REQUEST: {
1833 struct hsm_user_request *hur;
1836 hur = kzalloc(sizeof(*hur), GFP_NOFS);
1840 /* We don't know the true size yet; copy the fixed-size part */
1841 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1846 /* Compute the whole struct size */
1847 totalsize = hur_len(hur);
1852 /* Final size will be more than double totalsize */
1853 if (totalsize >= MDS_MAXREQSIZE / 3)
1856 OBD_ALLOC_LARGE(hur, totalsize);
1860 /* Copy the whole struct */
1861 if (copy_from_user(hur, (void *)arg, totalsize)) {
1862 OBD_FREE_LARGE(hur, totalsize);
1866 if (hur->hur_request.hr_action == HUA_RELEASE) {
1867 const struct lu_fid *fid;
1871 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1872 fid = &hur->hur_user_item[i].hui_fid;
1873 f = search_inode_for_lustre(inode->i_sb, fid);
1879 rc = ll_hsm_release(f);
1885 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1889 OBD_FREE_LARGE(hur, totalsize);
1893 case LL_IOC_HSM_PROGRESS: {
1894 struct hsm_progress_kernel hpk;
1895 struct hsm_progress hp;
1897 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1900 hpk.hpk_fid = hp.hp_fid;
1901 hpk.hpk_cookie = hp.hp_cookie;
1902 hpk.hpk_extent = hp.hp_extent;
1903 hpk.hpk_flags = hp.hp_flags;
1904 hpk.hpk_errval = hp.hp_errval;
1905 hpk.hpk_data_version = 0;
1907 /* File may not exist in Lustre; all progress
1908 * reported to Lustre root */
1909 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1913 case LL_IOC_HSM_CT_START:
1914 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1915 sizeof(struct lustre_kernelcomm));
1918 case LL_IOC_HSM_COPY_START: {
1919 struct hsm_copy *copy;
1922 copy = kzalloc(sizeof(*copy), GFP_NOFS);
1925 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1930 rc = ll_ioc_copy_start(inode->i_sb, copy);
1931 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1937 case LL_IOC_HSM_COPY_END: {
1938 struct hsm_copy *copy;
1941 copy = kzalloc(sizeof(*copy), GFP_NOFS);
1944 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1949 rc = ll_ioc_copy_end(inode->i_sb, copy);
1950 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1957 return obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL, (void *)arg);
1961 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1963 struct inode *inode = file->f_mapping->host;
1964 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1965 struct ll_sb_info *sbi = ll_i2sbi(inode);
1966 int api32 = ll_need_32bit_api(sbi);
1967 loff_t ret = -EINVAL;
1969 mutex_lock(&inode->i_mutex);
1974 offset += file->f_pos;
1980 offset += LL_DIR_END_OFF_32BIT;
1982 offset += LL_DIR_END_OFF;
1989 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1990 (!api32 && offset <= LL_DIR_END_OFF))) {
1991 if (offset != file->f_pos) {
1992 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1993 (!api32 && offset == LL_DIR_END_OFF))
1994 fd->lfd_pos = MDS_DIR_END_OFF;
1995 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1996 fd->lfd_pos = offset << 32;
1998 fd->lfd_pos = offset;
1999 file->f_pos = offset;
2000 file->f_version = 0;
2007 mutex_unlock(&inode->i_mutex);
2011 static int ll_dir_open(struct inode *inode, struct file *file)
2013 return ll_file_open(inode, file);
2016 static int ll_dir_release(struct inode *inode, struct file *file)
2018 return ll_file_release(inode, file);
2021 const struct file_operations ll_dir_operations = {
2022 .llseek = ll_dir_seek,
2023 .open = ll_dir_open,
2024 .release = ll_dir_release,
2025 .read = generic_read_dir,
2026 .iterate = ll_readdir,
2027 .unlocked_ioctl = ll_dir_ioctl,