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, 2015, 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/user_namespace.h>
42 #ifdef HAVE_UIDGID_HEADER
43 # include <linux/uidgid.h>
45 #include <asm/uaccess.h>
46 #include <linux/buffer_head.h> // for wait_on_buffer
47 #include <linux/pagevec.h>
49 #define DEBUG_SUBSYSTEM S_LLITE
51 #include <lustre/lustre_idl.h>
53 #include <obd_support.h>
54 #include <obd_class.h>
55 #include <lustre_ioctl.h>
56 #include <lustre_lib.h>
57 #include <lustre_dlm.h>
58 #include <lustre_fid.h>
59 #include <lustre_kernelcomm.h>
60 #include <lustre_swab.h>
62 #include "llite_internal.h"
65 * (new) readdir implementation overview.
67 * Original lustre readdir implementation cached exact copy of raw directory
68 * pages on the client. These pages were indexed in client page cache by
69 * logical offset in the directory file. This design, while very simple and
70 * intuitive had some inherent problems:
72 * . it implies that byte offset to the directory entry serves as a
73 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
74 * ext3/htree directory entries may move due to splits, and more
77 * . it is incompatible with the design of split directories for cmd3,
78 * that assumes that names are distributed across nodes based on their
79 * hash, and so readdir should be done in hash order.
81 * New readdir implementation does readdir in hash order, and uses hash of a
82 * file name as a telldir/seekdir cookie. This led to number of complications:
84 * . hash is not unique, so it cannot be used to index cached directory
85 * pages on the client (note, that it requires a whole pageful of hash
86 * collided entries to cause two pages to have identical hashes);
88 * . hash is not unique, so it cannot, strictly speaking, be used as an
89 * entry cookie. ext3/htree has the same problem and lustre implementation
90 * mimics their solution: seekdir(hash) positions directory at the first
91 * entry with the given hash.
97 * Client caches directory pages using hash of the first entry as an index. As
98 * noted above hash is not unique, so this solution doesn't work as is:
99 * special processing is needed for "page hash chains" (i.e., sequences of
100 * pages filled with entries all having the same hash value).
102 * First, such chains have to be detected. To this end, server returns to the
103 * client the hash of the first entry on the page next to one returned. When
104 * client detects that this hash is the same as hash of the first entry on the
105 * returned page, page hash collision has to be handled. Pages in the
106 * hash chain, except first one, are termed "overflow pages".
108 * Solution to index uniqueness problem is to not cache overflow
109 * pages. Instead, when page hash collision is detected, all overflow pages
110 * from emerging chain are immediately requested from the server and placed in
111 * a special data structure (struct ll_dir_chain). This data structure is used
112 * by ll_readdir() to process entries from overflow pages. When readdir
113 * invocation finishes, overflow pages are discarded. If page hash collision
114 * chain weren't completely processed, next call to readdir will again detect
115 * page hash collision, again read overflow pages in, process next portion of
116 * entries and again discard the pages. This is not as wasteful as it looks,
117 * because, given reasonable hash, page hash collisions are extremely rare.
119 * 1. directory positioning
121 * When seekdir(hash) is called, original
132 * identification of and access to overflow pages
136 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
137 * a header lu_dirpage which describes the start/end hash, and whether this
138 * page is empty (contains no dir entry) or hash collide with next page.
139 * After client receives reply, several pages will be integrated into dir page
140 * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
141 * lu_dirpage for this integrated page will be adjusted. See
142 * mdc_adjust_dirpages().
145 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
146 __u64 offset, struct ll_dir_chain *chain)
148 struct md_callback cb_op;
152 cb_op.md_blocking_ast = ll_md_blocking_ast;
153 rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
160 void ll_release_page(struct inode *inode, struct page *page,
165 /* Always remove the page for striped dir, because the page is
166 * built from temporarily in LMV layer */
167 if (inode != NULL && S_ISDIR(inode->i_mode) &&
168 ll_i2info(inode)->lli_lsm_md != NULL) {
175 if (likely(page->mapping != NULL))
176 truncate_complete_page(page->mapping, page);
183 * return IF_* type for given lu_dirent entry.
184 * IF_* flag shld be converted to particular OS file type in
185 * platform llite module.
187 static u16 ll_dirent_type_get(struct lu_dirent *ent)
190 struct luda_type *lt;
193 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
194 const unsigned align = sizeof(struct luda_type) - 1;
196 len = le16_to_cpu(ent->lde_namelen);
197 len = (len + align) & ~align;
198 lt = (void *)ent->lde_name + len;
199 type = IFTODT(le16_to_cpu(lt->lt_type));
205 #ifdef HAVE_DIR_CONTEXT
206 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
207 struct dir_context *ctx)
210 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
211 void *cookie, filldir_t filldir)
214 struct ll_sb_info *sbi = ll_i2sbi(inode);
216 bool is_api32 = ll_need_32bit_api(sbi);
217 bool is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
219 struct ll_dir_chain chain;
224 ll_dir_chain_init(&chain);
226 page = ll_get_dir_page(inode, op_data, pos, &chain);
228 while (rc == 0 && !done) {
229 struct lu_dirpage *dp;
230 struct lu_dirent *ent;
239 hash = MDS_DIR_END_OFF;
240 dp = page_address(page);
241 for (ent = lu_dirent_start(dp); ent != NULL && !done;
242 ent = lu_dirent_next(ent)) {
249 hash = le64_to_cpu(ent->lde_hash);
252 * Skip until we find target hash
257 namelen = le16_to_cpu(ent->lde_namelen);
264 if (is_api32 && is_hash64)
268 fid_le_to_cpu(&fid, &ent->lde_fid);
269 ino = cl_fid_build_ino(&fid, is_api32);
270 type = ll_dirent_type_get(ent);
271 /* For 'll_nfs_get_name_filldir()', it will try
272 * to access the 'ent' through its 'lde_name',
273 * so the parameter 'name' for 'filldir()' must
274 * be part of the 'ent'. */
275 #ifdef HAVE_DIR_CONTEXT
277 done = !dir_emit(ctx, ent->lde_name, namelen, ino,
280 done = filldir(cookie, ent->lde_name, namelen, lhash,
287 ll_release_page(inode, page, false);
291 next = le64_to_cpu(dp->ldp_hash_end);
293 if (pos == MDS_DIR_END_OFF) {
295 * End of directory reached.
298 ll_release_page(inode, page, false);
301 * Normal case: continue to the next
304 ll_release_page(inode, page,
305 le32_to_cpu(dp->ldp_flags) &
308 page = ll_get_dir_page(inode, op_data, pos,
312 #ifdef HAVE_DIR_CONTEXT
317 ll_dir_chain_fini(&chain);
321 #ifdef HAVE_DIR_CONTEXT
322 static int ll_iterate(struct file *filp, struct dir_context *ctx)
324 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
327 struct inode *inode = file_inode(filp);
328 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
329 struct ll_sb_info *sbi = ll_i2sbi(inode);
330 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
331 int api32 = ll_need_32bit_api(sbi);
332 struct md_op_data *op_data;
342 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
343 "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
344 inode, (unsigned long)pos, i_size_read(inode), api32);
346 if (pos == MDS_DIR_END_OFF)
352 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
353 LUSTRE_OPC_ANY, inode);
355 GOTO(out, rc = PTR_ERR(op_data));
357 if (unlikely(op_data->op_mea1 != NULL)) {
358 /* This is only needed for striped dir to fill ..,
359 * see lmv_read_entry */
360 if (file_dentry(filp)->d_parent != NULL &&
361 file_dentry(filp)->d_parent->d_inode != NULL) {
362 __u64 ibits = MDS_INODELOCK_UPDATE;
363 struct inode *parent =
364 file_dentry(filp)->d_parent->d_inode;
366 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
367 op_data->op_fid3 = *ll_inode2fid(parent);
370 /* If it can not find in cache, do lookup .. on the master
372 if (fid_is_zero(&op_data->op_fid3)) {
373 rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
375 ll_finish_md_op_data(op_data);
380 op_data->op_max_pages = sbi->ll_md_brw_pages;
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);
406 filp->f_version = inode->i_version;
410 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
415 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
416 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
418 struct mgs_send_param *msp;
425 strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
426 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
427 sizeof(struct mgs_send_param), msp, NULL);
429 CERROR("Failed to set parameter: %d\n", rc);
437 * Create striped directory with specified stripe(@lump)
439 * \param[in] dparent the parent of the directory.
440 * \param[in] lump the specified stripes.
441 * \param[in] dirname the name of the directory.
442 * \param[in] mode the specified mode of the directory.
444 * \retval =0 if striped directory is being created successfully.
445 * <0 if the creation is failed.
447 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
448 const char *dirname, umode_t mode)
450 struct inode *parent = dparent->d_inode;
451 struct ptlrpc_request *request = NULL;
452 struct md_op_data *op_data;
453 struct ll_sb_info *sbi = ll_i2sbi(parent);
454 struct inode *inode = NULL;
455 struct dentry dentry = {
459 .len = strlen(dirname),
460 .hash = full_name_hash(dirname, strlen(dirname)),
466 if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
469 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
470 "stripe_offset %d, stripe_count: %u\n",
471 PFID(ll_inode2fid(parent)), parent, dirname,
472 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
474 if (lump->lum_stripe_count > 1 &&
475 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
478 if (IS_DEADDIR(parent) &&
479 !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
482 if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
483 lustre_swab_lmv_user_md(lump);
485 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
486 mode &= ~current_umask();
487 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
488 op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
489 strlen(dirname), mode, LUSTRE_OPC_MKDIR,
492 RETURN(PTR_ERR(op_data));
494 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
495 /* selinux_dentry_init_security() uses dentry->d_parent and name
496 * to determine the security context for the file. So our fake
497 * dentry should be real enough for this purpose. */
498 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
499 &op_data->op_file_secctx_name,
500 &op_data->op_file_secctx,
501 &op_data->op_file_secctx_size);
503 GOTO(out_op_data, err);
506 op_data->op_cli_flags |= CLI_SET_MEA;
507 err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
508 from_kuid(&init_user_ns, current_fsuid()),
509 from_kgid(&init_user_ns, current_fsgid()),
510 cfs_curproc_cap_pack(), 0, &request);
512 GOTO(out_request, err);
514 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
516 err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
518 GOTO(out_inode, err);
520 dentry.d_inode = inode;
522 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
523 err = ll_inode_init_security(&dentry, inode, parent);
525 GOTO(out_inode, err);
532 ptlrpc_req_finished(request);
534 ll_finish_md_op_data(op_data);
539 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
542 struct ll_sb_info *sbi = ll_i2sbi(inode);
543 struct md_op_data *op_data;
544 struct ptlrpc_request *req = NULL;
546 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
547 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
548 struct obd_device *mgc = lsi->lsi_mgc;
555 * This is coming from userspace, so should be in
556 * local endian. But the MDS would like it in little
557 * endian, so we swab it before we send it.
559 switch (lump->lmm_magic) {
560 case LOV_USER_MAGIC_V1: {
561 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
562 lustre_swab_lov_user_md_v1(lump);
563 lum_size = sizeof(struct lov_user_md_v1);
566 case LOV_USER_MAGIC_V3: {
567 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
568 lustre_swab_lov_user_md_v3(
569 (struct lov_user_md_v3 *)lump);
570 lum_size = sizeof(struct lov_user_md_v3);
573 case LMV_USER_MAGIC: {
574 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
575 lustre_swab_lmv_user_md(
576 (struct lmv_user_md *)lump);
577 lum_size = sizeof(struct lmv_user_md);
581 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
582 " %#08x != %#08x nor %#08x\n",
583 lump->lmm_magic, LOV_USER_MAGIC_V1,
589 lum_size = sizeof(struct lov_user_md_v1);
592 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
593 LUSTRE_OPC_ANY, NULL);
595 RETURN(PTR_ERR(op_data));
597 /* swabbing is done in lov_setstripe() on server side */
598 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
599 ll_finish_md_op_data(op_data);
600 ptlrpc_req_finished(req);
604 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
606 * 2.9 server has stored filesystem default stripe in ROOT xattr,
607 * and it's stored into system config for backward compatibility.
609 * In the following we use the fact that LOV_USER_MAGIC_V1 and
610 * LOV_USER_MAGIC_V3 have the same initial fields so we do not
611 * need the make the distiction between the 2 versions
613 if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
617 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
619 GOTO(end, rc = -ENOMEM);
622 /* Get fsname and assume devname to be -MDT0000. */
623 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
624 strcat(buf, "-MDT0000.lov");
627 /* Set root stripesize */
628 sprintf(buf, ".stripesize=%u",
629 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
630 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
634 /* Set root stripecount */
635 sprintf(buf, ".stripecount=%hd",
636 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
637 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
641 /* Set root stripeoffset */
642 sprintf(buf, ".stripeoffset=%hd",
643 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
644 (typeof(lump->lmm_stripe_offset))(-1));
645 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
649 OBD_FREE(param, MGS_PARAM_MAXLEN);
656 * This function will be used to get default LOV/LMV/Default LMV
657 * @valid will be used to indicate which stripe it will retrieve
658 * OBD_MD_MEA LMV stripe EA
659 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
660 * otherwise Default LOV EA.
661 * Each time, it can only retrieve 1 stripe EA
663 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
664 struct ptlrpc_request **request, u64 valid)
666 struct ll_sb_info *sbi = ll_i2sbi(inode);
667 struct mdt_body *body;
668 struct lov_mds_md *lmm = NULL;
669 struct ptlrpc_request *req = NULL;
671 struct md_op_data *op_data;
674 rc = ll_get_default_mdsize(sbi, &lmm_size);
678 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
679 0, lmm_size, LUSTRE_OPC_ANY,
682 RETURN(PTR_ERR(op_data));
684 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
685 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
686 ll_finish_md_op_data(op_data);
688 CDEBUG(D_INFO, "md_getattr failed on inode "
689 DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
693 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
694 LASSERT(body != NULL);
696 lmm_size = body->mbo_eadatasize;
698 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
700 GOTO(out, rc = -ENODATA);
703 lmm = req_capsule_server_sized_get(&req->rq_pill,
704 &RMF_MDT_MD, lmm_size);
705 LASSERT(lmm != NULL);
708 * This is coming from the MDS, so is probably in
709 * little endian. We convert it to host endian before
710 * passing it to userspace.
712 /* We don't swab objects for directories */
713 switch (le32_to_cpu(lmm->lmm_magic)) {
715 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
716 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
719 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
720 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
723 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
724 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
727 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
728 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
731 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
736 *plmm_size = lmm_size;
741 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
743 struct md_op_data *op_data;
748 OBD_ALLOC_PTR(op_data);
752 op_data->op_flags |= MF_GET_MDT_IDX;
753 op_data->op_fid1 = *fid;
754 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
755 mdt_index = op_data->op_mds;
756 OBD_FREE_PTR(op_data);
764 * Get MDT index for the inode.
766 int ll_get_mdt_idx(struct inode *inode)
768 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
772 * Generic handler to do any pre-copy work.
774 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
775 * first information for it that real work has started.
777 * Moreover, for a ARCHIVE request, it will sample the file data version and
778 * store it in \a copy.
780 * \return 0 on success.
782 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
784 struct ll_sb_info *sbi = ll_s2sbi(sb);
785 struct hsm_progress_kernel hpk;
790 /* Forge a hsm_progress based on data from copy. */
791 hpk.hpk_fid = copy->hc_hai.hai_fid;
792 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
793 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
794 hpk.hpk_extent.length = 0;
797 hpk.hpk_data_version = 0;
800 /* For archive request, we need to read the current file version. */
801 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
803 __u64 data_version = 0;
805 /* Get inode for this fid */
806 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
808 hpk.hpk_flags |= HP_FLAG_RETRY;
809 /* hpk_errval is >= 0 */
810 hpk.hpk_errval = -PTR_ERR(inode);
811 GOTO(progress, rc = PTR_ERR(inode));
814 /* Read current file data version */
815 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
818 CDEBUG(D_HSM, "Could not read file data version of "
819 DFID" (rc = %d). Archive request ("
820 "%#llx) could not be done.\n",
821 PFID(©->hc_hai.hai_fid), rc,
822 copy->hc_hai.hai_cookie);
823 hpk.hpk_flags |= HP_FLAG_RETRY;
824 /* hpk_errval must be >= 0 */
825 hpk.hpk_errval = -rc;
829 /* Store in the hsm_copy for later copytool use.
830 * Always modified even if no lsm. */
831 copy->hc_data_version = data_version;
835 /* On error, the request should be considered as completed */
836 if (hpk.hpk_errval > 0)
837 hpk.hpk_flags |= HP_FLAG_COMPLETED;
839 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
842 /* Return first error */
843 RETURN(rc != 0 ? rc : rc2);
847 * Generic handler to do any post-copy work.
849 * It will send the last hsm_progress update to coordinator to inform it
850 * that copy is finished and whether it was successful or not.
853 * - for ARCHIVE request, it will sample the file data version and compare it
854 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
855 * will be considered as failed.
856 * - for RESTORE request, it will sample the file data version and send it to
857 * coordinator which is useful if the file was imported as 'released'.
859 * \return 0 on success.
861 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
863 struct ll_sb_info *sbi = ll_s2sbi(sb);
864 struct hsm_progress_kernel hpk;
869 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
870 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
871 * initialized if copy_end was called with copy == NULL.
874 /* Forge a hsm_progress based on data from copy. */
875 hpk.hpk_fid = copy->hc_hai.hai_fid;
876 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
877 hpk.hpk_extent = copy->hc_hai.hai_extent;
878 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
879 hpk.hpk_errval = copy->hc_errval;
880 hpk.hpk_data_version = 0;
882 /* For archive request, we need to check the file data was not changed.
884 * For restore request, we need to send the file data version, this is
885 * useful when the file was created using hsm_import.
887 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
888 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
889 (copy->hc_errval == 0)) {
891 __u64 data_version = 0;
893 /* Get lsm for this fid */
894 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
896 hpk.hpk_flags |= HP_FLAG_RETRY;
897 /* hpk_errval must be >= 0 */
898 hpk.hpk_errval = -PTR_ERR(inode);
899 GOTO(progress, rc = PTR_ERR(inode));
902 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
905 CDEBUG(D_HSM, "Could not read file data version. "
906 "Request could not be confirmed.\n");
907 if (hpk.hpk_errval == 0)
908 hpk.hpk_errval = -rc;
912 /* Store in the hsm_copy for later copytool use.
913 * Always modified even if no lsm. */
914 hpk.hpk_data_version = data_version;
916 /* File could have been stripped during archiving, so we need
917 * to check anyway. */
918 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
919 (copy->hc_data_version != data_version)) {
920 CDEBUG(D_HSM, "File data version mismatched. "
921 "File content was changed during archiving. "
922 DFID", start:%#llx current:%#llx\n",
923 PFID(©->hc_hai.hai_fid),
924 copy->hc_data_version, data_version);
925 /* File was changed, send error to cdt. Do not ask for
926 * retry because if a file is modified frequently,
927 * the cdt will loop on retried archive requests.
928 * The policy engine will ask for a new archive later
929 * when the file will not be modified for some tunable
931 hpk.hpk_flags &= ~HP_FLAG_RETRY;
933 /* hpk_errval must be >= 0 */
934 hpk.hpk_errval = -rc;
941 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
944 /* Return first error */
945 RETURN(rc != 0 ? rc : rc2);
949 static int copy_and_ioctl(int cmd, struct obd_export *exp,
950 const void __user *data, size_t size)
955 OBD_ALLOC(copy, size);
959 if (copy_from_user(copy, data, size)) {
964 rc = obd_iocontrol(cmd, exp, size, copy, NULL);
966 OBD_FREE(copy, size);
971 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
973 int cmd = qctl->qc_cmd;
974 int type = qctl->qc_type;
975 int id = qctl->qc_id;
976 int valid = qctl->qc_valid;
983 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
987 if (((type == USRQUOTA &&
988 !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
990 !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
991 (!cfs_capable(CFS_CAP_SYS_ADMIN)))
997 CERROR("unsupported quotactl op: %#x\n", cmd);
1001 if (valid != QC_GENERAL) {
1002 if (cmd == Q_GETINFO)
1003 qctl->qc_cmd = Q_GETOINFO;
1004 else if (cmd == Q_GETQUOTA)
1005 qctl->qc_cmd = Q_GETOQUOTA;
1011 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1012 sizeof(*qctl), qctl, NULL);
1015 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1016 sizeof(*qctl), qctl, NULL);
1019 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1020 sizeof(*qctl), qctl, NULL);
1022 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1024 sizeof(*qctl), qctl, NULL);
1036 struct obd_quotactl *oqctl;
1038 OBD_ALLOC_PTR(oqctl);
1042 QCTL_COPY(oqctl, qctl);
1043 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1045 OBD_FREE_PTR(oqctl);
1048 /* If QIF_SPACE is not set, client should collect the
1049 * space usage from OSSs by itself */
1050 if (cmd == Q_GETQUOTA &&
1051 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1052 !oqctl->qc_dqblk.dqb_curspace) {
1053 struct obd_quotactl *oqctl_tmp;
1055 OBD_ALLOC_PTR(oqctl_tmp);
1056 if (oqctl_tmp == NULL)
1057 GOTO(out, rc = -ENOMEM);
1059 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1060 oqctl_tmp->qc_id = oqctl->qc_id;
1061 oqctl_tmp->qc_type = oqctl->qc_type;
1063 /* collect space usage from OSTs */
1064 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1065 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1066 if (!rc || rc == -EREMOTEIO) {
1067 oqctl->qc_dqblk.dqb_curspace =
1068 oqctl_tmp->qc_dqblk.dqb_curspace;
1069 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1072 /* collect space & inode usage from MDTs */
1073 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1074 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1075 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1076 if (!rc || rc == -EREMOTEIO) {
1077 oqctl->qc_dqblk.dqb_curspace +=
1078 oqctl_tmp->qc_dqblk.dqb_curspace;
1079 oqctl->qc_dqblk.dqb_curinodes =
1080 oqctl_tmp->qc_dqblk.dqb_curinodes;
1081 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1083 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1086 OBD_FREE_PTR(oqctl_tmp);
1089 QCTL_COPY(qctl, oqctl);
1090 OBD_FREE_PTR(oqctl);
1096 /* This function tries to get a single name component,
1097 * to send to the server. No actual path traversal involved,
1098 * so we limit to NAME_MAX */
1099 static char *ll_getname(const char __user *filename)
1104 OBD_ALLOC(tmp, NAME_MAX + 1);
1107 return ERR_PTR(-ENOMEM);
1109 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1112 else if (len > NAME_MAX)
1113 ret = -ENAMETOOLONG;
1116 OBD_FREE(tmp, NAME_MAX + 1);
1122 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1124 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1126 struct dentry *dentry = file_dentry(file);
1127 struct inode *inode = file_inode(file);
1128 struct ll_sb_info *sbi = ll_i2sbi(inode);
1129 struct obd_ioctl_data *data;
1133 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1134 PFID(ll_inode2fid(inode)), inode, cmd);
1136 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1137 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1140 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1142 case FSFILT_IOC_GETFLAGS:
1143 case FSFILT_IOC_SETFLAGS:
1144 RETURN(ll_iocontrol(inode, file, cmd, arg));
1145 case FSFILT_IOC_GETVERSION_OLD:
1146 case FSFILT_IOC_GETVERSION:
1147 RETURN(put_user(inode->i_generation, (int __user *)arg));
1148 /* We need to special case any other ioctls we want to handle,
1149 * to send them to the MDS/OST as appropriate and to properly
1150 * network encode the arg field.
1151 case FSFILT_IOC_SETVERSION_OLD:
1152 case FSFILT_IOC_SETVERSION:
1154 case LL_IOC_GET_MDTIDX: {
1157 mdtidx = ll_get_mdt_idx(inode);
1161 if (put_user((int)mdtidx, (int __user *)arg))
1166 case IOC_MDC_LOOKUP: {
1167 int namelen, len = 0;
1171 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1176 filename = data->ioc_inlbuf1;
1177 namelen = strlen(filename);
1179 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1180 GOTO(out_free, rc = -EINVAL);
1183 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1185 CERROR("%s: lookup %.*s failed: rc = %d\n",
1186 ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1191 obd_ioctl_freedata(buf, len);
1194 case LL_IOC_LMV_SETSTRIPE: {
1195 struct lmv_user_md *lum;
1204 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1209 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1210 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1211 GOTO(lmv_out_free, rc = -EINVAL);
1213 filename = data->ioc_inlbuf1;
1214 namelen = data->ioc_inllen1;
1217 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1218 GOTO(lmv_out_free, rc = -EINVAL);
1220 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1221 lumlen = data->ioc_inllen2;
1223 if (lum->lum_magic != LMV_USER_MAGIC ||
1224 lumlen != sizeof(*lum)) {
1225 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1226 filename, lum->lum_magic, lumlen, -EFAULT);
1227 GOTO(lmv_out_free, rc = -EINVAL);
1230 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1231 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1233 mode = data->ioc_type;
1235 rc = ll_dir_setdirstripe(dentry, lum, filename, mode);
1237 obd_ioctl_freedata(buf, len);
1241 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1242 struct lmv_user_md lum;
1243 struct lmv_user_md __user *ulump =
1244 (struct lmv_user_md __user *)arg;
1247 if (copy_from_user(&lum, ulump, sizeof(lum)))
1250 if (lum.lum_magic != LMV_USER_MAGIC)
1253 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1257 case LL_IOC_LOV_SETSTRIPE: {
1258 struct lov_user_md_v3 lumv3;
1259 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1260 struct lov_user_md_v1 __user *lumv1p =
1261 (struct lov_user_md_v1 __user *)arg;
1262 struct lov_user_md_v3 __user *lumv3p =
1263 (struct lov_user_md_v3 __user *)arg;
1265 int set_default = 0;
1267 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1268 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1269 sizeof(lumv3p->lmm_objects[0]));
1270 /* first try with v1 which is smaller than v3 */
1271 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1274 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1275 if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1279 if (inode->i_sb->s_root == file_dentry(file))
1282 /* in v1 and v3 cases lumv1 points to data */
1283 rc = ll_dir_setstripe(inode, lumv1, set_default);
1287 case LL_IOC_LMV_GETSTRIPE: {
1288 struct lmv_user_md __user *ulmv =
1289 (struct lmv_user_md __user *)arg;
1290 struct lmv_user_md lum;
1291 struct ptlrpc_request *request = NULL;
1292 union lmv_mds_md *lmm = NULL;
1295 struct lmv_user_md *tmp = NULL;
1299 int max_stripe_count;
1303 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1306 max_stripe_count = lum.lum_stripe_count;
1307 /* lum_magic will indicate which stripe the ioctl will like
1308 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1309 * is for default LMV stripe */
1310 if (lum.lum_magic == LMV_MAGIC_V1)
1311 valid |= OBD_MD_MEA;
1312 else if (lum.lum_magic == LMV_USER_MAGIC)
1313 valid |= OBD_MD_DEFAULT_MEA;
1317 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1320 GOTO(finish_req, rc);
1322 /* Get default LMV EA */
1323 if (lum.lum_magic == LMV_USER_MAGIC) {
1324 if (lmmsize > sizeof(*ulmv))
1325 GOTO(finish_req, rc = -EINVAL);
1327 if (copy_to_user(ulmv, lmm, lmmsize))
1328 GOTO(finish_req, rc = -EFAULT);
1330 GOTO(finish_req, rc);
1333 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1334 if (max_stripe_count < stripe_count) {
1335 lum.lum_stripe_count = stripe_count;
1336 if (copy_to_user(ulmv, &lum, sizeof(lum)))
1337 GOTO(finish_req, rc = -EFAULT);
1338 GOTO(finish_req, rc = -E2BIG);
1341 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1342 OBD_ALLOC(tmp, lum_size);
1344 GOTO(finish_req, rc = -ENOMEM);
1346 mdt_index = ll_get_mdt_idx(inode);
1348 GOTO(out_tmp, rc = -ENOMEM);
1350 tmp->lum_magic = LMV_MAGIC_V1;
1351 tmp->lum_stripe_count = 0;
1352 tmp->lum_stripe_offset = mdt_index;
1353 for (i = 0; i < stripe_count; i++) {
1356 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1357 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1359 GOTO(out_tmp, rc = mdt_index);
1361 tmp->lum_objects[i].lum_mds = mdt_index;
1362 tmp->lum_objects[i].lum_fid = fid;
1363 tmp->lum_stripe_count++;
1366 if (copy_to_user(ulmv, tmp, lum_size))
1367 GOTO(out_tmp, rc = -EFAULT);
1369 OBD_FREE(tmp, lum_size);
1371 ptlrpc_req_finished(request);
1375 case LL_IOC_REMOVE_ENTRY: {
1376 char *filename = NULL;
1380 /* Here is a little hack to avoid sending REINT_RMENTRY to
1381 * unsupported server, which might crash the server(LU-2730),
1382 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1383 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1384 * server will support REINT_RMENTRY XXX*/
1385 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1388 filename = ll_getname((const char __user *)arg);
1389 if (IS_ERR(filename))
1390 RETURN(PTR_ERR(filename));
1392 namelen = strlen(filename);
1394 GOTO(out_rmdir, rc = -EINVAL);
1396 rc = ll_rmdir_entry(inode, filename, namelen);
1399 ll_putname(filename);
1402 case LL_IOC_LOV_SWAP_LAYOUTS:
1404 case IOC_OBD_STATFS:
1405 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1406 case LL_IOC_LOV_GETSTRIPE:
1407 case LL_IOC_MDC_GETINFO:
1408 case IOC_MDC_GETFILEINFO:
1409 case IOC_MDC_GETFILESTRIPE: {
1410 struct ptlrpc_request *request = NULL;
1411 struct lov_user_md __user *lump;
1412 struct lov_mds_md *lmm = NULL;
1413 struct mdt_body *body;
1414 char *filename = NULL;
1417 if (cmd == IOC_MDC_GETFILEINFO ||
1418 cmd == IOC_MDC_GETFILESTRIPE) {
1419 filename = ll_getname((const char __user *)arg);
1420 if (IS_ERR(filename))
1421 RETURN(PTR_ERR(filename));
1423 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1424 &lmmsize, &request);
1426 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1431 body = req_capsule_server_get(&request->rq_pill,
1433 LASSERT(body != NULL);
1439 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1440 cmd == LL_IOC_MDC_GETINFO))
1441 GOTO(skip_lmm, rc = 0);
1446 if (cmd == IOC_MDC_GETFILESTRIPE ||
1447 cmd == LL_IOC_LOV_GETSTRIPE) {
1448 lump = (struct lov_user_md __user *)arg;
1450 struct lov_user_mds_data __user *lmdp;
1451 lmdp = (struct lov_user_mds_data __user *)arg;
1452 lump = &lmdp->lmd_lmm;
1454 if (copy_to_user(lump, lmm, lmmsize)) {
1455 if (copy_to_user(lump, lmm, sizeof(*lump)))
1456 GOTO(out_req, rc = -EFAULT);
1460 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1461 struct lov_user_mds_data __user *lmdp;
1464 st.st_dev = inode->i_sb->s_dev;
1465 st.st_mode = body->mbo_mode;
1466 st.st_nlink = body->mbo_nlink;
1467 st.st_uid = body->mbo_uid;
1468 st.st_gid = body->mbo_gid;
1469 st.st_rdev = body->mbo_rdev;
1470 st.st_size = body->mbo_size;
1471 st.st_blksize = PAGE_SIZE;
1472 st.st_blocks = body->mbo_blocks;
1473 st.st_atime = body->mbo_atime;
1474 st.st_mtime = body->mbo_mtime;
1475 st.st_ctime = body->mbo_ctime;
1476 st.st_ino = cl_fid_build_ino(&body->mbo_fid1,
1480 lmdp = (struct lov_user_mds_data __user *)arg;
1481 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1482 GOTO(out_req, rc = -EFAULT);
1487 ptlrpc_req_finished(request);
1489 ll_putname(filename);
1492 case OBD_IOC_QUOTACTL: {
1493 struct if_quotactl *qctl;
1495 OBD_ALLOC_PTR(qctl);
1499 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1500 GOTO(out_quotactl, rc = -EFAULT);
1502 rc = quotactl_ioctl(sbi, qctl);
1505 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1512 case OBD_IOC_GETDTNAME:
1513 case OBD_IOC_GETMDNAME:
1514 RETURN(ll_get_obd_name(inode, cmd, arg));
1515 case LL_IOC_FLUSHCTX:
1516 RETURN(ll_flush_ctx(inode));
1517 case LL_IOC_GETOBDCOUNT: {
1519 struct obd_export *exp;
1521 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1524 /* get ost count when count is zero, get mdt count otherwise */
1525 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1526 vallen = sizeof(count);
1527 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1528 KEY_TGT_COUNT, &vallen, &count);
1530 CERROR("get target count failed: %d\n", rc);
1534 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1539 case LL_IOC_PATH2FID:
1540 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1541 sizeof(struct lu_fid)))
1544 case LL_IOC_GET_CONNECT_FLAGS: {
1545 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1546 (void __user *)arg));
1548 case OBD_IOC_CHANGELOG_SEND:
1549 case OBD_IOC_CHANGELOG_CLEAR:
1550 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1553 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1554 sizeof(struct ioc_changelog));
1556 case OBD_IOC_FID2PATH:
1557 RETURN(ll_fid2path(inode, (void __user *)arg));
1558 case LL_IOC_GETPARENT:
1559 RETURN(ll_getparent(file, (void __user *)arg));
1560 case LL_IOC_FID2MDTIDX: {
1561 struct obd_export *exp = ll_i2mdexp(inode);
1565 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1569 /* Call mdc_iocontrol */
1570 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1571 (__u32 __user *)&index);
1577 case LL_IOC_HSM_REQUEST: {
1578 struct hsm_user_request *hur;
1585 /* We don't know the true size yet; copy the fixed-size part */
1586 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1591 /* Compute the whole struct size */
1592 totalsize = hur_len(hur);
1597 /* Final size will be more than double totalsize */
1598 if (totalsize >= MDS_MAXREQSIZE / 3)
1601 OBD_ALLOC_LARGE(hur, totalsize);
1605 /* Copy the whole struct */
1606 if (copy_from_user(hur, (void __user *)arg, totalsize))
1607 GOTO(out_hur, rc = -EFAULT);
1609 if (hur->hur_request.hr_action == HUA_RELEASE) {
1610 const struct lu_fid *fid;
1614 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1615 fid = &hur->hur_user_item[i].hui_fid;
1616 f = search_inode_for_lustre(inode->i_sb, fid);
1622 rc = ll_hsm_release(f);
1628 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1633 OBD_FREE_LARGE(hur, totalsize);
1637 case LL_IOC_HSM_PROGRESS: {
1638 struct hsm_progress_kernel hpk;
1639 struct hsm_progress hp;
1641 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1644 hpk.hpk_fid = hp.hp_fid;
1645 hpk.hpk_cookie = hp.hp_cookie;
1646 hpk.hpk_extent = hp.hp_extent;
1647 hpk.hpk_flags = hp.hp_flags;
1648 hpk.hpk_errval = hp.hp_errval;
1649 hpk.hpk_data_version = 0;
1651 /* File may not exist in Lustre; all progress
1652 * reported to Lustre root */
1653 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1657 case LL_IOC_HSM_CT_START:
1658 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1661 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1662 sizeof(struct lustre_kernelcomm));
1665 case LL_IOC_HSM_COPY_START: {
1666 struct hsm_copy *copy;
1669 OBD_ALLOC_PTR(copy);
1672 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1677 rc = ll_ioc_copy_start(inode->i_sb, copy);
1678 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1684 case LL_IOC_HSM_COPY_END: {
1685 struct hsm_copy *copy;
1688 OBD_ALLOC_PTR(copy);
1691 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1696 rc = ll_ioc_copy_end(inode->i_sb, copy);
1697 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1703 case LL_IOC_MIGRATE: {
1705 const char *filename;
1711 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1715 data = (struct obd_ioctl_data *)buf;
1716 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1717 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1718 GOTO(migrate_free, rc = -EINVAL);
1720 filename = data->ioc_inlbuf1;
1721 namelen = data->ioc_inllen1;
1722 /* \0 is packed at the end of filename */
1723 if (namelen < 1 || namelen != strlen(filename) + 1)
1724 GOTO(migrate_free, rc = -EINVAL);
1726 if (data->ioc_inllen2 != sizeof(mdtidx))
1727 GOTO(migrate_free, rc = -EINVAL);
1728 mdtidx = *(int *)data->ioc_inlbuf2;
1730 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1732 obd_ioctl_freedata(buf, len);
1737 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1738 (void __user *)arg));
1742 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1744 struct inode *inode = file->f_mapping->host;
1745 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1746 struct ll_sb_info *sbi = ll_i2sbi(inode);
1747 int api32 = ll_need_32bit_api(sbi);
1748 loff_t ret = -EINVAL;
1756 offset += file->f_pos;
1762 offset += LL_DIR_END_OFF_32BIT;
1764 offset += LL_DIR_END_OFF;
1771 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1772 (!api32 && offset <= LL_DIR_END_OFF))) {
1773 if (offset != file->f_pos) {
1774 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1775 (!api32 && offset == LL_DIR_END_OFF))
1776 fd->lfd_pos = MDS_DIR_END_OFF;
1777 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1778 fd->lfd_pos = offset << 32;
1780 fd->lfd_pos = offset;
1781 file->f_pos = offset;
1782 file->f_version = 0;
1789 inode_unlock(inode);
1793 static int ll_dir_open(struct inode *inode, struct file *file)
1796 RETURN(ll_file_open(inode, file));
1799 static int ll_dir_release(struct inode *inode, struct file *file)
1802 RETURN(ll_file_release(inode, file));
1805 const struct file_operations ll_dir_operations = {
1806 .llseek = ll_dir_seek,
1807 .open = ll_dir_open,
1808 .release = ll_dir_release,
1809 .read = generic_read_dir,
1810 #ifdef HAVE_DIR_CONTEXT
1811 .iterate = ll_iterate,
1813 .readdir = ll_readdir,
1815 .unlocked_ioctl = ll_dir_ioctl,