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);
178 #ifdef HAVE_DIR_CONTEXT
179 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
180 struct dir_context *ctx)
183 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
184 void *cookie, filldir_t filldir)
187 struct ll_sb_info *sbi = ll_i2sbi(inode);
189 bool is_api32 = ll_need_32bit_api(sbi);
190 bool is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
192 struct ll_dir_chain chain;
197 ll_dir_chain_init(&chain);
199 page = ll_get_dir_page(inode, op_data, pos, &chain);
201 while (rc == 0 && !done) {
202 struct lu_dirpage *dp;
203 struct lu_dirent *ent;
212 hash = MDS_DIR_END_OFF;
213 dp = page_address(page);
214 for (ent = lu_dirent_start(dp); ent != NULL && !done;
215 ent = lu_dirent_next(ent)) {
222 hash = le64_to_cpu(ent->lde_hash);
223 if (hash < pos) /* Skip until we find target hash */
226 namelen = le16_to_cpu(ent->lde_namelen);
227 if (namelen == 0) /* Skip dummy record */
230 if (is_api32 && is_hash64)
234 fid_le_to_cpu(&fid, &ent->lde_fid);
235 ino = cl_fid_build_ino(&fid, is_api32);
236 type = IFTODT(lu_dirent_type_get(ent));
237 /* For ll_nfs_get_name_filldir(), it will try to access
238 * 'ent' through 'lde_name', so the parameter 'name'
239 * for 'filldir()' must be part of the 'ent'. */
240 #ifdef HAVE_DIR_CONTEXT
242 done = !dir_emit(ctx, ent->lde_name, namelen, ino,
245 done = filldir(cookie, ent->lde_name, namelen, lhash,
252 ll_release_page(inode, page, false);
256 next = le64_to_cpu(dp->ldp_hash_end);
258 if (pos == MDS_DIR_END_OFF) {
260 * End of directory reached.
263 ll_release_page(inode, page, false);
266 * Normal case: continue to the next
269 ll_release_page(inode, page,
270 le32_to_cpu(dp->ldp_flags) &
273 page = ll_get_dir_page(inode, op_data, pos,
277 #ifdef HAVE_DIR_CONTEXT
282 ll_dir_chain_fini(&chain);
286 #ifdef HAVE_DIR_CONTEXT
287 static int ll_iterate(struct file *filp, struct dir_context *ctx)
289 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
292 struct inode *inode = file_inode(filp);
293 struct ll_file_data *lfd = filp->private_data;
294 struct ll_sb_info *sbi = ll_i2sbi(inode);
295 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
296 int api32 = ll_need_32bit_api(sbi);
297 struct md_op_data *op_data;
298 struct lu_fid pfid = { 0 };
299 ktime_t kstart = ktime_get();
310 "VFS Op:inode="DFID"(%p) pos/size%lu/%llu 32bit_api %d\n",
311 PFID(ll_inode2fid(inode)),
312 inode, (unsigned long)pos, i_size_read(inode), api32);
314 if (pos == MDS_DIR_END_OFF)
320 if (unlikely(ll_dir_striped(inode))) {
322 * This is only needed for striped dir to fill ..,
323 * see lmv_read_page()
325 if (file_dentry(filp)->d_parent != NULL &&
326 file_dentry(filp)->d_parent->d_inode != NULL) {
327 __u64 ibits = MDS_INODELOCK_LOOKUP;
328 struct inode *parent =
329 file_dentry(filp)->d_parent->d_inode;
331 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
332 pfid = *ll_inode2fid(parent);
335 /* If it can not find in cache, do lookup .. on the master
337 if (fid_is_zero(&pfid)) {
338 rc = ll_dir_get_parent_fid(inode, &pfid);
344 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
345 LUSTRE_OPC_ANY, inode);
347 GOTO(out, rc = PTR_ERR(op_data));
349 /* foreign dirs are browsed out of Lustre */
350 if (unlikely(op_data->op_mea1 != NULL &&
351 op_data->op_mea1->lsm_md_magic == LMV_MAGIC_FOREIGN)) {
352 ll_finish_md_op_data(op_data);
356 op_data->op_fid3 = pfid;
358 #ifdef HAVE_DIR_CONTEXT
360 rc = ll_dir_read(inode, &pos, op_data, ctx);
363 rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
368 if (pos == MDS_DIR_END_OFF) {
370 pos = LL_DIR_END_OFF_32BIT;
372 pos = LL_DIR_END_OFF;
377 #ifdef HAVE_DIR_CONTEXT
382 ll_finish_md_op_data(op_data);
386 ll_stats_ops_tally(sbi, LPROC_LL_READDIR,
387 ktime_us_delta(ktime_get(), kstart));
393 * Create striped directory with specified stripe(@lump)
395 * \param[in] dparent the parent of the directory.
396 * \param[in] lump the specified stripes.
397 * \param[in] dirname the name of the directory.
398 * \param[in] mode the specified mode of the directory.
400 * \retval =0 if striped directory is being created successfully.
401 * <0 if the creation is failed.
403 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
404 size_t len, const char *dirname, umode_t mode)
406 struct inode *parent = dparent->d_inode;
407 struct ptlrpc_request *request = NULL;
408 struct md_op_data *op_data;
409 struct ll_sb_info *sbi = ll_i2sbi(parent);
410 struct inode *inode = NULL;
411 struct dentry dentry = {
415 .len = strlen(dirname),
416 .hash = ll_full_name_hash(dparent, dirname,
423 if (unlikely(!lmv_user_magic_supported(lump->lum_magic)))
426 if (lump->lum_magic != LMV_MAGIC_FOREIGN) {
428 "VFS Op:inode="DFID"(%p) name %s stripe_offset %d, stripe_count: %u\n",
429 PFID(ll_inode2fid(parent)), parent, dirname,
430 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
432 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lump;
435 "VFS Op:inode="DFID"(%p) name %s foreign, length %u, value '%.*s'\n",
436 PFID(ll_inode2fid(parent)), parent, dirname,
437 lfm->lfm_length, lfm->lfm_length, lfm->lfm_value);
440 if (lump->lum_stripe_count > 1 &&
441 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
444 if (IS_DEADDIR(parent) &&
445 !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
448 if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_CRUSH)) {
449 if ((lump->lum_hash_type & LMV_HASH_TYPE_MASK) ==
450 LMV_HASH_TYPE_CRUSH) {
451 /* if server doesn't support 'crush' hash type,
452 * switch to fnv_1a_64.
454 lump->lum_hash_type &= ~LMV_HASH_TYPE_MASK;
455 lump->lum_hash_type |= LMV_HASH_TYPE_FNV_1A_64;
456 } else if ((lump->lum_hash_type & LMV_HASH_TYPE_MASK) ==
457 LMV_HASH_TYPE_UNKNOWN) {
458 /* from 2.14 MDT will choose default hash type if client
459 * doesn't set a valid one, while old server doesn't
462 lump->lum_hash_type &= ~LMV_HASH_TYPE_MASK;
463 lump->lum_hash_type |= LMV_HASH_TYPE_DEFAULT;
467 if (unlikely(!lmv_user_magic_supported(cpu_to_le32(lump->lum_magic))))
468 lustre_swab_lmv_user_md(lump);
470 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
471 mode &= ~current_umask();
472 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
473 op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
474 strlen(dirname), mode, LUSTRE_OPC_MKDIR,
477 RETURN(PTR_ERR(op_data));
479 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
480 /* selinux_dentry_init_security() uses dentry->d_parent and name
481 * to determine the security context for the file. So our fake
482 * dentry should be real enough for this purpose. */
483 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
484 &op_data->op_file_secctx_name,
485 &op_data->op_file_secctx,
486 &op_data->op_file_secctx_size);
488 GOTO(out_op_data, err);
491 op_data->op_cli_flags |= CLI_SET_MEA;
492 err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
493 from_kuid(&init_user_ns, current_fsuid()),
494 from_kgid(&init_user_ns, current_fsgid()),
495 cfs_curproc_cap_pack(), 0, &request);
497 GOTO(out_request, err);
499 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
501 err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
503 GOTO(out_inode, err);
505 dentry.d_inode = inode;
507 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
509 err = security_inode_notifysecctx(inode,
510 op_data->op_file_secctx,
511 op_data->op_file_secctx_size);
514 err = ll_inode_init_security(&dentry, inode, parent);
517 GOTO(out_inode, err);
523 ptlrpc_req_finished(request);
525 ll_finish_md_op_data(op_data);
530 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
533 struct ll_sb_info *sbi = ll_i2sbi(inode);
534 struct md_op_data *op_data;
535 struct ptlrpc_request *req = NULL;
541 switch (lump->lmm_magic) {
542 case LOV_USER_MAGIC_V1:
543 lum_size = sizeof(struct lov_user_md_v1);
545 case LOV_USER_MAGIC_V3:
546 lum_size = sizeof(struct lov_user_md_v3);
548 case LOV_USER_MAGIC_COMP_V1:
549 lum_size = ((struct lov_comp_md_v1 *)lump)->lcm_size;
552 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
553 lustre_swab_lmv_user_md(
554 (struct lmv_user_md *)lump);
555 lum_size = sizeof(struct lmv_user_md);
557 case LOV_USER_MAGIC_SPECIFIC: {
558 struct lov_user_md_v3 *v3 =
559 (struct lov_user_md_v3 *)lump;
560 if (v3->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
562 lum_size = lov_user_md_size(v3->lmm_stripe_count,
563 LOV_USER_MAGIC_SPECIFIC);
567 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
568 " %#08x != %#08x nor %#08x\n",
569 lump->lmm_magic, LOV_USER_MAGIC_V1,
575 * This is coming from userspace, so should be in
576 * local endian. But the MDS would like it in little
577 * endian, so we swab it before we send it.
579 if ((__swab32(lump->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
580 le32_to_cpu(LOV_MAGIC_MAGIC))
581 lustre_swab_lov_user_md(lump, 0);
583 lum_size = sizeof(struct lov_user_md_v1);
586 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
587 LUSTRE_OPC_ANY, NULL);
589 RETURN(PTR_ERR(op_data));
591 /* swabbing is done in lov_setstripe() on server side */
592 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
593 ll_finish_md_op_data(op_data);
594 ptlrpc_req_finished(req);
601 static int ll_dir_get_default_layout(struct inode *inode, void **plmm,
603 struct ptlrpc_request **request, u64 valid,
604 enum get_default_layout_type type)
606 struct ll_sb_info *sbi = ll_i2sbi(inode);
607 struct mdt_body *body;
608 struct lov_mds_md *lmm = NULL;
609 struct ptlrpc_request *req = NULL;
611 struct md_op_data *op_data;
615 rc = ll_get_default_mdsize(sbi, &lmm_size);
619 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
620 0, lmm_size, LUSTRE_OPC_ANY,
623 RETURN(PTR_ERR(op_data));
625 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
627 if (type == GET_DEFAULT_LAYOUT_ROOT) {
628 lu_root_fid(&op_data->op_fid1);
629 fid = op_data->op_fid1;
631 fid = *ll_inode2fid(inode);
634 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
635 ll_finish_md_op_data(op_data);
637 CDEBUG(D_INFO, "md_getattr failed on inode "DFID": rc %d\n",
642 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
643 LASSERT(body != NULL);
645 lmm_size = body->mbo_eadatasize;
647 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
649 GOTO(out, rc = -ENODATA);
652 lmm = req_capsule_server_sized_get(&req->rq_pill,
653 &RMF_MDT_MD, lmm_size);
654 LASSERT(lmm != NULL);
657 * This is coming from the MDS, so is probably in
658 * little endian. We convert it to host endian before
659 * passing it to userspace.
661 /* We don't swab objects for directories */
662 switch (le32_to_cpu(lmm->lmm_magic)) {
665 case LOV_MAGIC_COMP_V1:
666 case LOV_USER_MAGIC_SPECIFIC:
667 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
668 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
671 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
672 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
675 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
676 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
678 case LMV_MAGIC_FOREIGN: {
679 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lmm;
681 if (LMV_MAGIC_FOREIGN != cpu_to_le32(LMV_MAGIC_FOREIGN)) {
682 __swab32s(&lfm->lfm_magic);
683 __swab32s(&lfm->lfm_length);
684 __swab32s(&lfm->lfm_type);
685 __swab32s(&lfm->lfm_flags);
690 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
695 *plmm_size = lmm_size;
701 * This function will be used to get default LOV/LMV/Default LMV
702 * @valid will be used to indicate which stripe it will retrieve.
703 * If the directory does not have its own default layout, then the
704 * function will request the default layout from root FID.
705 * OBD_MD_MEA LMV stripe EA
706 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
707 * otherwise Default LOV EA.
708 * Each time, it can only retrieve 1 stripe EA
710 int ll_dir_getstripe_default(struct inode *inode, void **plmm, int *plmm_size,
711 struct ptlrpc_request **request,
712 struct ptlrpc_request **root_request,
715 struct ptlrpc_request *req = NULL;
716 struct ptlrpc_request *root_req = NULL;
717 struct lov_mds_md *lmm = NULL;
722 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
724 if (rc == -ENODATA && !fid_is_root(ll_inode2fid(inode)) &&
725 !(valid & (OBD_MD_MEA|OBD_MD_DEFAULT_MEA)) && root_request != NULL)
726 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
728 GET_DEFAULT_LAYOUT_ROOT);
731 *plmm_size = lmm_size;
733 if (root_request != NULL)
734 *root_request = root_req;
740 * This function will be used to get default LOV/LMV/Default LMV
741 * @valid will be used to indicate which stripe it will retrieve
742 * OBD_MD_MEA LMV stripe EA
743 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
744 * otherwise Default LOV EA.
745 * Each time, it can only retrieve 1 stripe EA
747 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
748 struct ptlrpc_request **request, u64 valid)
750 struct ptlrpc_request *req = NULL;
751 struct lov_mds_md *lmm = NULL;
756 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
760 *plmm_size = lmm_size;
766 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
768 struct md_op_data *op_data;
773 OBD_ALLOC_PTR(op_data);
777 op_data->op_flags |= MF_GET_MDT_IDX;
778 op_data->op_fid1 = *fid;
779 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
780 mdt_index = op_data->op_mds;
781 OBD_FREE_PTR(op_data);
789 * Get MDT index for the inode.
791 int ll_get_mdt_idx(struct inode *inode)
793 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
797 * Generic handler to do any pre-copy work.
799 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
800 * first information for it that real work has started.
802 * Moreover, for a ARCHIVE request, it will sample the file data version and
803 * store it in \a copy.
805 * \return 0 on success.
807 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
809 struct ll_sb_info *sbi = ll_s2sbi(sb);
810 struct hsm_progress_kernel hpk;
815 /* Forge a hsm_progress based on data from copy. */
816 hpk.hpk_fid = copy->hc_hai.hai_fid;
817 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
818 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
819 hpk.hpk_extent.length = 0;
822 hpk.hpk_data_version = 0;
825 /* For archive request, we need to read the current file version. */
826 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
828 __u64 data_version = 0;
830 /* Get inode for this fid */
831 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
833 hpk.hpk_flags |= HP_FLAG_RETRY;
834 /* hpk_errval is >= 0 */
835 hpk.hpk_errval = -PTR_ERR(inode);
836 GOTO(progress, rc = PTR_ERR(inode));
839 /* Read current file data version */
840 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
843 CDEBUG(D_HSM, "Could not read file data version of "
844 DFID" (rc = %d). Archive request ("
845 "%#llx) could not be done.\n",
846 PFID(©->hc_hai.hai_fid), rc,
847 copy->hc_hai.hai_cookie);
848 hpk.hpk_flags |= HP_FLAG_RETRY;
849 /* hpk_errval must be >= 0 */
850 hpk.hpk_errval = -rc;
854 /* Store in the hsm_copy for later copytool use.
855 * Always modified even if no lsm. */
856 copy->hc_data_version = data_version;
860 /* On error, the request should be considered as completed */
861 if (hpk.hpk_errval > 0)
862 hpk.hpk_flags |= HP_FLAG_COMPLETED;
864 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
867 /* Return first error */
868 RETURN(rc != 0 ? rc : rc2);
872 * Generic handler to do any post-copy work.
874 * It will send the last hsm_progress update to coordinator to inform it
875 * that copy is finished and whether it was successful or not.
878 * - for ARCHIVE request, it will sample the file data version and compare it
879 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
880 * will be considered as failed.
881 * - for RESTORE request, it will sample the file data version and send it to
882 * coordinator which is useful if the file was imported as 'released'.
884 * \return 0 on success.
886 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
888 struct ll_sb_info *sbi = ll_s2sbi(sb);
889 struct hsm_progress_kernel hpk;
894 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
895 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
896 * initialized if copy_end was called with copy == NULL.
899 /* Forge a hsm_progress based on data from copy. */
900 hpk.hpk_fid = copy->hc_hai.hai_fid;
901 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
902 hpk.hpk_extent = copy->hc_hai.hai_extent;
903 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
904 hpk.hpk_errval = copy->hc_errval;
905 hpk.hpk_data_version = 0;
907 /* For archive request, we need to check the file data was not changed.
909 * For restore request, we need to send the file data version, this is
910 * useful when the file was created using hsm_import.
912 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
913 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
914 (copy->hc_errval == 0)) {
916 __u64 data_version = 0;
918 /* Get lsm for this fid */
919 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
921 hpk.hpk_flags |= HP_FLAG_RETRY;
922 /* hpk_errval must be >= 0 */
923 hpk.hpk_errval = -PTR_ERR(inode);
924 GOTO(progress, rc = PTR_ERR(inode));
927 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
930 CDEBUG(D_HSM, "Could not read file data version. "
931 "Request could not be confirmed.\n");
932 if (hpk.hpk_errval == 0)
933 hpk.hpk_errval = -rc;
937 /* Store in the hsm_copy for later copytool use.
938 * Always modified even if no lsm. */
939 hpk.hpk_data_version = data_version;
941 /* File could have been stripped during archiving, so we need
942 * to check anyway. */
943 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
944 (copy->hc_data_version != data_version)) {
945 CDEBUG(D_HSM, "File data version mismatched. "
946 "File content was changed during archiving. "
947 DFID", start:%#llx current:%#llx\n",
948 PFID(©->hc_hai.hai_fid),
949 copy->hc_data_version, data_version);
950 /* File was changed, send error to cdt. Do not ask for
951 * retry because if a file is modified frequently,
952 * the cdt will loop on retried archive requests.
953 * The policy engine will ask for a new archive later
954 * when the file will not be modified for some tunable
956 hpk.hpk_flags &= ~HP_FLAG_RETRY;
958 /* hpk_errval must be >= 0 */
959 hpk.hpk_errval = -rc;
966 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
969 /* Return first error */
970 RETURN(rc != 0 ? rc : rc2);
974 static int copy_and_ct_start(int cmd, struct obd_export *exp,
975 const struct lustre_kernelcomm __user *data)
977 struct lustre_kernelcomm *lk;
978 struct lustre_kernelcomm *tmp;
979 size_t size = sizeof(*lk);
984 /* copy data from userspace to get numbers of archive_id */
989 if (copy_from_user(lk, data, size))
990 GOTO(out_lk, rc = -EFAULT);
992 if (lk->lk_flags & LK_FLG_STOP)
995 if (!(lk->lk_flags & LK_FLG_DATANR)) {
996 __u32 archive_mask = lk->lk_data_count;
999 /* old hsm agent to old MDS */
1000 if (!exp_connect_archive_id_array(exp))
1003 /* old hsm agent to new MDS */
1004 lk->lk_flags |= LK_FLG_DATANR;
1006 if (archive_mask == 0)
1009 count = hweight32(archive_mask);
1010 new_size = offsetof(struct lustre_kernelcomm, lk_data[count]);
1011 OBD_ALLOC(tmp, new_size);
1013 GOTO(out_lk, rc = -ENOMEM);
1015 memcpy(tmp, lk, size);
1016 tmp->lk_data_count = count;
1022 for (i = 0; i < sizeof(archive_mask) * 8; i++) {
1023 if (BIT(i) & archive_mask) {
1024 lk->lk_data[count] = i + 1;
1031 /* new hsm agent to new mds */
1032 if (lk->lk_data_count > 0) {
1033 new_size = offsetof(struct lustre_kernelcomm,
1034 lk_data[lk->lk_data_count]);
1035 OBD_ALLOC(tmp, new_size);
1037 GOTO(out_lk, rc = -ENOMEM);
1043 if (copy_from_user(lk, data, size))
1044 GOTO(out_lk, rc = -EFAULT);
1047 /* new hsm agent to old MDS */
1048 if (!exp_connect_archive_id_array(exp)) {
1051 if (lk->lk_data_count > LL_HSM_ORIGIN_MAX_ARCHIVE)
1052 GOTO(out_lk, rc = -EINVAL);
1054 for (i = 0; i < lk->lk_data_count; i++) {
1055 if (lk->lk_data[i] > LL_HSM_ORIGIN_MAX_ARCHIVE) {
1057 CERROR("%s: archive id %d requested but only "
1058 "[0 - %zu] supported: rc = %d\n",
1059 exp->exp_obd->obd_name, lk->lk_data[i],
1060 LL_HSM_ORIGIN_MAX_ARCHIVE, rc);
1064 if (lk->lk_data[i] == 0) {
1069 archives |= (1 << (lk->lk_data[i] - 1));
1071 lk->lk_flags &= ~LK_FLG_DATANR;
1072 lk->lk_data_count = archives;
1075 rc = obd_iocontrol(cmd, exp, size, lk, NULL);
1081 static int check_owner(int type, int id)
1085 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
1089 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1098 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1100 int cmd = qctl->qc_cmd;
1101 int type = qctl->qc_type;
1102 int id = qctl->qc_id;
1103 int valid = qctl->qc_valid;
1110 case LUSTRE_Q_SETDEFAULT:
1111 case LUSTRE_Q_SETQUOTAPOOL:
1112 case LUSTRE_Q_SETINFOPOOL:
1113 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1117 case LUSTRE_Q_GETDEFAULT:
1118 case LUSTRE_Q_GETQUOTAPOOL:
1119 if (check_owner(type, id) &&
1120 (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1124 case LUSTRE_Q_GETINFOPOOL:
1127 CERROR("unsupported quotactl op: %#x\n", cmd);
1131 if (valid != QC_GENERAL) {
1132 if (cmd == Q_GETINFO)
1133 qctl->qc_cmd = Q_GETOINFO;
1134 else if (cmd == Q_GETQUOTA ||
1135 cmd == LUSTRE_Q_GETQUOTAPOOL)
1136 qctl->qc_cmd = Q_GETOQUOTA;
1142 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1143 sizeof(*qctl), qctl, NULL);
1146 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1147 sizeof(*qctl), qctl, NULL);
1150 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1151 sizeof(*qctl), qctl, NULL);
1153 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1155 sizeof(*qctl), qctl, NULL);
1167 struct obd_quotactl *oqctl;
1168 int oqctl_len = sizeof(*oqctl);
1170 if (LUSTRE_Q_CMD_IS_POOL(cmd))
1171 oqctl_len += LOV_MAXPOOLNAME + 1;
1173 OBD_ALLOC(oqctl, oqctl_len);
1177 QCTL_COPY(oqctl, qctl);
1178 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1180 OBD_FREE(oqctl, oqctl_len);
1183 /* If QIF_SPACE is not set, client should collect the
1184 * space usage from OSSs by itself */
1185 if ((cmd == Q_GETQUOTA || cmd == LUSTRE_Q_GETQUOTAPOOL) &&
1186 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1187 !oqctl->qc_dqblk.dqb_curspace) {
1188 struct obd_quotactl *oqctl_tmp;
1190 OBD_ALLOC_PTR(oqctl_tmp);
1191 if (oqctl_tmp == NULL)
1192 GOTO(out, rc = -ENOMEM);
1194 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1195 oqctl_tmp->qc_id = oqctl->qc_id;
1196 oqctl_tmp->qc_type = oqctl->qc_type;
1198 /* collect space usage from OSTs */
1199 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1200 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1201 if (!rc || rc == -EREMOTEIO) {
1202 oqctl->qc_dqblk.dqb_curspace =
1203 oqctl_tmp->qc_dqblk.dqb_curspace;
1204 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1207 /* collect space & inode usage from MDTs */
1208 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1209 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1210 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1211 if (!rc || rc == -EREMOTEIO) {
1212 oqctl->qc_dqblk.dqb_curspace +=
1213 oqctl_tmp->qc_dqblk.dqb_curspace;
1214 oqctl->qc_dqblk.dqb_curinodes =
1215 oqctl_tmp->qc_dqblk.dqb_curinodes;
1216 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1218 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1221 OBD_FREE_PTR(oqctl_tmp);
1224 QCTL_COPY(qctl, oqctl);
1225 OBD_FREE(oqctl, oqctl_len);
1231 int ll_rmfid(struct file *file, void __user *arg)
1233 const struct fid_array __user *ufa = arg;
1234 struct fid_array *lfa = NULL;
1237 int i, rc, *rcs = NULL;
1240 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
1241 !(ll_i2sbi(file_inode(file))->ll_flags & LL_SBI_USER_FID2PATH))
1243 /* Only need to get the buflen */
1244 if (get_user(nr, &ufa->fa_nr))
1246 /* DoS protection */
1247 if (nr > OBD_MAX_FIDS_IN_ARRAY)
1250 size = offsetof(struct fid_array, fa_fids[nr]);
1251 OBD_ALLOC(lfa, size);
1254 OBD_ALLOC_PTR_ARRAY(rcs, nr);
1256 GOTO(free_lfa, rc = -ENOMEM);
1258 if (copy_from_user(lfa, arg, size))
1259 GOTO(free_rcs, rc = -EFAULT);
1261 /* Call mdc_iocontrol */
1262 rc = md_rmfid(ll_i2mdexp(file_inode(file)), lfa, rcs, NULL);
1264 for (i = 0; i < nr; i++)
1266 lfa->fa_fids[i].f_ver = rcs[i];
1267 if (copy_to_user(arg, lfa, size))
1272 OBD_FREE_PTR_ARRAY(rcs, nr);
1274 OBD_FREE(lfa, size);
1279 /* This function tries to get a single name component,
1280 * to send to the server. No actual path traversal involved,
1281 * so we limit to NAME_MAX */
1282 static char *ll_getname(const char __user *filename)
1287 OBD_ALLOC(tmp, NAME_MAX + 1);
1290 return ERR_PTR(-ENOMEM);
1292 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1295 else if (len > NAME_MAX)
1296 ret = -ENAMETOOLONG;
1299 OBD_FREE(tmp, NAME_MAX + 1);
1305 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1307 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1309 struct dentry *dentry = file_dentry(file);
1310 struct inode *inode = file_inode(file);
1311 struct ll_sb_info *sbi = ll_i2sbi(inode);
1312 struct obd_ioctl_data *data;
1316 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1317 PFID(ll_inode2fid(inode)), inode, cmd);
1319 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1320 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1323 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1325 case FS_IOC_GETFLAGS:
1326 case FS_IOC_SETFLAGS:
1327 RETURN(ll_iocontrol(inode, file, cmd, arg));
1328 case FSFILT_IOC_GETVERSION:
1329 case FS_IOC_GETVERSION:
1330 RETURN(put_user(inode->i_generation, (int __user *)arg));
1331 /* We need to special case any other ioctls we want to handle,
1332 * to send them to the MDS/OST as appropriate and to properly
1333 * network encode the arg field. */
1334 case FS_IOC_SETVERSION:
1337 case LL_IOC_GET_MDTIDX: {
1340 mdtidx = ll_get_mdt_idx(inode);
1344 if (put_user((int)mdtidx, (int __user *)arg))
1349 case IOC_MDC_LOOKUP: {
1350 int namelen, len = 0;
1354 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1359 filename = data->ioc_inlbuf1;
1360 namelen = strlen(filename);
1362 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1363 GOTO(out_free, rc = -EINVAL);
1366 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1368 CERROR("%s: lookup %.*s failed: rc = %d\n",
1369 sbi->ll_fsname, namelen, filename, rc);
1373 OBD_FREE_LARGE(buf, len);
1376 case LL_IOC_LMV_SETSTRIPE: {
1377 struct lmv_user_md *lum;
1386 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1391 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1392 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1393 GOTO(lmv_out_free, rc = -EINVAL);
1395 filename = data->ioc_inlbuf1;
1396 namelen = data->ioc_inllen1;
1399 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1400 GOTO(lmv_out_free, rc = -EINVAL);
1402 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1403 lumlen = data->ioc_inllen2;
1405 if (!lmv_user_magic_supported(lum->lum_magic)) {
1406 CERROR("%s: wrong lum magic %x : rc = %d\n", filename,
1407 lum->lum_magic, -EINVAL);
1408 GOTO(lmv_out_free, rc = -EINVAL);
1411 if ((lum->lum_magic == LMV_USER_MAGIC ||
1412 lum->lum_magic == LMV_USER_MAGIC_SPECIFIC) &&
1413 lumlen < sizeof(*lum)) {
1414 CERROR("%s: wrong lum size %d for magic %x : rc = %d\n",
1415 filename, lumlen, lum->lum_magic, -EINVAL);
1416 GOTO(lmv_out_free, rc = -EINVAL);
1419 if (lum->lum_magic == LMV_MAGIC_FOREIGN &&
1420 lumlen < sizeof(struct lmv_foreign_md)) {
1421 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1422 filename, lum->lum_magic, lumlen, -EFAULT);
1423 GOTO(lmv_out_free, rc = -EINVAL);
1426 mode = data->ioc_type;
1427 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1429 OBD_FREE_LARGE(buf, len);
1433 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1434 struct lmv_user_md lum;
1435 struct lmv_user_md __user *ulump =
1436 (struct lmv_user_md __user *)arg;
1439 if (copy_from_user(&lum, ulump, sizeof(lum)))
1442 if (lum.lum_magic != LMV_USER_MAGIC)
1445 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1449 case LL_IOC_LOV_SETSTRIPE_NEW:
1450 case LL_IOC_LOV_SETSTRIPE: {
1451 struct lov_user_md_v3 *lumv3 = NULL;
1452 struct lov_user_md_v1 lumv1;
1453 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1454 struct lov_user_md_v1 __user *lumv1p =
1455 (struct lov_user_md_v1 __user *)arg;
1456 struct lov_user_md_v3 __user *lumv3p =
1457 (struct lov_user_md_v3 __user *)arg;
1460 int set_default = 0;
1462 BUILD_BUG_ON(sizeof(struct lov_user_md_v3) <=
1463 sizeof(struct lov_comp_md_v1));
1464 BUILD_BUG_ON(sizeof(*lumv3) != sizeof(*lumv3p));
1465 /* first try with v1 which is smaller than v3 */
1466 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1469 if (inode->i_sb->s_root == file_dentry(file))
1472 switch (lumv1.lmm_magic) {
1473 case LOV_USER_MAGIC_V3:
1474 case LOV_USER_MAGIC_SPECIFIC:
1475 lum_size = ll_lov_user_md_size(&lumv1);
1478 OBD_ALLOC(lumv3, lum_size);
1481 if (copy_from_user(lumv3, lumv3p, lum_size))
1482 GOTO(out, rc = -EFAULT);
1483 lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1485 case LOV_USER_MAGIC_V1:
1488 GOTO(out, rc = -ENOTSUPP);
1491 /* in v1 and v3 cases lumv1 points to data */
1492 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1495 OBD_FREE(lumv3, lum_size);
1498 case LL_IOC_LMV_GETSTRIPE: {
1499 struct lmv_user_md __user *ulmv =
1500 (struct lmv_user_md __user *)arg;
1501 struct lmv_user_md lum;
1502 struct ptlrpc_request *request = NULL;
1503 struct ptlrpc_request *root_request = NULL;
1504 union lmv_mds_md *lmm = NULL;
1507 struct lmv_user_md *tmp = NULL;
1511 int max_stripe_count;
1515 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1518 max_stripe_count = lum.lum_stripe_count;
1519 /* lum_magic will indicate which stripe the ioctl will like
1520 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1521 * is for default LMV stripe */
1522 if (lum.lum_magic == LMV_MAGIC_V1)
1523 valid |= OBD_MD_MEA;
1524 else if (lum.lum_magic == LMV_USER_MAGIC)
1525 valid |= OBD_MD_DEFAULT_MEA;
1529 rc = ll_dir_getstripe_default(inode, (void **)&lmm, &lmmsize,
1530 &request, &root_request, valid);
1532 GOTO(finish_req, rc);
1534 /* Get default LMV EA */
1535 if (lum.lum_magic == LMV_USER_MAGIC) {
1536 if (lmmsize > sizeof(*ulmv))
1537 GOTO(finish_req, rc = -EINVAL);
1539 if (copy_to_user(ulmv, lmm, lmmsize))
1540 GOTO(finish_req, rc = -EFAULT);
1542 GOTO(finish_req, rc);
1545 /* if foreign LMV case, fake stripes number */
1546 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1547 struct lmv_foreign_md *lfm;
1549 lfm = (struct lmv_foreign_md *)lmm;
1550 if (lfm->lfm_length < XATTR_SIZE_MAX -
1551 offsetof(typeof(*lfm), lfm_value)) {
1552 __u32 size = lfm->lfm_length +
1553 offsetof(typeof(*lfm), lfm_value);
1555 stripe_count = lmv_foreign_to_md_stripes(size);
1557 CERROR("invalid %d foreign size returned\n",
1562 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1564 if (max_stripe_count < stripe_count) {
1565 lum.lum_stripe_count = stripe_count;
1566 if (copy_to_user(ulmv, &lum, sizeof(lum)))
1567 GOTO(finish_req, rc = -EFAULT);
1568 GOTO(finish_req, rc = -E2BIG);
1571 /* enough room on user side and foreign case */
1572 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1573 struct lmv_foreign_md *lfm;
1576 lfm = (struct lmv_foreign_md *)lmm;
1577 size = lfm->lfm_length +
1578 offsetof(struct lmv_foreign_md, lfm_value);
1579 if (copy_to_user(ulmv, lfm, size))
1580 GOTO(finish_req, rc = -EFAULT);
1581 GOTO(finish_req, rc);
1584 lum_size = lmv_user_md_size(stripe_count,
1585 LMV_USER_MAGIC_SPECIFIC);
1586 OBD_ALLOC(tmp, lum_size);
1588 GOTO(finish_req, rc = -ENOMEM);
1590 mdt_index = ll_get_mdt_idx(inode);
1592 GOTO(out_tmp, rc = -ENOMEM);
1594 tmp->lum_magic = LMV_MAGIC_V1;
1595 tmp->lum_stripe_count = 0;
1596 tmp->lum_stripe_offset = mdt_index;
1597 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1598 for (i = 0; i < stripe_count; i++) {
1601 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1602 if (fid_is_sane(&fid)) {
1603 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1605 GOTO(out_tmp, rc = mdt_index);
1607 tmp->lum_objects[i].lum_mds = mdt_index;
1608 tmp->lum_objects[i].lum_fid = fid;
1611 tmp->lum_stripe_count++;
1614 if (copy_to_user(ulmv, tmp, lum_size))
1615 GOTO(out_tmp, rc = -EFAULT);
1617 OBD_FREE(tmp, lum_size);
1619 ptlrpc_req_finished(request);
1620 ptlrpc_req_finished(root_request);
1624 case LL_IOC_REMOVE_ENTRY: {
1625 char *filename = NULL;
1629 /* Here is a little hack to avoid sending REINT_RMENTRY to
1630 * unsupported server, which might crash the server(LU-2730),
1631 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1632 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1633 * server will support REINT_RMENTRY XXX*/
1634 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1637 filename = ll_getname((const char __user *)arg);
1638 if (IS_ERR(filename))
1639 RETURN(PTR_ERR(filename));
1641 namelen = strlen(filename);
1643 GOTO(out_rmdir, rc = -EINVAL);
1645 rc = ll_rmdir_entry(inode, filename, namelen);
1648 ll_putname(filename);
1652 RETURN(ll_rmfid(file, (void __user *)arg));
1653 case LL_IOC_LOV_SWAP_LAYOUTS:
1655 case IOC_OBD_STATFS:
1656 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1657 case LL_IOC_LOV_GETSTRIPE:
1658 case LL_IOC_LOV_GETSTRIPE_NEW:
1659 case LL_IOC_MDC_GETINFO:
1660 case LL_IOC_MDC_GETINFO_OLD:
1661 case IOC_MDC_GETFILEINFO:
1662 case IOC_MDC_GETFILEINFO_OLD:
1663 case IOC_MDC_GETFILESTRIPE: {
1664 struct ptlrpc_request *request = NULL;
1665 struct ptlrpc_request *root_request = NULL;
1666 struct lov_user_md __user *lump;
1667 struct lov_mds_md *lmm = NULL;
1668 struct mdt_body *body;
1669 char *filename = NULL;
1670 lstat_t __user *statp = NULL;
1671 lstatx_t __user *stxp = NULL;
1672 __u64 __user *flagsp = NULL;
1673 __u32 __user *lmmsizep = NULL;
1674 struct lu_fid __user *fidp = NULL;
1677 if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1678 cmd == IOC_MDC_GETFILEINFO ||
1679 cmd == IOC_MDC_GETFILESTRIPE) {
1680 filename = ll_getname((const char __user *)arg);
1681 if (IS_ERR(filename))
1682 RETURN(PTR_ERR(filename));
1684 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1685 &lmmsize, &request);
1687 rc = ll_dir_getstripe_default(inode, (void **)&lmm,
1693 body = req_capsule_server_get(&request->rq_pill,
1695 LASSERT(body != NULL);
1700 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1701 cmd == LL_IOC_MDC_GETINFO ||
1702 cmd == IOC_MDC_GETFILEINFO_OLD ||
1703 cmd == LL_IOC_MDC_GETINFO_OLD)) {
1711 if (cmd == IOC_MDC_GETFILESTRIPE ||
1712 cmd == LL_IOC_LOV_GETSTRIPE ||
1713 cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1714 lump = (struct lov_user_md __user *)arg;
1715 } else if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1716 cmd == LL_IOC_MDC_GETINFO_OLD){
1717 struct lov_user_mds_data_v1 __user *lmdp;
1719 lmdp = (struct lov_user_mds_data_v1 __user *)arg;
1720 statp = &lmdp->lmd_st;
1721 lump = &lmdp->lmd_lmm;
1723 struct lov_user_mds_data __user *lmdp;
1725 lmdp = (struct lov_user_mds_data __user *)arg;
1726 fidp = &lmdp->lmd_fid;
1727 stxp = &lmdp->lmd_stx;
1728 flagsp = &lmdp->lmd_flags;
1729 lmmsizep = &lmdp->lmd_lmmsize;
1730 lump = &lmdp->lmd_lmm;
1734 /* If the file has no striping then zero out *lump so
1735 * that the caller isn't confused by garbage. */
1736 if (clear_user(lump, sizeof(*lump)))
1737 GOTO(out_req, rc = -EFAULT);
1738 } else if (copy_to_user(lump, lmm, lmmsize)) {
1739 if (copy_to_user(lump, lmm, sizeof(*lump)))
1740 GOTO(out_req, rc = -EFAULT);
1744 if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1745 cmd == LL_IOC_MDC_GETINFO_OLD) {
1748 st.st_dev = inode->i_sb->s_dev;
1749 st.st_mode = body->mbo_mode;
1750 st.st_nlink = body->mbo_nlink;
1751 st.st_uid = body->mbo_uid;
1752 st.st_gid = body->mbo_gid;
1753 st.st_rdev = body->mbo_rdev;
1754 st.st_size = body->mbo_size;
1755 st.st_blksize = PAGE_SIZE;
1756 st.st_blocks = body->mbo_blocks;
1757 st.st_atime = body->mbo_atime;
1758 st.st_mtime = body->mbo_mtime;
1759 st.st_ctime = body->mbo_ctime;
1760 st.st_ino = cl_fid_build_ino(&body->mbo_fid1,
1764 if (copy_to_user(statp, &st, sizeof(st)))
1765 GOTO(out_req, rc = -EFAULT);
1766 } else if (cmd == IOC_MDC_GETFILEINFO ||
1767 cmd == LL_IOC_MDC_GETINFO) {
1768 lstatx_t stx = { 0 };
1769 __u64 valid = body->mbo_valid;
1771 stx.stx_blksize = PAGE_SIZE;
1772 stx.stx_nlink = body->mbo_nlink;
1773 stx.stx_uid = body->mbo_uid;
1774 stx.stx_gid = body->mbo_gid;
1775 stx.stx_mode = body->mbo_mode;
1776 stx.stx_ino = cl_fid_build_ino(&body->mbo_fid1,
1779 stx.stx_size = body->mbo_size;
1780 stx.stx_blocks = body->mbo_blocks;
1781 stx.stx_atime.tv_sec = body->mbo_atime;
1782 stx.stx_ctime.tv_sec = body->mbo_ctime;
1783 stx.stx_mtime.tv_sec = body->mbo_mtime;
1784 stx.stx_btime.tv_sec = body->mbo_btime;
1785 stx.stx_rdev_major = MAJOR(body->mbo_rdev);
1786 stx.stx_rdev_minor = MINOR(body->mbo_rdev);
1787 stx.stx_dev_major = MAJOR(inode->i_sb->s_dev);
1788 stx.stx_dev_minor = MINOR(inode->i_sb->s_dev);
1789 stx.stx_mask |= STATX_BASIC_STATS | STATX_BTIME;
1792 * For a striped directory, the size and blocks returned
1793 * from MDT is not correct.
1794 * The size and blocks are aggregated by client across
1796 * Thus for a striped directory, do not return the valid
1797 * FLSIZE and FLBLOCKS flags to the caller.
1798 * However, this whould be better decided by the MDS
1799 * instead of the client.
1801 if (cmd == LL_IOC_MDC_GETINFO &&
1802 ll_i2info(inode)->lli_lsm_md != NULL)
1803 valid &= ~(OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
1805 if (flagsp && copy_to_user(flagsp, &valid,
1807 GOTO(out_req, rc = -EFAULT);
1809 if (fidp && copy_to_user(fidp, &body->mbo_fid1,
1811 GOTO(out_req, rc = -EFAULT);
1813 if (!(valid & OBD_MD_FLSIZE))
1814 stx.stx_mask &= ~STATX_SIZE;
1815 if (!(valid & OBD_MD_FLBLOCKS))
1816 stx.stx_mask &= ~STATX_BLOCKS;
1818 if (stxp && copy_to_user(stxp, &stx, sizeof(stx)))
1819 GOTO(out_req, rc = -EFAULT);
1821 if (lmmsizep && copy_to_user(lmmsizep, &lmmsize,
1823 GOTO(out_req, rc = -EFAULT);
1828 ptlrpc_req_finished(request);
1829 ptlrpc_req_finished(root_request);
1831 ll_putname(filename);
1834 case OBD_IOC_QUOTACTL: {
1835 struct if_quotactl *qctl;
1836 int qctl_len = sizeof(*qctl) + LOV_MAXPOOLNAME + 1;
1838 OBD_ALLOC(qctl, qctl_len);
1842 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1843 GOTO(out_quotactl, rc = -EFAULT);
1845 if (LUSTRE_Q_CMD_IS_POOL(qctl->qc_cmd)) {
1846 char __user *from = (char __user *)arg +
1847 offsetof(typeof(*qctl), qc_poolname);
1848 if (copy_from_user(qctl->qc_poolname, from,
1849 LOV_MAXPOOLNAME + 1))
1850 GOTO(out_quotactl, rc = -EFAULT);
1853 rc = quotactl_ioctl(sbi, qctl);
1855 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1859 OBD_FREE(qctl, qctl_len);
1862 case OBD_IOC_GETDTNAME:
1863 case OBD_IOC_GETMDNAME:
1864 RETURN(ll_get_obd_name(inode, cmd, arg));
1865 case LL_IOC_FLUSHCTX:
1866 RETURN(ll_flush_ctx(inode));
1867 case LL_IOC_GETOBDCOUNT: {
1869 struct obd_export *exp;
1871 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1874 /* get ost count when count is zero, get mdt count otherwise */
1875 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1876 vallen = sizeof(count);
1877 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1878 KEY_TGT_COUNT, &vallen, &count);
1880 CERROR("get target count failed: %d\n", rc);
1884 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1889 case LL_IOC_PATH2FID:
1890 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1891 sizeof(struct lu_fid)))
1894 case LL_IOC_GET_CONNECT_FLAGS: {
1895 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1896 (void __user *)arg));
1898 case OBD_IOC_FID2PATH:
1899 RETURN(ll_fid2path(inode, (void __user *)arg));
1900 case LL_IOC_GETPARENT:
1901 RETURN(ll_getparent(file, (void __user *)arg));
1902 case LL_IOC_FID2MDTIDX: {
1903 struct obd_export *exp = ll_i2mdexp(inode);
1907 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1911 /* Call mdc_iocontrol */
1912 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1913 (__u32 __user *)&index);
1919 case LL_IOC_HSM_REQUEST: {
1920 struct hsm_user_request *hur;
1927 /* We don't know the true size yet; copy the fixed-size part */
1928 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1933 /* Compute the whole struct size */
1934 totalsize = hur_len(hur);
1939 /* Final size will be more than double totalsize */
1940 if (totalsize >= MDS_MAXREQSIZE / 3)
1943 OBD_ALLOC_LARGE(hur, totalsize);
1947 /* Copy the whole struct */
1948 if (copy_from_user(hur, (void __user *)arg, totalsize))
1949 GOTO(out_hur, rc = -EFAULT);
1951 if (hur->hur_request.hr_action == HUA_RELEASE) {
1952 const struct lu_fid *fid;
1956 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1957 fid = &hur->hur_user_item[i].hui_fid;
1958 f = search_inode_for_lustre(inode->i_sb, fid);
1964 rc = ll_hsm_release(f);
1970 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1975 OBD_FREE_LARGE(hur, totalsize);
1979 case LL_IOC_HSM_PROGRESS: {
1980 struct hsm_progress_kernel hpk;
1981 struct hsm_progress hp;
1983 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1986 hpk.hpk_fid = hp.hp_fid;
1987 hpk.hpk_cookie = hp.hp_cookie;
1988 hpk.hpk_extent = hp.hp_extent;
1989 hpk.hpk_flags = hp.hp_flags;
1990 hpk.hpk_errval = hp.hp_errval;
1991 hpk.hpk_data_version = 0;
1993 /* File may not exist in Lustre; all progress
1994 * reported to Lustre root */
1995 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1999 case LL_IOC_HSM_CT_START:
2000 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2003 rc = copy_and_ct_start(cmd, sbi->ll_md_exp,
2004 (struct lustre_kernelcomm __user *)arg);
2007 case LL_IOC_HSM_COPY_START: {
2008 struct hsm_copy *copy;
2011 OBD_ALLOC_PTR(copy);
2014 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2019 rc = ll_ioc_copy_start(inode->i_sb, copy);
2020 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2026 case LL_IOC_HSM_COPY_END: {
2027 struct hsm_copy *copy;
2030 OBD_ALLOC_PTR(copy);
2033 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2038 rc = ll_ioc_copy_end(inode->i_sb, copy);
2039 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2045 case LL_IOC_MIGRATE: {
2046 struct lmv_user_md *lum;
2053 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
2057 data = (struct obd_ioctl_data *)buf;
2058 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
2059 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
2060 GOTO(migrate_free, rc = -EINVAL);
2062 filename = data->ioc_inlbuf1;
2063 namelen = data->ioc_inllen1;
2065 if (namelen < 1 || namelen != strlen(filename) + 1) {
2066 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
2067 GOTO(migrate_free, rc = -EINVAL);
2070 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
2071 if (lum->lum_magic != LMV_USER_MAGIC &&
2072 lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
2074 CERROR("%s: wrong lum magic %x: rc = %d\n",
2075 filename, lum->lum_magic, rc);
2076 GOTO(migrate_free, rc);
2079 rc = ll_migrate(inode, file, lum, filename);
2081 OBD_FREE_LARGE(buf, len);
2085 case LL_IOC_FSGETXATTR:
2086 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
2087 case LL_IOC_FSSETXATTR:
2088 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
2089 case LL_IOC_PCC_DETACH_BY_FID: {
2090 struct lu_pcc_detach_fid *detach;
2092 struct inode *inode2;
2095 OBD_ALLOC_PTR(detach);
2099 if (copy_from_user(detach,
2100 (const struct lu_pcc_detach_fid __user *)arg,
2102 GOTO(out_detach, rc = -EFAULT);
2104 fid = &detach->pccd_fid;
2105 ino = cl_fid_build_ino(fid, ll_need_32bit_api(sbi));
2106 inode2 = ilookup5(inode->i_sb, ino, ll_test_inode_by_fid, fid);
2108 /* Target inode is not in inode cache, and PCC file
2109 * has aleady released, return immdiately.
2111 GOTO(out_detach, rc = 0);
2113 if (!S_ISREG(inode2->i_mode))
2114 GOTO(out_iput, rc = -EINVAL);
2116 if (!inode_owner_or_capable(inode2))
2117 GOTO(out_iput, rc = -EPERM);
2119 rc = pcc_ioctl_detach(inode2, detach->pccd_opt);
2123 OBD_FREE_PTR(detach);
2127 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
2128 (void __user *)arg));
2132 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
2134 struct inode *inode = file->f_mapping->host;
2135 struct ll_file_data *fd = file->private_data;
2136 struct ll_sb_info *sbi = ll_i2sbi(inode);
2137 int api32 = ll_need_32bit_api(sbi);
2138 loff_t ret = -EINVAL;
2146 offset += file->f_pos;
2152 offset += LL_DIR_END_OFF_32BIT;
2154 offset += LL_DIR_END_OFF;
2161 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
2162 (!api32 && offset <= LL_DIR_END_OFF))) {
2163 if (offset != file->f_pos) {
2164 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
2165 (!api32 && offset == LL_DIR_END_OFF))
2166 fd->lfd_pos = MDS_DIR_END_OFF;
2167 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
2168 fd->lfd_pos = offset << 32;
2170 fd->lfd_pos = offset;
2171 file->f_pos = offset;
2172 file->f_version = 0;
2179 inode_unlock(inode);
2183 static int ll_dir_open(struct inode *inode, struct file *file)
2186 RETURN(ll_file_open(inode, file));
2189 static int ll_dir_release(struct inode *inode, struct file *file)
2192 RETURN(ll_file_release(inode, file));
2195 const struct file_operations ll_dir_operations = {
2196 .llseek = ll_dir_seek,
2197 .open = ll_dir_open,
2198 .release = ll_dir_release,
2199 .read = generic_read_dir,
2200 #ifdef HAVE_DIR_CONTEXT
2201 .iterate_shared = ll_iterate,
2203 .readdir = ll_readdir,
2205 .unlocked_ioctl = ll_dir_ioctl,