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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 * Directory code for lustre client.
38 #include <linux/pagemap.h>
40 #include <linux/version.h>
41 #include <linux/security.h>
42 #include <linux/user_namespace.h>
43 #include <linux/uidgid.h>
44 #include <linux/uaccess.h>
45 #include <linux/buffer_head.h> // for wait_on_buffer
46 #include <linux/pagevec.h>
48 #define DEBUG_SUBSYSTEM S_LLITE
50 #include <obd_support.h>
51 #include <obd_class.h>
52 #include <uapi/linux/lustre/lustre_ioctl.h>
53 #include <lustre_lib.h>
54 #include <lustre_dlm.h>
55 #include <lustre_fid.h>
56 #include <lustre_kernelcomm.h>
57 #include <lustre_swab.h>
59 #include "llite_internal.h"
62 * (new) readdir implementation overview.
64 * Original lustre readdir implementation cached exact copy of raw directory
65 * pages on the client. These pages were indexed in client page cache by
66 * logical offset in the directory file. This design, while very simple and
67 * intuitive had some inherent problems:
69 * . it implies that byte offset to the directory entry serves as a
70 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
71 * ext3/htree directory entries may move due to splits, and more
74 * . it is incompatible with the design of split directories for cmd3,
75 * that assumes that names are distributed across nodes based on their
76 * hash, and so readdir should be done in hash order.
78 * New readdir implementation does readdir in hash order, and uses hash of a
79 * file name as a telldir/seekdir cookie. This led to number of complications:
81 * . hash is not unique, so it cannot be used to index cached directory
82 * pages on the client (note, that it requires a whole pageful of hash
83 * collided entries to cause two pages to have identical hashes);
85 * . hash is not unique, so it cannot, strictly speaking, be used as an
86 * entry cookie. ext3/htree has the same problem and lustre implementation
87 * mimics their solution: seekdir(hash) positions directory at the first
88 * entry with the given hash.
94 * Client caches directory pages using hash of the first entry as an index. As
95 * noted above hash is not unique, so this solution doesn't work as is:
96 * special processing is needed for "page hash chains" (i.e., sequences of
97 * pages filled with entries all having the same hash value).
99 * First, such chains have to be detected. To this end, server returns to the
100 * client the hash of the first entry on the page next to one returned. When
101 * client detects that this hash is the same as hash of the first entry on the
102 * returned page, page hash collision has to be handled. Pages in the
103 * hash chain, except first one, are termed "overflow pages".
105 * Solution to index uniqueness problem is to not cache overflow
106 * pages. Instead, when page hash collision is detected, all overflow pages
107 * from emerging chain are immediately requested from the server and placed in
108 * a special data structure (struct ll_dir_chain). This data structure is used
109 * by ll_readdir() to process entries from overflow pages. When readdir
110 * invocation finishes, overflow pages are discarded. If page hash collision
111 * chain weren't completely processed, next call to readdir will again detect
112 * page hash collision, again read overflow pages in, process next portion of
113 * entries and again discard the pages. This is not as wasteful as it looks,
114 * because, given reasonable hash, page hash collisions are extremely rare.
116 * 1. directory positioning
118 * When seekdir(hash) is called, original
129 * identification of and access to overflow pages
133 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
134 * a header lu_dirpage which describes the start/end hash, and whether this
135 * page is empty (contains no dir entry) or hash collide with next page.
136 * After client receives reply, several pages will be integrated into dir page
137 * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
138 * lu_dirpage for this integrated page will be adjusted. See
139 * mdc_adjust_dirpages().
142 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
143 __u64 offset, struct ll_dir_chain *chain)
145 struct md_callback cb_op;
149 cb_op.md_blocking_ast = ll_md_blocking_ast;
150 rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
157 void ll_release_page(struct inode *inode, struct page *page,
162 /* Always remove the page for striped dir, because the page is
163 * built from temporarily in LMV layer */
164 if (inode && ll_dir_striped(inode)) {
171 if (likely(page->mapping != NULL))
172 delete_from_page_cache(page);
179 * return IF_* type for given lu_dirent entry.
180 * IF_* flag shld be converted to particular OS file type in
181 * platform llite module.
183 static u16 ll_dirent_type_get(struct lu_dirent *ent)
186 struct luda_type *lt;
189 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
190 const unsigned align = sizeof(struct luda_type) - 1;
192 len = le16_to_cpu(ent->lde_namelen);
193 len = (len + align) & ~align;
194 lt = (void *)ent->lde_name + len;
195 type = IFTODT(le16_to_cpu(lt->lt_type));
201 #ifdef HAVE_DIR_CONTEXT
202 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
203 struct dir_context *ctx)
206 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
207 void *cookie, filldir_t filldir)
210 struct ll_sb_info *sbi = ll_i2sbi(inode);
212 bool is_api32 = ll_need_32bit_api(sbi);
213 bool is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
215 struct ll_dir_chain chain;
220 ll_dir_chain_init(&chain);
222 page = ll_get_dir_page(inode, op_data, pos, &chain);
224 while (rc == 0 && !done) {
225 struct lu_dirpage *dp;
226 struct lu_dirent *ent;
235 hash = MDS_DIR_END_OFF;
236 dp = page_address(page);
237 for (ent = lu_dirent_start(dp); ent != NULL && !done;
238 ent = lu_dirent_next(ent)) {
245 hash = le64_to_cpu(ent->lde_hash);
246 if (hash < pos) /* Skip until we find target hash */
249 namelen = le16_to_cpu(ent->lde_namelen);
250 if (namelen == 0) /* Skip dummy record */
253 if (is_api32 && is_hash64)
257 fid_le_to_cpu(&fid, &ent->lde_fid);
258 ino = cl_fid_build_ino(&fid, is_api32);
259 type = ll_dirent_type_get(ent);
260 /* For ll_nfs_get_name_filldir(), it will try to access
261 * 'ent' through 'lde_name', so the parameter 'name'
262 * for 'filldir()' must be part of the 'ent'. */
263 #ifdef HAVE_DIR_CONTEXT
265 done = !dir_emit(ctx, ent->lde_name, namelen, ino,
268 done = filldir(cookie, ent->lde_name, namelen, lhash,
275 ll_release_page(inode, page, false);
279 next = le64_to_cpu(dp->ldp_hash_end);
281 if (pos == MDS_DIR_END_OFF) {
283 * End of directory reached.
286 ll_release_page(inode, page, false);
289 * Normal case: continue to the next
292 ll_release_page(inode, page,
293 le32_to_cpu(dp->ldp_flags) &
296 page = ll_get_dir_page(inode, op_data, pos,
300 #ifdef HAVE_DIR_CONTEXT
305 ll_dir_chain_fini(&chain);
309 #ifdef HAVE_DIR_CONTEXT
310 static int ll_iterate(struct file *filp, struct dir_context *ctx)
312 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
315 struct inode *inode = file_inode(filp);
316 struct ll_file_data *lfd = filp->private_data;
317 struct ll_sb_info *sbi = ll_i2sbi(inode);
318 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
319 int api32 = ll_need_32bit_api(sbi);
320 struct md_op_data *op_data;
321 struct lu_fid pfid = { 0 };
322 ktime_t kstart = ktime_get();
333 "VFS Op:inode="DFID"(%p) pos/size%lu/%llu 32bit_api %d\n",
334 PFID(ll_inode2fid(inode)),
335 inode, (unsigned long)pos, i_size_read(inode), api32);
337 if (pos == MDS_DIR_END_OFF)
343 if (unlikely(ll_dir_striped(inode))) {
345 * This is only needed for striped dir to fill ..,
346 * see lmv_read_page()
348 if (file_dentry(filp)->d_parent != NULL &&
349 file_dentry(filp)->d_parent->d_inode != NULL) {
350 __u64 ibits = MDS_INODELOCK_LOOKUP;
351 struct inode *parent =
352 file_dentry(filp)->d_parent->d_inode;
354 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
355 pfid = *ll_inode2fid(parent);
358 /* If it can not find in cache, do lookup .. on the master
360 if (fid_is_zero(&pfid)) {
361 rc = ll_dir_get_parent_fid(inode, &pfid);
367 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
368 LUSTRE_OPC_ANY, inode);
370 GOTO(out, rc = PTR_ERR(op_data));
372 /* foreign dirs are browsed out of Lustre */
373 if (unlikely(op_data->op_mea1 != NULL &&
374 op_data->op_mea1->lsm_md_magic == LMV_MAGIC_FOREIGN)) {
375 ll_finish_md_op_data(op_data);
379 op_data->op_fid3 = pfid;
381 #ifdef HAVE_DIR_CONTEXT
383 rc = ll_dir_read(inode, &pos, op_data, ctx);
386 rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
391 if (pos == MDS_DIR_END_OFF) {
393 pos = LL_DIR_END_OFF_32BIT;
395 pos = LL_DIR_END_OFF;
400 #ifdef HAVE_DIR_CONTEXT
405 ll_finish_md_op_data(op_data);
409 ll_stats_ops_tally(sbi, LPROC_LL_READDIR,
410 ktime_us_delta(ktime_get(), kstart));
416 * Create striped directory with specified stripe(@lump)
418 * \param[in] dparent the parent of the directory.
419 * \param[in] lump the specified stripes.
420 * \param[in] dirname the name of the directory.
421 * \param[in] mode the specified mode of the directory.
423 * \retval =0 if striped directory is being created successfully.
424 * <0 if the creation is failed.
426 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
427 size_t len, const char *dirname, umode_t mode)
429 struct inode *parent = dparent->d_inode;
430 struct ptlrpc_request *request = NULL;
431 struct md_op_data *op_data;
432 struct ll_sb_info *sbi = ll_i2sbi(parent);
433 struct inode *inode = NULL;
434 struct dentry dentry = {
438 .len = strlen(dirname),
439 .hash = ll_full_name_hash(dparent, dirname,
446 if (unlikely(!lmv_user_magic_supported(lump->lum_magic)))
449 if (lump->lum_magic != LMV_MAGIC_FOREIGN) {
451 "VFS Op:inode="DFID"(%p) name %s stripe_offset %d, stripe_count: %u\n",
452 PFID(ll_inode2fid(parent)), parent, dirname,
453 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
455 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lump;
458 "VFS Op:inode="DFID"(%p) name %s foreign, length %u, value '%.*s'\n",
459 PFID(ll_inode2fid(parent)), parent, dirname,
460 lfm->lfm_length, lfm->lfm_length, lfm->lfm_value);
463 if (lump->lum_stripe_count > 1 &&
464 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
467 if (IS_DEADDIR(parent) &&
468 !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
471 if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_CRUSH)) {
472 if ((lump->lum_hash_type & LMV_HASH_TYPE_MASK) ==
473 LMV_HASH_TYPE_CRUSH) {
474 /* if server doesn't support 'crush' hash type,
475 * switch to fnv_1a_64.
477 lump->lum_hash_type &= ~LMV_HASH_TYPE_MASK;
478 lump->lum_hash_type |= LMV_HASH_TYPE_FNV_1A_64;
479 } else if ((lump->lum_hash_type & LMV_HASH_TYPE_MASK) ==
480 LMV_HASH_TYPE_UNKNOWN) {
481 /* from 2.14 MDT will choose default hash type if client
482 * doesn't set a valid one, while old server doesn't
485 lump->lum_hash_type &= ~LMV_HASH_TYPE_MASK;
486 lump->lum_hash_type |= LMV_HASH_TYPE_DEFAULT;
490 if (unlikely(!lmv_user_magic_supported(cpu_to_le32(lump->lum_magic))))
491 lustre_swab_lmv_user_md(lump);
493 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
494 mode &= ~current_umask();
495 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
496 op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
497 strlen(dirname), mode, LUSTRE_OPC_MKDIR,
500 RETURN(PTR_ERR(op_data));
502 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
503 /* selinux_dentry_init_security() uses dentry->d_parent and name
504 * to determine the security context for the file. So our fake
505 * dentry should be real enough for this purpose. */
506 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
507 &op_data->op_file_secctx_name,
508 &op_data->op_file_secctx,
509 &op_data->op_file_secctx_size);
511 GOTO(out_op_data, err);
514 op_data->op_cli_flags |= CLI_SET_MEA;
515 err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
516 from_kuid(&init_user_ns, current_fsuid()),
517 from_kgid(&init_user_ns, current_fsgid()),
518 cfs_curproc_cap_pack(), 0, &request);
520 GOTO(out_request, err);
522 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
524 err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
526 GOTO(out_inode, err);
528 dentry.d_inode = inode;
530 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
532 err = security_inode_notifysecctx(inode,
533 op_data->op_file_secctx,
534 op_data->op_file_secctx_size);
537 err = ll_inode_init_security(&dentry, inode, parent);
540 GOTO(out_inode, err);
546 ptlrpc_req_finished(request);
548 ll_finish_md_op_data(op_data);
553 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
556 struct ll_sb_info *sbi = ll_i2sbi(inode);
557 struct md_op_data *op_data;
558 struct ptlrpc_request *req = NULL;
564 switch (lump->lmm_magic) {
565 case LOV_USER_MAGIC_V1:
566 lum_size = sizeof(struct lov_user_md_v1);
568 case LOV_USER_MAGIC_V3:
569 lum_size = sizeof(struct lov_user_md_v3);
571 case LOV_USER_MAGIC_COMP_V1:
572 lum_size = ((struct lov_comp_md_v1 *)lump)->lcm_size;
575 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
576 lustre_swab_lmv_user_md(
577 (struct lmv_user_md *)lump);
578 lum_size = sizeof(struct lmv_user_md);
580 case LOV_USER_MAGIC_SPECIFIC: {
581 struct lov_user_md_v3 *v3 =
582 (struct lov_user_md_v3 *)lump;
583 if (v3->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
585 lum_size = lov_user_md_size(v3->lmm_stripe_count,
586 LOV_USER_MAGIC_SPECIFIC);
590 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
591 " %#08x != %#08x nor %#08x\n",
592 lump->lmm_magic, LOV_USER_MAGIC_V1,
598 * This is coming from userspace, so should be in
599 * local endian. But the MDS would like it in little
600 * endian, so we swab it before we send it.
602 if ((__swab32(lump->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
603 le32_to_cpu(LOV_MAGIC_MAGIC))
604 lustre_swab_lov_user_md(lump, 0);
606 lum_size = sizeof(struct lov_user_md_v1);
609 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
610 LUSTRE_OPC_ANY, NULL);
612 RETURN(PTR_ERR(op_data));
614 /* swabbing is done in lov_setstripe() on server side */
615 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
616 ll_finish_md_op_data(op_data);
617 ptlrpc_req_finished(req);
624 static int ll_dir_get_default_layout(struct inode *inode, void **plmm,
626 struct ptlrpc_request **request, u64 valid,
627 enum get_default_layout_type type)
629 struct ll_sb_info *sbi = ll_i2sbi(inode);
630 struct mdt_body *body;
631 struct lov_mds_md *lmm = NULL;
632 struct ptlrpc_request *req = NULL;
634 struct md_op_data *op_data;
638 rc = ll_get_default_mdsize(sbi, &lmm_size);
642 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
643 0, lmm_size, LUSTRE_OPC_ANY,
646 RETURN(PTR_ERR(op_data));
648 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
650 if (type == GET_DEFAULT_LAYOUT_ROOT) {
651 lu_root_fid(&op_data->op_fid1);
652 fid = op_data->op_fid1;
654 fid = *ll_inode2fid(inode);
657 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
658 ll_finish_md_op_data(op_data);
660 CDEBUG(D_INFO, "md_getattr failed on inode "DFID": rc %d\n",
665 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
666 LASSERT(body != NULL);
668 lmm_size = body->mbo_eadatasize;
670 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
672 GOTO(out, rc = -ENODATA);
675 lmm = req_capsule_server_sized_get(&req->rq_pill,
676 &RMF_MDT_MD, lmm_size);
677 LASSERT(lmm != NULL);
680 * This is coming from the MDS, so is probably in
681 * little endian. We convert it to host endian before
682 * passing it to userspace.
684 /* We don't swab objects for directories */
685 switch (le32_to_cpu(lmm->lmm_magic)) {
688 case LOV_MAGIC_COMP_V1:
689 case LOV_USER_MAGIC_SPECIFIC:
690 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
691 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
694 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
695 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
698 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
699 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
701 case LMV_MAGIC_FOREIGN: {
702 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lmm;
704 if (LMV_MAGIC_FOREIGN != cpu_to_le32(LMV_MAGIC_FOREIGN)) {
705 __swab32s(&lfm->lfm_magic);
706 __swab32s(&lfm->lfm_length);
707 __swab32s(&lfm->lfm_type);
708 __swab32s(&lfm->lfm_flags);
713 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
718 *plmm_size = lmm_size;
724 * This function will be used to get default LOV/LMV/Default LMV
725 * @valid will be used to indicate which stripe it will retrieve.
726 * If the directory does not have its own default layout, then the
727 * function will request the default layout from root FID.
728 * OBD_MD_MEA LMV stripe EA
729 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
730 * otherwise Default LOV EA.
731 * Each time, it can only retrieve 1 stripe EA
733 int ll_dir_getstripe_default(struct inode *inode, void **plmm, int *plmm_size,
734 struct ptlrpc_request **request,
735 struct ptlrpc_request **root_request,
738 struct ptlrpc_request *req = NULL;
739 struct ptlrpc_request *root_req = NULL;
740 struct lov_mds_md *lmm = NULL;
745 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
747 if (rc == -ENODATA && !fid_is_root(ll_inode2fid(inode)) &&
748 !(valid & (OBD_MD_MEA|OBD_MD_DEFAULT_MEA)) && root_request != NULL)
749 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
751 GET_DEFAULT_LAYOUT_ROOT);
754 *plmm_size = lmm_size;
756 if (root_request != NULL)
757 *root_request = root_req;
763 * This function will be used to get default LOV/LMV/Default LMV
764 * @valid will be used to indicate which stripe it will retrieve
765 * OBD_MD_MEA LMV stripe EA
766 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
767 * otherwise Default LOV EA.
768 * Each time, it can only retrieve 1 stripe EA
770 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
771 struct ptlrpc_request **request, u64 valid)
773 struct ptlrpc_request *req = NULL;
774 struct lov_mds_md *lmm = NULL;
779 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
783 *plmm_size = lmm_size;
789 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
791 struct md_op_data *op_data;
796 OBD_ALLOC_PTR(op_data);
800 op_data->op_flags |= MF_GET_MDT_IDX;
801 op_data->op_fid1 = *fid;
802 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
803 mdt_index = op_data->op_mds;
804 OBD_FREE_PTR(op_data);
812 * Get MDT index for the inode.
814 int ll_get_mdt_idx(struct inode *inode)
816 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
820 * Generic handler to do any pre-copy work.
822 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
823 * first information for it that real work has started.
825 * Moreover, for a ARCHIVE request, it will sample the file data version and
826 * store it in \a copy.
828 * \return 0 on success.
830 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
832 struct ll_sb_info *sbi = ll_s2sbi(sb);
833 struct hsm_progress_kernel hpk;
838 /* Forge a hsm_progress based on data from copy. */
839 hpk.hpk_fid = copy->hc_hai.hai_fid;
840 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
841 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
842 hpk.hpk_extent.length = 0;
845 hpk.hpk_data_version = 0;
848 /* For archive request, we need to read the current file version. */
849 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
851 __u64 data_version = 0;
853 /* Get inode for this fid */
854 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
856 hpk.hpk_flags |= HP_FLAG_RETRY;
857 /* hpk_errval is >= 0 */
858 hpk.hpk_errval = -PTR_ERR(inode);
859 GOTO(progress, rc = PTR_ERR(inode));
862 /* Read current file data version */
863 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
866 CDEBUG(D_HSM, "Could not read file data version of "
867 DFID" (rc = %d). Archive request ("
868 "%#llx) could not be done.\n",
869 PFID(©->hc_hai.hai_fid), rc,
870 copy->hc_hai.hai_cookie);
871 hpk.hpk_flags |= HP_FLAG_RETRY;
872 /* hpk_errval must be >= 0 */
873 hpk.hpk_errval = -rc;
877 /* Store in the hsm_copy for later copytool use.
878 * Always modified even if no lsm. */
879 copy->hc_data_version = data_version;
883 /* On error, the request should be considered as completed */
884 if (hpk.hpk_errval > 0)
885 hpk.hpk_flags |= HP_FLAG_COMPLETED;
887 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
890 /* Return first error */
891 RETURN(rc != 0 ? rc : rc2);
895 * Generic handler to do any post-copy work.
897 * It will send the last hsm_progress update to coordinator to inform it
898 * that copy is finished and whether it was successful or not.
901 * - for ARCHIVE request, it will sample the file data version and compare it
902 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
903 * will be considered as failed.
904 * - for RESTORE request, it will sample the file data version and send it to
905 * coordinator which is useful if the file was imported as 'released'.
907 * \return 0 on success.
909 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
911 struct ll_sb_info *sbi = ll_s2sbi(sb);
912 struct hsm_progress_kernel hpk;
917 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
918 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
919 * initialized if copy_end was called with copy == NULL.
922 /* Forge a hsm_progress based on data from copy. */
923 hpk.hpk_fid = copy->hc_hai.hai_fid;
924 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
925 hpk.hpk_extent = copy->hc_hai.hai_extent;
926 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
927 hpk.hpk_errval = copy->hc_errval;
928 hpk.hpk_data_version = 0;
930 /* For archive request, we need to check the file data was not changed.
932 * For restore request, we need to send the file data version, this is
933 * useful when the file was created using hsm_import.
935 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
936 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
937 (copy->hc_errval == 0)) {
939 __u64 data_version = 0;
941 /* Get lsm for this fid */
942 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
944 hpk.hpk_flags |= HP_FLAG_RETRY;
945 /* hpk_errval must be >= 0 */
946 hpk.hpk_errval = -PTR_ERR(inode);
947 GOTO(progress, rc = PTR_ERR(inode));
950 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
953 CDEBUG(D_HSM, "Could not read file data version. "
954 "Request could not be confirmed.\n");
955 if (hpk.hpk_errval == 0)
956 hpk.hpk_errval = -rc;
960 /* Store in the hsm_copy for later copytool use.
961 * Always modified even if no lsm. */
962 hpk.hpk_data_version = data_version;
964 /* File could have been stripped during archiving, so we need
965 * to check anyway. */
966 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
967 (copy->hc_data_version != data_version)) {
968 CDEBUG(D_HSM, "File data version mismatched. "
969 "File content was changed during archiving. "
970 DFID", start:%#llx current:%#llx\n",
971 PFID(©->hc_hai.hai_fid),
972 copy->hc_data_version, data_version);
973 /* File was changed, send error to cdt. Do not ask for
974 * retry because if a file is modified frequently,
975 * the cdt will loop on retried archive requests.
976 * The policy engine will ask for a new archive later
977 * when the file will not be modified for some tunable
979 hpk.hpk_flags &= ~HP_FLAG_RETRY;
981 /* hpk_errval must be >= 0 */
982 hpk.hpk_errval = -rc;
989 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
992 /* Return first error */
993 RETURN(rc != 0 ? rc : rc2);
997 static int copy_and_ct_start(int cmd, struct obd_export *exp,
998 const struct lustre_kernelcomm __user *data)
1000 struct lustre_kernelcomm *lk;
1001 struct lustre_kernelcomm *tmp;
1002 size_t size = sizeof(*lk);
1007 /* copy data from userspace to get numbers of archive_id */
1008 OBD_ALLOC(lk, size);
1012 if (copy_from_user(lk, data, size))
1013 GOTO(out_lk, rc = -EFAULT);
1015 if (lk->lk_flags & LK_FLG_STOP)
1018 if (!(lk->lk_flags & LK_FLG_DATANR)) {
1019 __u32 archive_mask = lk->lk_data_count;
1022 /* old hsm agent to old MDS */
1023 if (!exp_connect_archive_id_array(exp))
1026 /* old hsm agent to new MDS */
1027 lk->lk_flags |= LK_FLG_DATANR;
1029 if (archive_mask == 0)
1032 count = hweight32(archive_mask);
1033 new_size = offsetof(struct lustre_kernelcomm, lk_data[count]);
1034 OBD_ALLOC(tmp, new_size);
1036 GOTO(out_lk, rc = -ENOMEM);
1038 memcpy(tmp, lk, size);
1039 tmp->lk_data_count = count;
1045 for (i = 0; i < sizeof(archive_mask) * 8; i++) {
1046 if ((1 << i) & archive_mask) {
1047 lk->lk_data[count] = i + 1;
1054 /* new hsm agent to new mds */
1055 if (lk->lk_data_count > 0) {
1056 new_size = offsetof(struct lustre_kernelcomm,
1057 lk_data[lk->lk_data_count]);
1058 OBD_ALLOC(tmp, new_size);
1060 GOTO(out_lk, rc = -ENOMEM);
1066 if (copy_from_user(lk, data, size))
1067 GOTO(out_lk, rc = -EFAULT);
1070 /* new hsm agent to old MDS */
1071 if (!exp_connect_archive_id_array(exp)) {
1074 if (lk->lk_data_count > LL_HSM_ORIGIN_MAX_ARCHIVE)
1075 GOTO(out_lk, rc = -EINVAL);
1077 for (i = 0; i < lk->lk_data_count; i++) {
1078 if (lk->lk_data[i] > LL_HSM_ORIGIN_MAX_ARCHIVE) {
1080 CERROR("%s: archive id %d requested but only "
1081 "[0 - %zu] supported: rc = %d\n",
1082 exp->exp_obd->obd_name, lk->lk_data[i],
1083 LL_HSM_ORIGIN_MAX_ARCHIVE, rc);
1087 if (lk->lk_data[i] == 0) {
1092 archives |= (1 << (lk->lk_data[i] - 1));
1094 lk->lk_flags &= ~LK_FLG_DATANR;
1095 lk->lk_data_count = archives;
1098 rc = obd_iocontrol(cmd, exp, size, lk, NULL);
1104 static int check_owner(int type, int id)
1108 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
1112 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1121 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1123 int cmd = qctl->qc_cmd;
1124 int type = qctl->qc_type;
1125 int id = qctl->qc_id;
1126 int valid = qctl->qc_valid;
1133 case LUSTRE_Q_SETDEFAULT:
1134 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1138 case LUSTRE_Q_GETDEFAULT:
1139 if (check_owner(type, id) &&
1140 (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1146 CERROR("unsupported quotactl op: %#x\n", cmd);
1150 if (valid != QC_GENERAL) {
1151 if (cmd == Q_GETINFO)
1152 qctl->qc_cmd = Q_GETOINFO;
1153 else if (cmd == Q_GETQUOTA)
1154 qctl->qc_cmd = Q_GETOQUOTA;
1160 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1161 sizeof(*qctl), qctl, NULL);
1164 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1165 sizeof(*qctl), qctl, NULL);
1168 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1169 sizeof(*qctl), qctl, NULL);
1171 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1173 sizeof(*qctl), qctl, NULL);
1185 struct obd_quotactl *oqctl;
1187 OBD_ALLOC_PTR(oqctl);
1191 QCTL_COPY(oqctl, qctl);
1192 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1194 OBD_FREE_PTR(oqctl);
1197 /* If QIF_SPACE is not set, client should collect the
1198 * space usage from OSSs by itself */
1199 if (cmd == Q_GETQUOTA &&
1200 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1201 !oqctl->qc_dqblk.dqb_curspace) {
1202 struct obd_quotactl *oqctl_tmp;
1204 OBD_ALLOC_PTR(oqctl_tmp);
1205 if (oqctl_tmp == NULL)
1206 GOTO(out, rc = -ENOMEM);
1208 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1209 oqctl_tmp->qc_id = oqctl->qc_id;
1210 oqctl_tmp->qc_type = oqctl->qc_type;
1212 /* collect space usage from OSTs */
1213 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1214 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1215 if (!rc || rc == -EREMOTEIO) {
1216 oqctl->qc_dqblk.dqb_curspace =
1217 oqctl_tmp->qc_dqblk.dqb_curspace;
1218 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1221 /* collect space & inode usage from MDTs */
1222 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1223 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1224 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1225 if (!rc || rc == -EREMOTEIO) {
1226 oqctl->qc_dqblk.dqb_curspace +=
1227 oqctl_tmp->qc_dqblk.dqb_curspace;
1228 oqctl->qc_dqblk.dqb_curinodes =
1229 oqctl_tmp->qc_dqblk.dqb_curinodes;
1230 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1232 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1235 OBD_FREE_PTR(oqctl_tmp);
1238 QCTL_COPY(qctl, oqctl);
1239 OBD_FREE_PTR(oqctl);
1245 int ll_rmfid(struct file *file, void __user *arg)
1247 const struct fid_array __user *ufa = arg;
1248 struct fid_array *lfa = NULL;
1251 int i, rc, *rcs = NULL;
1254 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
1255 !(ll_i2sbi(file_inode(file))->ll_flags & LL_SBI_USER_FID2PATH))
1257 /* Only need to get the buflen */
1258 if (get_user(nr, &ufa->fa_nr))
1260 /* DoS protection */
1261 if (nr > OBD_MAX_FIDS_IN_ARRAY)
1264 size = offsetof(struct fid_array, fa_fids[nr]);
1265 OBD_ALLOC(lfa, size);
1268 OBD_ALLOC(rcs, sizeof(int) * nr);
1270 GOTO(free_lfa, rc = -ENOMEM);
1272 if (copy_from_user(lfa, arg, size))
1273 GOTO(free_rcs, rc = -EFAULT);
1275 /* Call mdc_iocontrol */
1276 rc = md_rmfid(ll_i2mdexp(file_inode(file)), lfa, rcs, NULL);
1278 for (i = 0; i < nr; i++)
1280 lfa->fa_fids[i].f_ver = rcs[i];
1281 if (copy_to_user(arg, lfa, size))
1286 OBD_FREE(rcs, sizeof(int) * nr);
1288 OBD_FREE(lfa, size);
1293 /* This function tries to get a single name component,
1294 * to send to the server. No actual path traversal involved,
1295 * so we limit to NAME_MAX */
1296 static char *ll_getname(const char __user *filename)
1301 OBD_ALLOC(tmp, NAME_MAX + 1);
1304 return ERR_PTR(-ENOMEM);
1306 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1309 else if (len > NAME_MAX)
1310 ret = -ENAMETOOLONG;
1313 OBD_FREE(tmp, NAME_MAX + 1);
1319 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1321 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1323 struct dentry *dentry = file_dentry(file);
1324 struct inode *inode = file_inode(file);
1325 struct ll_sb_info *sbi = ll_i2sbi(inode);
1326 struct obd_ioctl_data *data;
1330 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1331 PFID(ll_inode2fid(inode)), inode, cmd);
1333 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1334 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1337 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1339 case FS_IOC_GETFLAGS:
1340 case FS_IOC_SETFLAGS:
1341 RETURN(ll_iocontrol(inode, file, cmd, arg));
1342 case FSFILT_IOC_GETVERSION:
1343 case FS_IOC_GETVERSION:
1344 RETURN(put_user(inode->i_generation, (int __user *)arg));
1345 /* We need to special case any other ioctls we want to handle,
1346 * to send them to the MDS/OST as appropriate and to properly
1347 * network encode the arg field. */
1348 case FS_IOC_SETVERSION:
1351 case LL_IOC_GET_MDTIDX: {
1354 mdtidx = ll_get_mdt_idx(inode);
1358 if (put_user((int)mdtidx, (int __user *)arg))
1363 case IOC_MDC_LOOKUP: {
1364 int namelen, len = 0;
1368 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1373 filename = data->ioc_inlbuf1;
1374 namelen = strlen(filename);
1376 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1377 GOTO(out_free, rc = -EINVAL);
1380 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1382 CERROR("%s: lookup %.*s failed: rc = %d\n",
1383 sbi->ll_fsname, namelen, filename, rc);
1387 OBD_FREE_LARGE(buf, len);
1390 case LL_IOC_LMV_SETSTRIPE: {
1391 struct lmv_user_md *lum;
1400 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1405 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1406 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1407 GOTO(lmv_out_free, rc = -EINVAL);
1409 filename = data->ioc_inlbuf1;
1410 namelen = data->ioc_inllen1;
1413 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1414 GOTO(lmv_out_free, rc = -EINVAL);
1416 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1417 lumlen = data->ioc_inllen2;
1419 if (!lmv_user_magic_supported(lum->lum_magic)) {
1420 CERROR("%s: wrong lum magic %x : rc = %d\n", filename,
1421 lum->lum_magic, -EINVAL);
1422 GOTO(lmv_out_free, rc = -EINVAL);
1425 if ((lum->lum_magic == LMV_USER_MAGIC ||
1426 lum->lum_magic == LMV_USER_MAGIC_SPECIFIC) &&
1427 lumlen < sizeof(*lum)) {
1428 CERROR("%s: wrong lum size %d for magic %x : rc = %d\n",
1429 filename, lumlen, lum->lum_magic, -EINVAL);
1430 GOTO(lmv_out_free, rc = -EINVAL);
1433 if (lum->lum_magic == LMV_MAGIC_FOREIGN &&
1434 lumlen < sizeof(struct lmv_foreign_md)) {
1435 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1436 filename, lum->lum_magic, lumlen, -EFAULT);
1437 GOTO(lmv_out_free, rc = -EINVAL);
1440 mode = data->ioc_type;
1441 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1443 OBD_FREE_LARGE(buf, len);
1447 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1448 struct lmv_user_md lum;
1449 struct lmv_user_md __user *ulump =
1450 (struct lmv_user_md __user *)arg;
1453 if (copy_from_user(&lum, ulump, sizeof(lum)))
1456 if (lum.lum_magic != LMV_USER_MAGIC)
1459 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1463 case LL_IOC_LOV_SETSTRIPE_NEW:
1464 case LL_IOC_LOV_SETSTRIPE: {
1465 struct lov_user_md_v3 *lumv3 = NULL;
1466 struct lov_user_md_v1 lumv1;
1467 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1468 struct lov_user_md_v1 __user *lumv1p =
1469 (struct lov_user_md_v1 __user *)arg;
1470 struct lov_user_md_v3 __user *lumv3p =
1471 (struct lov_user_md_v3 __user *)arg;
1474 int set_default = 0;
1476 BUILD_BUG_ON(sizeof(struct lov_user_md_v3) <=
1477 sizeof(struct lov_comp_md_v1));
1478 BUILD_BUG_ON(sizeof(*lumv3) != sizeof(*lumv3p));
1479 /* first try with v1 which is smaller than v3 */
1480 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1483 if (inode->i_sb->s_root == file_dentry(file))
1486 switch (lumv1.lmm_magic) {
1487 case LOV_USER_MAGIC_V3:
1488 case LOV_USER_MAGIC_SPECIFIC:
1489 lum_size = ll_lov_user_md_size(&lumv1);
1492 OBD_ALLOC(lumv3, lum_size);
1495 if (copy_from_user(lumv3, lumv3p, lum_size))
1496 GOTO(out, rc = -EFAULT);
1497 lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1499 case LOV_USER_MAGIC_V1:
1502 GOTO(out, rc = -ENOTSUPP);
1505 /* in v1 and v3 cases lumv1 points to data */
1506 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1509 OBD_FREE(lumv3, lum_size);
1512 case LL_IOC_LMV_GETSTRIPE: {
1513 struct lmv_user_md __user *ulmv =
1514 (struct lmv_user_md __user *)arg;
1515 struct lmv_user_md lum;
1516 struct ptlrpc_request *request = NULL;
1517 struct ptlrpc_request *root_request = NULL;
1518 union lmv_mds_md *lmm = NULL;
1521 struct lmv_user_md *tmp = NULL;
1525 int max_stripe_count;
1529 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1532 max_stripe_count = lum.lum_stripe_count;
1533 /* lum_magic will indicate which stripe the ioctl will like
1534 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1535 * is for default LMV stripe */
1536 if (lum.lum_magic == LMV_MAGIC_V1)
1537 valid |= OBD_MD_MEA;
1538 else if (lum.lum_magic == LMV_USER_MAGIC)
1539 valid |= OBD_MD_DEFAULT_MEA;
1543 rc = ll_dir_getstripe_default(inode, (void **)&lmm, &lmmsize,
1544 &request, &root_request, valid);
1546 GOTO(finish_req, rc);
1548 /* Get default LMV EA */
1549 if (lum.lum_magic == LMV_USER_MAGIC) {
1550 if (lmmsize > sizeof(*ulmv))
1551 GOTO(finish_req, rc = -EINVAL);
1553 if (copy_to_user(ulmv, lmm, lmmsize))
1554 GOTO(finish_req, rc = -EFAULT);
1556 GOTO(finish_req, rc);
1559 /* if foreign LMV case, fake stripes number */
1560 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1561 struct lmv_foreign_md *lfm;
1563 lfm = (struct lmv_foreign_md *)lmm;
1564 if (lfm->lfm_length < XATTR_SIZE_MAX -
1565 offsetof(typeof(*lfm), lfm_value)) {
1566 __u32 size = lfm->lfm_length +
1567 offsetof(typeof(*lfm), lfm_value);
1569 stripe_count = lmv_foreign_to_md_stripes(size);
1571 CERROR("invalid %d foreign size returned\n",
1576 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1578 if (max_stripe_count < stripe_count) {
1579 lum.lum_stripe_count = stripe_count;
1580 if (copy_to_user(ulmv, &lum, sizeof(lum)))
1581 GOTO(finish_req, rc = -EFAULT);
1582 GOTO(finish_req, rc = -E2BIG);
1585 /* enough room on user side and foreign case */
1586 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1587 struct lmv_foreign_md *lfm;
1590 lfm = (struct lmv_foreign_md *)lmm;
1591 size = lfm->lfm_length +
1592 offsetof(struct lmv_foreign_md, lfm_value);
1593 if (copy_to_user(ulmv, lfm, size))
1594 GOTO(finish_req, rc = -EFAULT);
1595 GOTO(finish_req, rc);
1598 lum_size = lmv_user_md_size(stripe_count,
1599 LMV_USER_MAGIC_SPECIFIC);
1600 OBD_ALLOC(tmp, lum_size);
1602 GOTO(finish_req, rc = -ENOMEM);
1604 mdt_index = ll_get_mdt_idx(inode);
1606 GOTO(out_tmp, rc = -ENOMEM);
1608 tmp->lum_magic = LMV_MAGIC_V1;
1609 tmp->lum_stripe_count = 0;
1610 tmp->lum_stripe_offset = mdt_index;
1611 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1612 for (i = 0; i < stripe_count; i++) {
1615 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1616 if (fid_is_sane(&fid)) {
1617 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1619 GOTO(out_tmp, rc = mdt_index);
1621 tmp->lum_objects[i].lum_mds = mdt_index;
1622 tmp->lum_objects[i].lum_fid = fid;
1625 tmp->lum_stripe_count++;
1628 if (copy_to_user(ulmv, tmp, lum_size))
1629 GOTO(out_tmp, rc = -EFAULT);
1631 OBD_FREE(tmp, lum_size);
1633 ptlrpc_req_finished(request);
1634 ptlrpc_req_finished(root_request);
1638 case LL_IOC_REMOVE_ENTRY: {
1639 char *filename = NULL;
1643 /* Here is a little hack to avoid sending REINT_RMENTRY to
1644 * unsupported server, which might crash the server(LU-2730),
1645 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1646 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1647 * server will support REINT_RMENTRY XXX*/
1648 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1651 filename = ll_getname((const char __user *)arg);
1652 if (IS_ERR(filename))
1653 RETURN(PTR_ERR(filename));
1655 namelen = strlen(filename);
1657 GOTO(out_rmdir, rc = -EINVAL);
1659 rc = ll_rmdir_entry(inode, filename, namelen);
1662 ll_putname(filename);
1666 RETURN(ll_rmfid(file, (void __user *)arg));
1667 case LL_IOC_LOV_SWAP_LAYOUTS:
1669 case IOC_OBD_STATFS:
1670 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1671 case LL_IOC_LOV_GETSTRIPE:
1672 case LL_IOC_LOV_GETSTRIPE_NEW:
1673 case LL_IOC_MDC_GETINFO:
1674 case LL_IOC_MDC_GETINFO_OLD:
1675 case IOC_MDC_GETFILEINFO:
1676 case IOC_MDC_GETFILEINFO_OLD:
1677 case IOC_MDC_GETFILESTRIPE: {
1678 struct ptlrpc_request *request = NULL;
1679 struct ptlrpc_request *root_request = NULL;
1680 struct lov_user_md __user *lump;
1681 struct lov_mds_md *lmm = NULL;
1682 struct mdt_body *body;
1683 char *filename = NULL;
1684 lstat_t __user *statp = NULL;
1685 lstatx_t __user *stxp = NULL;
1686 __u64 __user *flagsp = NULL;
1687 __u32 __user *lmmsizep = NULL;
1688 struct lu_fid __user *fidp = NULL;
1691 if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1692 cmd == IOC_MDC_GETFILEINFO ||
1693 cmd == IOC_MDC_GETFILESTRIPE) {
1694 filename = ll_getname((const char __user *)arg);
1695 if (IS_ERR(filename))
1696 RETURN(PTR_ERR(filename));
1698 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1699 &lmmsize, &request);
1701 rc = ll_dir_getstripe_default(inode, (void **)&lmm,
1707 body = req_capsule_server_get(&request->rq_pill,
1709 LASSERT(body != NULL);
1714 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1715 cmd == LL_IOC_MDC_GETINFO ||
1716 cmd == IOC_MDC_GETFILEINFO_OLD ||
1717 cmd == LL_IOC_MDC_GETINFO_OLD)) {
1725 if (cmd == IOC_MDC_GETFILESTRIPE ||
1726 cmd == LL_IOC_LOV_GETSTRIPE ||
1727 cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1728 lump = (struct lov_user_md __user *)arg;
1729 } else if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1730 cmd == LL_IOC_MDC_GETINFO_OLD){
1731 struct lov_user_mds_data_v1 __user *lmdp;
1733 lmdp = (struct lov_user_mds_data_v1 __user *)arg;
1734 statp = &lmdp->lmd_st;
1735 lump = &lmdp->lmd_lmm;
1737 struct lov_user_mds_data __user *lmdp;
1739 lmdp = (struct lov_user_mds_data __user *)arg;
1740 fidp = &lmdp->lmd_fid;
1741 stxp = &lmdp->lmd_stx;
1742 flagsp = &lmdp->lmd_flags;
1743 lmmsizep = &lmdp->lmd_lmmsize;
1744 lump = &lmdp->lmd_lmm;
1748 /* If the file has no striping then zero out *lump so
1749 * that the caller isn't confused by garbage. */
1750 if (clear_user(lump, sizeof(*lump)))
1751 GOTO(out_req, rc = -EFAULT);
1752 } else if (copy_to_user(lump, lmm, lmmsize)) {
1753 if (copy_to_user(lump, lmm, sizeof(*lump)))
1754 GOTO(out_req, rc = -EFAULT);
1758 if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1759 cmd == LL_IOC_MDC_GETINFO_OLD) {
1762 st.st_dev = inode->i_sb->s_dev;
1763 st.st_mode = body->mbo_mode;
1764 st.st_nlink = body->mbo_nlink;
1765 st.st_uid = body->mbo_uid;
1766 st.st_gid = body->mbo_gid;
1767 st.st_rdev = body->mbo_rdev;
1768 st.st_size = body->mbo_size;
1769 st.st_blksize = PAGE_SIZE;
1770 st.st_blocks = body->mbo_blocks;
1771 st.st_atime = body->mbo_atime;
1772 st.st_mtime = body->mbo_mtime;
1773 st.st_ctime = body->mbo_ctime;
1774 st.st_ino = cl_fid_build_ino(&body->mbo_fid1,
1778 if (copy_to_user(statp, &st, sizeof(st)))
1779 GOTO(out_req, rc = -EFAULT);
1780 } else if (cmd == IOC_MDC_GETFILEINFO ||
1781 cmd == LL_IOC_MDC_GETINFO) {
1782 lstatx_t stx = { 0 };
1783 __u64 valid = body->mbo_valid;
1785 stx.stx_blksize = PAGE_SIZE;
1786 stx.stx_nlink = body->mbo_nlink;
1787 stx.stx_uid = body->mbo_uid;
1788 stx.stx_gid = body->mbo_gid;
1789 stx.stx_mode = body->mbo_mode;
1790 stx.stx_ino = cl_fid_build_ino(&body->mbo_fid1,
1793 stx.stx_size = body->mbo_size;
1794 stx.stx_blocks = body->mbo_blocks;
1795 stx.stx_atime.tv_sec = body->mbo_atime;
1796 stx.stx_ctime.tv_sec = body->mbo_ctime;
1797 stx.stx_mtime.tv_sec = body->mbo_mtime;
1798 stx.stx_rdev_major = MAJOR(body->mbo_rdev);
1799 stx.stx_rdev_minor = MINOR(body->mbo_rdev);
1800 stx.stx_dev_major = MAJOR(inode->i_sb->s_dev);
1801 stx.stx_dev_minor = MINOR(inode->i_sb->s_dev);
1802 stx.stx_mask |= STATX_BASIC_STATS;
1805 * For a striped directory, the size and blocks returned
1806 * from MDT is not correct.
1807 * The size and blocks are aggregated by client across
1809 * Thus for a striped directory, do not return the valid
1810 * FLSIZE and FLBLOCKS flags to the caller.
1811 * However, this whould be better decided by the MDS
1812 * instead of the client.
1814 if (cmd == LL_IOC_MDC_GETINFO &&
1815 ll_i2info(inode)->lli_lsm_md != NULL)
1816 valid &= ~(OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
1818 if (flagsp && copy_to_user(flagsp, &valid,
1820 GOTO(out_req, rc = -EFAULT);
1822 if (fidp && copy_to_user(fidp, &body->mbo_fid1,
1824 GOTO(out_req, rc = -EFAULT);
1826 if (!(valid & OBD_MD_FLSIZE))
1827 stx.stx_mask &= ~STATX_SIZE;
1828 if (!(valid & OBD_MD_FLBLOCKS))
1829 stx.stx_mask &= ~STATX_BLOCKS;
1831 if (stxp && copy_to_user(stxp, &stx, sizeof(stx)))
1832 GOTO(out_req, rc = -EFAULT);
1834 if (lmmsizep && copy_to_user(lmmsizep, &lmmsize,
1836 GOTO(out_req, rc = -EFAULT);
1841 ptlrpc_req_finished(request);
1842 ptlrpc_req_finished(root_request);
1844 ll_putname(filename);
1847 case OBD_IOC_QUOTACTL: {
1848 struct if_quotactl *qctl;
1850 OBD_ALLOC_PTR(qctl);
1854 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1855 GOTO(out_quotactl, rc = -EFAULT);
1857 rc = quotactl_ioctl(sbi, qctl);
1860 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1867 case OBD_IOC_GETDTNAME:
1868 case OBD_IOC_GETMDNAME:
1869 RETURN(ll_get_obd_name(inode, cmd, arg));
1870 case LL_IOC_FLUSHCTX:
1871 RETURN(ll_flush_ctx(inode));
1872 case LL_IOC_GETOBDCOUNT: {
1874 struct obd_export *exp;
1876 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1879 /* get ost count when count is zero, get mdt count otherwise */
1880 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1881 vallen = sizeof(count);
1882 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1883 KEY_TGT_COUNT, &vallen, &count);
1885 CERROR("get target count failed: %d\n", rc);
1889 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1894 case LL_IOC_PATH2FID:
1895 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1896 sizeof(struct lu_fid)))
1899 case LL_IOC_GET_CONNECT_FLAGS: {
1900 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1901 (void __user *)arg));
1903 case OBD_IOC_FID2PATH:
1904 RETURN(ll_fid2path(inode, (void __user *)arg));
1905 case LL_IOC_GETPARENT:
1906 RETURN(ll_getparent(file, (void __user *)arg));
1907 case LL_IOC_FID2MDTIDX: {
1908 struct obd_export *exp = ll_i2mdexp(inode);
1912 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1916 /* Call mdc_iocontrol */
1917 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1918 (__u32 __user *)&index);
1924 case LL_IOC_HSM_REQUEST: {
1925 struct hsm_user_request *hur;
1932 /* We don't know the true size yet; copy the fixed-size part */
1933 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1938 /* Compute the whole struct size */
1939 totalsize = hur_len(hur);
1944 /* Final size will be more than double totalsize */
1945 if (totalsize >= MDS_MAXREQSIZE / 3)
1948 OBD_ALLOC_LARGE(hur, totalsize);
1952 /* Copy the whole struct */
1953 if (copy_from_user(hur, (void __user *)arg, totalsize))
1954 GOTO(out_hur, rc = -EFAULT);
1956 if (hur->hur_request.hr_action == HUA_RELEASE) {
1957 const struct lu_fid *fid;
1961 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1962 fid = &hur->hur_user_item[i].hui_fid;
1963 f = search_inode_for_lustre(inode->i_sb, fid);
1969 rc = ll_hsm_release(f);
1975 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1980 OBD_FREE_LARGE(hur, totalsize);
1984 case LL_IOC_HSM_PROGRESS: {
1985 struct hsm_progress_kernel hpk;
1986 struct hsm_progress hp;
1988 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1991 hpk.hpk_fid = hp.hp_fid;
1992 hpk.hpk_cookie = hp.hp_cookie;
1993 hpk.hpk_extent = hp.hp_extent;
1994 hpk.hpk_flags = hp.hp_flags;
1995 hpk.hpk_errval = hp.hp_errval;
1996 hpk.hpk_data_version = 0;
1998 /* File may not exist in Lustre; all progress
1999 * reported to Lustre root */
2000 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
2004 case LL_IOC_HSM_CT_START:
2005 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2008 rc = copy_and_ct_start(cmd, sbi->ll_md_exp,
2009 (struct lustre_kernelcomm __user *)arg);
2012 case LL_IOC_HSM_COPY_START: {
2013 struct hsm_copy *copy;
2016 OBD_ALLOC_PTR(copy);
2019 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2024 rc = ll_ioc_copy_start(inode->i_sb, copy);
2025 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2031 case LL_IOC_HSM_COPY_END: {
2032 struct hsm_copy *copy;
2035 OBD_ALLOC_PTR(copy);
2038 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2043 rc = ll_ioc_copy_end(inode->i_sb, copy);
2044 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2050 case LL_IOC_MIGRATE: {
2051 struct lmv_user_md *lum;
2058 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
2062 data = (struct obd_ioctl_data *)buf;
2063 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
2064 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
2065 GOTO(migrate_free, rc = -EINVAL);
2067 filename = data->ioc_inlbuf1;
2068 namelen = data->ioc_inllen1;
2070 if (namelen < 1 || namelen != strlen(filename) + 1) {
2071 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
2072 GOTO(migrate_free, rc = -EINVAL);
2075 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
2076 if (lum->lum_magic != LMV_USER_MAGIC &&
2077 lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
2079 CERROR("%s: wrong lum magic %x: rc = %d\n",
2080 filename, lum->lum_magic, rc);
2081 GOTO(migrate_free, rc);
2084 rc = ll_migrate(inode, file, lum, filename);
2086 OBD_FREE_LARGE(buf, len);
2090 case LL_IOC_FSGETXATTR:
2091 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
2092 case LL_IOC_FSSETXATTR:
2093 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
2094 case LL_IOC_PCC_DETACH_BY_FID: {
2095 struct lu_pcc_detach_fid *detach;
2097 struct inode *inode2;
2100 OBD_ALLOC_PTR(detach);
2104 if (copy_from_user(detach,
2105 (const struct lu_pcc_detach_fid __user *)arg,
2107 GOTO(out_detach, rc = -EFAULT);
2109 fid = &detach->pccd_fid;
2110 ino = cl_fid_build_ino(fid, ll_need_32bit_api(sbi));
2111 inode2 = ilookup5(inode->i_sb, ino, ll_test_inode_by_fid, fid);
2113 /* Target inode is not in inode cache, and PCC file
2114 * has aleady released, return immdiately.
2116 GOTO(out_detach, rc = 0);
2118 if (!S_ISREG(inode2->i_mode))
2119 GOTO(out_iput, rc = -EINVAL);
2121 if (!inode_owner_or_capable(inode2))
2122 GOTO(out_iput, rc = -EPERM);
2124 rc = pcc_ioctl_detach(inode2, detach->pccd_opt);
2128 OBD_FREE_PTR(detach);
2132 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
2133 (void __user *)arg));
2137 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
2139 struct inode *inode = file->f_mapping->host;
2140 struct ll_file_data *fd = file->private_data;
2141 struct ll_sb_info *sbi = ll_i2sbi(inode);
2142 int api32 = ll_need_32bit_api(sbi);
2143 loff_t ret = -EINVAL;
2151 offset += file->f_pos;
2157 offset += LL_DIR_END_OFF_32BIT;
2159 offset += LL_DIR_END_OFF;
2166 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
2167 (!api32 && offset <= LL_DIR_END_OFF))) {
2168 if (offset != file->f_pos) {
2169 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
2170 (!api32 && offset == LL_DIR_END_OFF))
2171 fd->lfd_pos = MDS_DIR_END_OFF;
2172 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
2173 fd->lfd_pos = offset << 32;
2175 fd->lfd_pos = offset;
2176 file->f_pos = offset;
2177 file->f_version = 0;
2184 inode_unlock(inode);
2188 static int ll_dir_open(struct inode *inode, struct file *file)
2191 RETURN(ll_file_open(inode, file));
2194 static int ll_dir_release(struct inode *inode, struct file *file)
2197 RETURN(ll_file_release(inode, file));
2200 const struct file_operations ll_dir_operations = {
2201 .llseek = ll_dir_seek,
2202 .open = ll_dir_open,
2203 .release = ll_dir_release,
2204 .read = generic_read_dir,
2205 #ifdef HAVE_DIR_CONTEXT
2206 .iterate_shared = ll_iterate,
2208 .readdir = ll_readdir,
2210 .unlocked_ioctl = ll_dir_ioctl,