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 (BIT(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 case LUSTRE_Q_SETQUOTAPOOL:
1135 case LUSTRE_Q_SETINFOPOOL:
1136 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1140 case LUSTRE_Q_GETDEFAULT:
1141 case LUSTRE_Q_GETQUOTAPOOL:
1142 if (check_owner(type, id) &&
1143 (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1147 case LUSTRE_Q_GETINFOPOOL:
1150 CERROR("unsupported quotactl op: %#x\n", cmd);
1154 if (valid != QC_GENERAL) {
1155 if (cmd == Q_GETINFO)
1156 qctl->qc_cmd = Q_GETOINFO;
1157 else if (cmd == Q_GETQUOTA ||
1158 cmd == LUSTRE_Q_GETQUOTAPOOL)
1159 qctl->qc_cmd = Q_GETOQUOTA;
1165 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1166 sizeof(*qctl), qctl, NULL);
1169 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1170 sizeof(*qctl), qctl, NULL);
1173 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1174 sizeof(*qctl), qctl, NULL);
1176 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1178 sizeof(*qctl), qctl, NULL);
1190 struct obd_quotactl *oqctl;
1191 int oqctl_len = sizeof(*oqctl);
1193 if (LUSTRE_Q_CMD_IS_POOL(cmd))
1194 oqctl_len += LOV_MAXPOOLNAME + 1;
1196 OBD_ALLOC(oqctl, oqctl_len);
1200 QCTL_COPY(oqctl, qctl);
1201 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1203 OBD_FREE(oqctl, oqctl_len);
1206 /* If QIF_SPACE is not set, client should collect the
1207 * space usage from OSSs by itself */
1208 if ((cmd == Q_GETQUOTA || cmd == LUSTRE_Q_GETQUOTAPOOL) &&
1209 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1210 !oqctl->qc_dqblk.dqb_curspace) {
1211 struct obd_quotactl *oqctl_tmp;
1213 OBD_ALLOC_PTR(oqctl_tmp);
1214 if (oqctl_tmp == NULL)
1215 GOTO(out, rc = -ENOMEM);
1217 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1218 oqctl_tmp->qc_id = oqctl->qc_id;
1219 oqctl_tmp->qc_type = oqctl->qc_type;
1221 /* collect space usage from OSTs */
1222 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1223 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1224 if (!rc || rc == -EREMOTEIO) {
1225 oqctl->qc_dqblk.dqb_curspace =
1226 oqctl_tmp->qc_dqblk.dqb_curspace;
1227 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1230 /* collect space & inode usage from MDTs */
1231 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1232 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1233 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1234 if (!rc || rc == -EREMOTEIO) {
1235 oqctl->qc_dqblk.dqb_curspace +=
1236 oqctl_tmp->qc_dqblk.dqb_curspace;
1237 oqctl->qc_dqblk.dqb_curinodes =
1238 oqctl_tmp->qc_dqblk.dqb_curinodes;
1239 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1241 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1244 OBD_FREE_PTR(oqctl_tmp);
1247 QCTL_COPY(qctl, oqctl);
1248 OBD_FREE(oqctl, oqctl_len);
1254 int ll_rmfid(struct file *file, void __user *arg)
1256 const struct fid_array __user *ufa = arg;
1257 struct fid_array *lfa = NULL;
1260 int i, rc, *rcs = NULL;
1263 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
1264 !(ll_i2sbi(file_inode(file))->ll_flags & LL_SBI_USER_FID2PATH))
1266 /* Only need to get the buflen */
1267 if (get_user(nr, &ufa->fa_nr))
1269 /* DoS protection */
1270 if (nr > OBD_MAX_FIDS_IN_ARRAY)
1273 size = offsetof(struct fid_array, fa_fids[nr]);
1274 OBD_ALLOC(lfa, size);
1277 OBD_ALLOC_PTR_ARRAY(rcs, nr);
1279 GOTO(free_lfa, rc = -ENOMEM);
1281 if (copy_from_user(lfa, arg, size))
1282 GOTO(free_rcs, rc = -EFAULT);
1284 /* Call mdc_iocontrol */
1285 rc = md_rmfid(ll_i2mdexp(file_inode(file)), lfa, rcs, NULL);
1287 for (i = 0; i < nr; i++)
1289 lfa->fa_fids[i].f_ver = rcs[i];
1290 if (copy_to_user(arg, lfa, size))
1295 OBD_FREE_PTR_ARRAY(rcs, nr);
1297 OBD_FREE(lfa, size);
1302 /* This function tries to get a single name component,
1303 * to send to the server. No actual path traversal involved,
1304 * so we limit to NAME_MAX */
1305 static char *ll_getname(const char __user *filename)
1310 OBD_ALLOC(tmp, NAME_MAX + 1);
1313 return ERR_PTR(-ENOMEM);
1315 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1318 else if (len > NAME_MAX)
1319 ret = -ENAMETOOLONG;
1322 OBD_FREE(tmp, NAME_MAX + 1);
1328 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1330 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1332 struct dentry *dentry = file_dentry(file);
1333 struct inode *inode = file_inode(file);
1334 struct ll_sb_info *sbi = ll_i2sbi(inode);
1335 struct obd_ioctl_data *data;
1339 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1340 PFID(ll_inode2fid(inode)), inode, cmd);
1342 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1343 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1346 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1348 case FS_IOC_GETFLAGS:
1349 case FS_IOC_SETFLAGS:
1350 RETURN(ll_iocontrol(inode, file, cmd, arg));
1351 case FSFILT_IOC_GETVERSION:
1352 case FS_IOC_GETVERSION:
1353 RETURN(put_user(inode->i_generation, (int __user *)arg));
1354 /* We need to special case any other ioctls we want to handle,
1355 * to send them to the MDS/OST as appropriate and to properly
1356 * network encode the arg field. */
1357 case FS_IOC_SETVERSION:
1360 case LL_IOC_GET_MDTIDX: {
1363 mdtidx = ll_get_mdt_idx(inode);
1367 if (put_user((int)mdtidx, (int __user *)arg))
1372 case IOC_MDC_LOOKUP: {
1373 int namelen, len = 0;
1377 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1382 filename = data->ioc_inlbuf1;
1383 namelen = strlen(filename);
1385 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1386 GOTO(out_free, rc = -EINVAL);
1389 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1391 CERROR("%s: lookup %.*s failed: rc = %d\n",
1392 sbi->ll_fsname, namelen, filename, rc);
1396 OBD_FREE_LARGE(buf, len);
1399 case LL_IOC_LMV_SETSTRIPE: {
1400 struct lmv_user_md *lum;
1409 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1414 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1415 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1416 GOTO(lmv_out_free, rc = -EINVAL);
1418 filename = data->ioc_inlbuf1;
1419 namelen = data->ioc_inllen1;
1422 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1423 GOTO(lmv_out_free, rc = -EINVAL);
1425 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1426 lumlen = data->ioc_inllen2;
1428 if (!lmv_user_magic_supported(lum->lum_magic)) {
1429 CERROR("%s: wrong lum magic %x : rc = %d\n", filename,
1430 lum->lum_magic, -EINVAL);
1431 GOTO(lmv_out_free, rc = -EINVAL);
1434 if ((lum->lum_magic == LMV_USER_MAGIC ||
1435 lum->lum_magic == LMV_USER_MAGIC_SPECIFIC) &&
1436 lumlen < sizeof(*lum)) {
1437 CERROR("%s: wrong lum size %d for magic %x : rc = %d\n",
1438 filename, lumlen, lum->lum_magic, -EINVAL);
1439 GOTO(lmv_out_free, rc = -EINVAL);
1442 if (lum->lum_magic == LMV_MAGIC_FOREIGN &&
1443 lumlen < sizeof(struct lmv_foreign_md)) {
1444 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1445 filename, lum->lum_magic, lumlen, -EFAULT);
1446 GOTO(lmv_out_free, rc = -EINVAL);
1449 mode = data->ioc_type;
1450 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1452 OBD_FREE_LARGE(buf, len);
1456 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1457 struct lmv_user_md lum;
1458 struct lmv_user_md __user *ulump =
1459 (struct lmv_user_md __user *)arg;
1462 if (copy_from_user(&lum, ulump, sizeof(lum)))
1465 if (lum.lum_magic != LMV_USER_MAGIC)
1468 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1472 case LL_IOC_LOV_SETSTRIPE_NEW:
1473 case LL_IOC_LOV_SETSTRIPE: {
1474 struct lov_user_md_v3 *lumv3 = NULL;
1475 struct lov_user_md_v1 lumv1;
1476 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1477 struct lov_user_md_v1 __user *lumv1p =
1478 (struct lov_user_md_v1 __user *)arg;
1479 struct lov_user_md_v3 __user *lumv3p =
1480 (struct lov_user_md_v3 __user *)arg;
1483 int set_default = 0;
1485 BUILD_BUG_ON(sizeof(struct lov_user_md_v3) <=
1486 sizeof(struct lov_comp_md_v1));
1487 BUILD_BUG_ON(sizeof(*lumv3) != sizeof(*lumv3p));
1488 /* first try with v1 which is smaller than v3 */
1489 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1492 if (inode->i_sb->s_root == file_dentry(file))
1495 switch (lumv1.lmm_magic) {
1496 case LOV_USER_MAGIC_V3:
1497 case LOV_USER_MAGIC_SPECIFIC:
1498 lum_size = ll_lov_user_md_size(&lumv1);
1501 OBD_ALLOC(lumv3, lum_size);
1504 if (copy_from_user(lumv3, lumv3p, lum_size))
1505 GOTO(out, rc = -EFAULT);
1506 lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1508 case LOV_USER_MAGIC_V1:
1511 GOTO(out, rc = -ENOTSUPP);
1514 /* in v1 and v3 cases lumv1 points to data */
1515 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1518 OBD_FREE(lumv3, lum_size);
1521 case LL_IOC_LMV_GETSTRIPE: {
1522 struct lmv_user_md __user *ulmv =
1523 (struct lmv_user_md __user *)arg;
1524 struct lmv_user_md lum;
1525 struct ptlrpc_request *request = NULL;
1526 struct ptlrpc_request *root_request = NULL;
1527 union lmv_mds_md *lmm = NULL;
1530 struct lmv_user_md *tmp = NULL;
1534 int max_stripe_count;
1538 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1541 max_stripe_count = lum.lum_stripe_count;
1542 /* lum_magic will indicate which stripe the ioctl will like
1543 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1544 * is for default LMV stripe */
1545 if (lum.lum_magic == LMV_MAGIC_V1)
1546 valid |= OBD_MD_MEA;
1547 else if (lum.lum_magic == LMV_USER_MAGIC)
1548 valid |= OBD_MD_DEFAULT_MEA;
1552 rc = ll_dir_getstripe_default(inode, (void **)&lmm, &lmmsize,
1553 &request, &root_request, valid);
1555 GOTO(finish_req, rc);
1557 /* Get default LMV EA */
1558 if (lum.lum_magic == LMV_USER_MAGIC) {
1559 if (lmmsize > sizeof(*ulmv))
1560 GOTO(finish_req, rc = -EINVAL);
1562 if (copy_to_user(ulmv, lmm, lmmsize))
1563 GOTO(finish_req, rc = -EFAULT);
1565 GOTO(finish_req, rc);
1568 /* if foreign LMV case, fake stripes number */
1569 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1570 struct lmv_foreign_md *lfm;
1572 lfm = (struct lmv_foreign_md *)lmm;
1573 if (lfm->lfm_length < XATTR_SIZE_MAX -
1574 offsetof(typeof(*lfm), lfm_value)) {
1575 __u32 size = lfm->lfm_length +
1576 offsetof(typeof(*lfm), lfm_value);
1578 stripe_count = lmv_foreign_to_md_stripes(size);
1580 CERROR("invalid %d foreign size returned\n",
1585 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1587 if (max_stripe_count < stripe_count) {
1588 lum.lum_stripe_count = stripe_count;
1589 if (copy_to_user(ulmv, &lum, sizeof(lum)))
1590 GOTO(finish_req, rc = -EFAULT);
1591 GOTO(finish_req, rc = -E2BIG);
1594 /* enough room on user side and foreign case */
1595 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1596 struct lmv_foreign_md *lfm;
1599 lfm = (struct lmv_foreign_md *)lmm;
1600 size = lfm->lfm_length +
1601 offsetof(struct lmv_foreign_md, lfm_value);
1602 if (copy_to_user(ulmv, lfm, size))
1603 GOTO(finish_req, rc = -EFAULT);
1604 GOTO(finish_req, rc);
1607 lum_size = lmv_user_md_size(stripe_count,
1608 LMV_USER_MAGIC_SPECIFIC);
1609 OBD_ALLOC(tmp, lum_size);
1611 GOTO(finish_req, rc = -ENOMEM);
1613 mdt_index = ll_get_mdt_idx(inode);
1615 GOTO(out_tmp, rc = -ENOMEM);
1617 tmp->lum_magic = LMV_MAGIC_V1;
1618 tmp->lum_stripe_count = 0;
1619 tmp->lum_stripe_offset = mdt_index;
1620 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1621 for (i = 0; i < stripe_count; i++) {
1624 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1625 if (fid_is_sane(&fid)) {
1626 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1628 GOTO(out_tmp, rc = mdt_index);
1630 tmp->lum_objects[i].lum_mds = mdt_index;
1631 tmp->lum_objects[i].lum_fid = fid;
1634 tmp->lum_stripe_count++;
1637 if (copy_to_user(ulmv, tmp, lum_size))
1638 GOTO(out_tmp, rc = -EFAULT);
1640 OBD_FREE(tmp, lum_size);
1642 ptlrpc_req_finished(request);
1643 ptlrpc_req_finished(root_request);
1647 case LL_IOC_REMOVE_ENTRY: {
1648 char *filename = NULL;
1652 /* Here is a little hack to avoid sending REINT_RMENTRY to
1653 * unsupported server, which might crash the server(LU-2730),
1654 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1655 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1656 * server will support REINT_RMENTRY XXX*/
1657 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1660 filename = ll_getname((const char __user *)arg);
1661 if (IS_ERR(filename))
1662 RETURN(PTR_ERR(filename));
1664 namelen = strlen(filename);
1666 GOTO(out_rmdir, rc = -EINVAL);
1668 rc = ll_rmdir_entry(inode, filename, namelen);
1671 ll_putname(filename);
1675 RETURN(ll_rmfid(file, (void __user *)arg));
1676 case LL_IOC_LOV_SWAP_LAYOUTS:
1678 case IOC_OBD_STATFS:
1679 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1680 case LL_IOC_LOV_GETSTRIPE:
1681 case LL_IOC_LOV_GETSTRIPE_NEW:
1682 case LL_IOC_MDC_GETINFO:
1683 case LL_IOC_MDC_GETINFO_OLD:
1684 case IOC_MDC_GETFILEINFO:
1685 case IOC_MDC_GETFILEINFO_OLD:
1686 case IOC_MDC_GETFILESTRIPE: {
1687 struct ptlrpc_request *request = NULL;
1688 struct ptlrpc_request *root_request = NULL;
1689 struct lov_user_md __user *lump;
1690 struct lov_mds_md *lmm = NULL;
1691 struct mdt_body *body;
1692 char *filename = NULL;
1693 lstat_t __user *statp = NULL;
1694 lstatx_t __user *stxp = NULL;
1695 __u64 __user *flagsp = NULL;
1696 __u32 __user *lmmsizep = NULL;
1697 struct lu_fid __user *fidp = NULL;
1700 if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1701 cmd == IOC_MDC_GETFILEINFO ||
1702 cmd == IOC_MDC_GETFILESTRIPE) {
1703 filename = ll_getname((const char __user *)arg);
1704 if (IS_ERR(filename))
1705 RETURN(PTR_ERR(filename));
1707 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1708 &lmmsize, &request);
1710 rc = ll_dir_getstripe_default(inode, (void **)&lmm,
1716 body = req_capsule_server_get(&request->rq_pill,
1718 LASSERT(body != NULL);
1723 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1724 cmd == LL_IOC_MDC_GETINFO ||
1725 cmd == IOC_MDC_GETFILEINFO_OLD ||
1726 cmd == LL_IOC_MDC_GETINFO_OLD)) {
1734 if (cmd == IOC_MDC_GETFILESTRIPE ||
1735 cmd == LL_IOC_LOV_GETSTRIPE ||
1736 cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1737 lump = (struct lov_user_md __user *)arg;
1738 } else if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1739 cmd == LL_IOC_MDC_GETINFO_OLD){
1740 struct lov_user_mds_data_v1 __user *lmdp;
1742 lmdp = (struct lov_user_mds_data_v1 __user *)arg;
1743 statp = &lmdp->lmd_st;
1744 lump = &lmdp->lmd_lmm;
1746 struct lov_user_mds_data __user *lmdp;
1748 lmdp = (struct lov_user_mds_data __user *)arg;
1749 fidp = &lmdp->lmd_fid;
1750 stxp = &lmdp->lmd_stx;
1751 flagsp = &lmdp->lmd_flags;
1752 lmmsizep = &lmdp->lmd_lmmsize;
1753 lump = &lmdp->lmd_lmm;
1757 /* If the file has no striping then zero out *lump so
1758 * that the caller isn't confused by garbage. */
1759 if (clear_user(lump, sizeof(*lump)))
1760 GOTO(out_req, rc = -EFAULT);
1761 } else if (copy_to_user(lump, lmm, lmmsize)) {
1762 if (copy_to_user(lump, lmm, sizeof(*lump)))
1763 GOTO(out_req, rc = -EFAULT);
1767 if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1768 cmd == LL_IOC_MDC_GETINFO_OLD) {
1771 st.st_dev = inode->i_sb->s_dev;
1772 st.st_mode = body->mbo_mode;
1773 st.st_nlink = body->mbo_nlink;
1774 st.st_uid = body->mbo_uid;
1775 st.st_gid = body->mbo_gid;
1776 st.st_rdev = body->mbo_rdev;
1777 st.st_size = body->mbo_size;
1778 st.st_blksize = PAGE_SIZE;
1779 st.st_blocks = body->mbo_blocks;
1780 st.st_atime = body->mbo_atime;
1781 st.st_mtime = body->mbo_mtime;
1782 st.st_ctime = body->mbo_ctime;
1783 st.st_ino = cl_fid_build_ino(&body->mbo_fid1,
1787 if (copy_to_user(statp, &st, sizeof(st)))
1788 GOTO(out_req, rc = -EFAULT);
1789 } else if (cmd == IOC_MDC_GETFILEINFO ||
1790 cmd == LL_IOC_MDC_GETINFO) {
1791 lstatx_t stx = { 0 };
1792 __u64 valid = body->mbo_valid;
1794 stx.stx_blksize = PAGE_SIZE;
1795 stx.stx_nlink = body->mbo_nlink;
1796 stx.stx_uid = body->mbo_uid;
1797 stx.stx_gid = body->mbo_gid;
1798 stx.stx_mode = body->mbo_mode;
1799 stx.stx_ino = cl_fid_build_ino(&body->mbo_fid1,
1802 stx.stx_size = body->mbo_size;
1803 stx.stx_blocks = body->mbo_blocks;
1804 stx.stx_atime.tv_sec = body->mbo_atime;
1805 stx.stx_ctime.tv_sec = body->mbo_ctime;
1806 stx.stx_mtime.tv_sec = body->mbo_mtime;
1807 stx.stx_btime.tv_sec = body->mbo_btime;
1808 stx.stx_rdev_major = MAJOR(body->mbo_rdev);
1809 stx.stx_rdev_minor = MINOR(body->mbo_rdev);
1810 stx.stx_dev_major = MAJOR(inode->i_sb->s_dev);
1811 stx.stx_dev_minor = MINOR(inode->i_sb->s_dev);
1812 stx.stx_mask |= STATX_BASIC_STATS | STATX_BTIME;
1815 * For a striped directory, the size and blocks returned
1816 * from MDT is not correct.
1817 * The size and blocks are aggregated by client across
1819 * Thus for a striped directory, do not return the valid
1820 * FLSIZE and FLBLOCKS flags to the caller.
1821 * However, this whould be better decided by the MDS
1822 * instead of the client.
1824 if (cmd == LL_IOC_MDC_GETINFO &&
1825 ll_i2info(inode)->lli_lsm_md != NULL)
1826 valid &= ~(OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
1828 if (flagsp && copy_to_user(flagsp, &valid,
1830 GOTO(out_req, rc = -EFAULT);
1832 if (fidp && copy_to_user(fidp, &body->mbo_fid1,
1834 GOTO(out_req, rc = -EFAULT);
1836 if (!(valid & OBD_MD_FLSIZE))
1837 stx.stx_mask &= ~STATX_SIZE;
1838 if (!(valid & OBD_MD_FLBLOCKS))
1839 stx.stx_mask &= ~STATX_BLOCKS;
1841 if (stxp && copy_to_user(stxp, &stx, sizeof(stx)))
1842 GOTO(out_req, rc = -EFAULT);
1844 if (lmmsizep && copy_to_user(lmmsizep, &lmmsize,
1846 GOTO(out_req, rc = -EFAULT);
1851 ptlrpc_req_finished(request);
1852 ptlrpc_req_finished(root_request);
1854 ll_putname(filename);
1857 case OBD_IOC_QUOTACTL: {
1858 struct if_quotactl *qctl;
1859 int qctl_len = sizeof(*qctl) + LOV_MAXPOOLNAME + 1;
1861 OBD_ALLOC(qctl, qctl_len);
1865 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1866 GOTO(out_quotactl, rc = -EFAULT);
1868 if (LUSTRE_Q_CMD_IS_POOL(qctl->qc_cmd)) {
1869 char __user *from = (char __user *)arg +
1870 offsetof(typeof(*qctl), qc_poolname);
1871 if (copy_from_user(qctl->qc_poolname, from,
1872 LOV_MAXPOOLNAME + 1))
1873 GOTO(out_quotactl, rc = -EFAULT);
1876 rc = quotactl_ioctl(sbi, qctl);
1878 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1882 OBD_FREE(qctl, qctl_len);
1885 case OBD_IOC_GETDTNAME:
1886 case OBD_IOC_GETMDNAME:
1887 RETURN(ll_get_obd_name(inode, cmd, arg));
1888 case LL_IOC_FLUSHCTX:
1889 RETURN(ll_flush_ctx(inode));
1890 case LL_IOC_GETOBDCOUNT: {
1892 struct obd_export *exp;
1894 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1897 /* get ost count when count is zero, get mdt count otherwise */
1898 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1899 vallen = sizeof(count);
1900 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1901 KEY_TGT_COUNT, &vallen, &count);
1903 CERROR("get target count failed: %d\n", rc);
1907 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1912 case LL_IOC_PATH2FID:
1913 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1914 sizeof(struct lu_fid)))
1917 case LL_IOC_GET_CONNECT_FLAGS: {
1918 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1919 (void __user *)arg));
1921 case OBD_IOC_FID2PATH:
1922 RETURN(ll_fid2path(inode, (void __user *)arg));
1923 case LL_IOC_GETPARENT:
1924 RETURN(ll_getparent(file, (void __user *)arg));
1925 case LL_IOC_FID2MDTIDX: {
1926 struct obd_export *exp = ll_i2mdexp(inode);
1930 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1934 /* Call mdc_iocontrol */
1935 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1936 (__u32 __user *)&index);
1942 case LL_IOC_HSM_REQUEST: {
1943 struct hsm_user_request *hur;
1950 /* We don't know the true size yet; copy the fixed-size part */
1951 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1956 /* Compute the whole struct size */
1957 totalsize = hur_len(hur);
1962 /* Final size will be more than double totalsize */
1963 if (totalsize >= MDS_MAXREQSIZE / 3)
1966 OBD_ALLOC_LARGE(hur, totalsize);
1970 /* Copy the whole struct */
1971 if (copy_from_user(hur, (void __user *)arg, totalsize))
1972 GOTO(out_hur, rc = -EFAULT);
1974 if (hur->hur_request.hr_action == HUA_RELEASE) {
1975 const struct lu_fid *fid;
1979 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1980 fid = &hur->hur_user_item[i].hui_fid;
1981 f = search_inode_for_lustre(inode->i_sb, fid);
1987 rc = ll_hsm_release(f);
1993 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1998 OBD_FREE_LARGE(hur, totalsize);
2002 case LL_IOC_HSM_PROGRESS: {
2003 struct hsm_progress_kernel hpk;
2004 struct hsm_progress hp;
2006 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
2009 hpk.hpk_fid = hp.hp_fid;
2010 hpk.hpk_cookie = hp.hp_cookie;
2011 hpk.hpk_extent = hp.hp_extent;
2012 hpk.hpk_flags = hp.hp_flags;
2013 hpk.hpk_errval = hp.hp_errval;
2014 hpk.hpk_data_version = 0;
2016 /* File may not exist in Lustre; all progress
2017 * reported to Lustre root */
2018 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
2022 case LL_IOC_HSM_CT_START:
2023 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2026 rc = copy_and_ct_start(cmd, sbi->ll_md_exp,
2027 (struct lustre_kernelcomm __user *)arg);
2030 case LL_IOC_HSM_COPY_START: {
2031 struct hsm_copy *copy;
2034 OBD_ALLOC_PTR(copy);
2037 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2042 rc = ll_ioc_copy_start(inode->i_sb, copy);
2043 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2049 case LL_IOC_HSM_COPY_END: {
2050 struct hsm_copy *copy;
2053 OBD_ALLOC_PTR(copy);
2056 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2061 rc = ll_ioc_copy_end(inode->i_sb, copy);
2062 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2068 case LL_IOC_MIGRATE: {
2069 struct lmv_user_md *lum;
2076 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
2080 data = (struct obd_ioctl_data *)buf;
2081 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
2082 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
2083 GOTO(migrate_free, rc = -EINVAL);
2085 filename = data->ioc_inlbuf1;
2086 namelen = data->ioc_inllen1;
2088 if (namelen < 1 || namelen != strlen(filename) + 1) {
2089 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
2090 GOTO(migrate_free, rc = -EINVAL);
2093 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
2094 if (lum->lum_magic != LMV_USER_MAGIC &&
2095 lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
2097 CERROR("%s: wrong lum magic %x: rc = %d\n",
2098 filename, lum->lum_magic, rc);
2099 GOTO(migrate_free, rc);
2102 rc = ll_migrate(inode, file, lum, filename);
2104 OBD_FREE_LARGE(buf, len);
2108 case LL_IOC_FSGETXATTR:
2109 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
2110 case LL_IOC_FSSETXATTR:
2111 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
2112 case LL_IOC_PCC_DETACH_BY_FID: {
2113 struct lu_pcc_detach_fid *detach;
2115 struct inode *inode2;
2118 OBD_ALLOC_PTR(detach);
2122 if (copy_from_user(detach,
2123 (const struct lu_pcc_detach_fid __user *)arg,
2125 GOTO(out_detach, rc = -EFAULT);
2127 fid = &detach->pccd_fid;
2128 ino = cl_fid_build_ino(fid, ll_need_32bit_api(sbi));
2129 inode2 = ilookup5(inode->i_sb, ino, ll_test_inode_by_fid, fid);
2131 /* Target inode is not in inode cache, and PCC file
2132 * has aleady released, return immdiately.
2134 GOTO(out_detach, rc = 0);
2136 if (!S_ISREG(inode2->i_mode))
2137 GOTO(out_iput, rc = -EINVAL);
2139 if (!inode_owner_or_capable(inode2))
2140 GOTO(out_iput, rc = -EPERM);
2142 rc = pcc_ioctl_detach(inode2, detach->pccd_opt);
2146 OBD_FREE_PTR(detach);
2150 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
2151 (void __user *)arg));
2155 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
2157 struct inode *inode = file->f_mapping->host;
2158 struct ll_file_data *fd = file->private_data;
2159 struct ll_sb_info *sbi = ll_i2sbi(inode);
2160 int api32 = ll_need_32bit_api(sbi);
2161 loff_t ret = -EINVAL;
2169 offset += file->f_pos;
2175 offset += LL_DIR_END_OFF_32BIT;
2177 offset += LL_DIR_END_OFF;
2184 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
2185 (!api32 && offset <= LL_DIR_END_OFF))) {
2186 if (offset != file->f_pos) {
2187 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
2188 (!api32 && offset == LL_DIR_END_OFF))
2189 fd->lfd_pos = MDS_DIR_END_OFF;
2190 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
2191 fd->lfd_pos = offset << 32;
2193 fd->lfd_pos = offset;
2194 file->f_pos = offset;
2195 file->f_version = 0;
2202 inode_unlock(inode);
2206 static int ll_dir_open(struct inode *inode, struct file *file)
2209 RETURN(ll_file_open(inode, file));
2212 static int ll_dir_release(struct inode *inode, struct file *file)
2215 RETURN(ll_file_release(inode, file));
2218 const struct file_operations ll_dir_operations = {
2219 .llseek = ll_dir_seek,
2220 .open = ll_dir_open,
2221 .release = ll_dir_release,
2222 .read = generic_read_dir,
2223 #ifdef HAVE_DIR_CONTEXT
2224 .iterate_shared = ll_iterate,
2226 .readdir = ll_readdir,
2228 .unlocked_ioctl = ll_dir_ioctl,