4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Directory code for lustre client.
42 #include <linux/pagemap.h>
44 #include <linux/version.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>
52 #include <obd_support.h>
53 #include <obd_class.h>
54 #include <lustre_ioctl.h>
55 #include <lustre_lib.h>
56 #include <lustre_dlm.h>
57 #include <lustre_fid.h>
58 #include <lustre_kernelcomm.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_CACHE_SIZE (if PAGE_CACHE_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 != NULL && S_ISDIR(inode->i_mode) &&
165 ll_i2info(inode)->lli_lsm_md != NULL) {
172 if (likely(page->mapping != NULL))
173 truncate_complete_page(page->mapping, page);
176 page_cache_release(page);
180 * return IF_* type for given lu_dirent entry.
181 * IF_* flag shld be converted to particular OS file type in
182 * platform llite module.
184 static u16 ll_dirent_type_get(struct lu_dirent *ent)
187 struct luda_type *lt;
190 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
191 const unsigned align = sizeof(struct luda_type) - 1;
193 len = le16_to_cpu(ent->lde_namelen);
194 len = (len + align) & ~align;
195 lt = (void *)ent->lde_name + len;
196 type = IFTODT(le16_to_cpu(lt->lt_type));
202 #ifdef HAVE_DIR_CONTEXT
203 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
204 struct dir_context *ctx)
207 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
208 void *cookie, filldir_t filldir)
211 struct ll_sb_info *sbi = ll_i2sbi(inode);
213 bool is_api32 = ll_need_32bit_api(sbi);
214 bool is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
216 struct ll_dir_chain chain;
221 ll_dir_chain_init(&chain);
223 page = ll_get_dir_page(inode, op_data, pos, &chain);
225 while (rc == 0 && !done) {
226 struct lu_dirpage *dp;
227 struct lu_dirent *ent;
236 hash = MDS_DIR_END_OFF;
237 dp = page_address(page);
238 for (ent = lu_dirent_start(dp); ent != NULL && !done;
239 ent = lu_dirent_next(ent)) {
246 hash = le64_to_cpu(ent->lde_hash);
249 * Skip until we find target hash
254 namelen = le16_to_cpu(ent->lde_namelen);
261 if (is_api32 && is_hash64)
265 fid_le_to_cpu(&fid, &ent->lde_fid);
266 ino = cl_fid_build_ino(&fid, is_api32);
267 type = ll_dirent_type_get(ent);
268 /* For 'll_nfs_get_name_filldir()', it will try
269 * to access the 'ent' through its 'lde_name',
270 * so the parameter 'name' for 'filldir()' must
271 * be part of the 'ent'. */
272 #ifdef HAVE_DIR_CONTEXT
274 done = !dir_emit(ctx, ent->lde_name, namelen, ino,
277 done = filldir(cookie, ent->lde_name, namelen, lhash,
284 ll_release_page(inode, page, false);
288 next = le64_to_cpu(dp->ldp_hash_end);
290 if (pos == MDS_DIR_END_OFF) {
292 * End of directory reached.
295 ll_release_page(inode, page, false);
298 * Normal case: continue to the next
301 ll_release_page(inode, page,
302 le32_to_cpu(dp->ldp_flags) &
305 page = ll_get_dir_page(inode, op_data, pos,
309 #ifdef HAVE_DIR_CONTEXT
314 ll_dir_chain_fini(&chain);
318 #ifdef HAVE_DIR_CONTEXT
319 static int ll_iterate(struct file *filp, struct dir_context *ctx)
321 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
324 struct inode *inode = filp->f_dentry->d_inode;
325 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
326 struct ll_sb_info *sbi = ll_i2sbi(inode);
327 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
328 int api32 = ll_need_32bit_api(sbi);
329 struct md_op_data *op_data;
339 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
340 "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
341 inode, (unsigned long)pos, i_size_read(inode), api32);
343 if (pos == MDS_DIR_END_OFF)
349 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
350 LUSTRE_OPC_ANY, inode);
352 GOTO(out, rc = PTR_ERR(op_data));
354 if (unlikely(op_data->op_mea1 != NULL)) {
355 /* This is only needed for striped dir to fill ..,
356 * see lmv_read_entry */
357 if (filp->f_dentry->d_parent != NULL &&
358 filp->f_dentry->d_parent->d_inode != NULL) {
359 __u64 ibits = MDS_INODELOCK_UPDATE;
360 struct inode *parent =
361 filp->f_dentry->d_parent->d_inode;
363 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
364 op_data->op_fid3 = *ll_inode2fid(parent);
367 /* If it can not find in cache, do lookup .. on the master
369 if (fid_is_zero(&op_data->op_fid3)) {
370 rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
372 ll_finish_md_op_data(op_data);
377 op_data->op_max_pages = sbi->ll_md_brw_pages;
378 #ifdef HAVE_DIR_CONTEXT
380 rc = ll_dir_read(inode, &pos, op_data, ctx);
383 rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
388 if (pos == MDS_DIR_END_OFF) {
390 pos = LL_DIR_END_OFF_32BIT;
392 pos = LL_DIR_END_OFF;
397 #ifdef HAVE_DIR_CONTEXT
402 ll_finish_md_op_data(op_data);
403 filp->f_version = inode->i_version;
407 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
412 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
414 struct mgs_send_param *msp;
421 strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
422 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
423 sizeof(struct mgs_send_param), msp, NULL);
425 CERROR("Failed to set parameter: %d\n", rc);
432 * Create striped directory with specified stripe(@lump)
434 * param[in]parent the parent of the directory.
435 * param[in]lump the specified stripes.
436 * param[in]dirname the name of the directory.
437 * param[in]mode the specified mode of the directory.
439 * retval =0 if striped directory is being created successfully.
440 * <0 if the creation is failed.
442 static int ll_dir_setdirstripe(struct inode *parent, struct lmv_user_md *lump,
443 const char *dirname, umode_t mode)
445 struct ptlrpc_request *request = NULL;
446 struct md_op_data *op_data;
447 struct ll_sb_info *sbi = ll_i2sbi(parent);
451 if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
454 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
455 "stripe_offset %d, stripe_count: %u\n",
456 PFID(ll_inode2fid(parent)), parent, dirname,
457 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
459 if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
460 lustre_swab_lmv_user_md(lump);
462 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
463 mode &= ~current_umask();
464 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
465 op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
466 strlen(dirname), mode, LUSTRE_OPC_MKDIR,
469 GOTO(err_exit, err = PTR_ERR(op_data));
471 op_data->op_cli_flags |= CLI_SET_MEA;
472 err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
473 from_kuid(&init_user_ns, current_fsuid()),
474 from_kgid(&init_user_ns, current_fsgid()),
475 cfs_curproc_cap_pack(), 0, &request);
476 ll_finish_md_op_data(op_data);
480 ptlrpc_req_finished(request);
484 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
487 struct ll_sb_info *sbi = ll_i2sbi(inode);
488 struct md_op_data *op_data;
489 struct ptlrpc_request *req = NULL;
491 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
492 struct obd_device *mgc = lsi->lsi_mgc;
498 * This is coming from userspace, so should be in
499 * local endian. But the MDS would like it in little
500 * endian, so we swab it before we send it.
502 switch (lump->lmm_magic) {
503 case LOV_USER_MAGIC_V1: {
504 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
505 lustre_swab_lov_user_md_v1(lump);
506 lum_size = sizeof(struct lov_user_md_v1);
509 case LOV_USER_MAGIC_V3: {
510 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
511 lustre_swab_lov_user_md_v3(
512 (struct lov_user_md_v3 *)lump);
513 lum_size = sizeof(struct lov_user_md_v3);
516 case LMV_USER_MAGIC: {
517 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
518 lustre_swab_lmv_user_md(
519 (struct lmv_user_md *)lump);
520 lum_size = sizeof(struct lmv_user_md);
524 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
525 " %#08x != %#08x nor %#08x\n",
526 lump->lmm_magic, LOV_USER_MAGIC_V1,
532 lum_size = sizeof(struct lov_user_md_v1);
535 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
536 LUSTRE_OPC_ANY, NULL);
538 RETURN(PTR_ERR(op_data));
540 /* swabbing is done in lov_setstripe() on server side */
541 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
542 ll_finish_md_op_data(op_data);
543 ptlrpc_req_finished(req);
545 if (rc != -EPERM && rc != -EACCES)
546 CERROR("mdc_setattr fails: rc = %d\n", rc);
549 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
550 LOV_USER_MAGIC_V3 have the same initial fields so we do not
551 need the make the distiction between the 2 versions */
552 if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
556 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
558 GOTO(end, rc = -ENOMEM);
561 /* Get fsname and assume devname to be -MDT0000. */
562 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
563 strcat(buf, "-MDT0000.lov");
566 /* Set root stripesize */
567 sprintf(buf, ".stripesize=%u",
568 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
569 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
573 /* Set root stripecount */
574 sprintf(buf, ".stripecount=%hd",
575 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
576 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
580 /* Set root stripeoffset */
581 sprintf(buf, ".stripeoffset=%hd",
582 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
583 (typeof(lump->lmm_stripe_offset))(-1));
584 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
588 OBD_FREE(param, MGS_PARAM_MAXLEN);
594 * This function will be used to get default LOV/LMV/Default LMV
595 * @valid will be used to indicate which stripe it will retrieve
596 * OBD_MD_MEA LMV stripe EA
597 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
598 * otherwise Default LOV EA.
599 * Each time, it can only retrieve 1 stripe EA
601 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
602 struct ptlrpc_request **request, u64 valid)
604 struct ll_sb_info *sbi = ll_i2sbi(inode);
605 struct mdt_body *body;
606 struct lov_mds_md *lmm = NULL;
607 struct ptlrpc_request *req = NULL;
609 struct md_op_data *op_data;
612 rc = ll_get_default_mdsize(sbi, &lmm_size);
616 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
617 0, lmm_size, LUSTRE_OPC_ANY,
620 RETURN(PTR_ERR(op_data));
622 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
623 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
624 ll_finish_md_op_data(op_data);
626 CDEBUG(D_INFO, "md_getattr failed on inode "
627 DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
631 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
632 LASSERT(body != NULL);
634 lmm_size = body->mbo_eadatasize;
636 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
638 GOTO(out, rc = -ENODATA);
641 lmm = req_capsule_server_sized_get(&req->rq_pill,
642 &RMF_MDT_MD, lmm_size);
643 LASSERT(lmm != NULL);
646 * This is coming from the MDS, so is probably in
647 * little endian. We convert it to host endian before
648 * passing it to userspace.
650 /* We don't swab objects for directories */
651 switch (le32_to_cpu(lmm->lmm_magic)) {
653 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
654 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
657 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
658 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
661 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
662 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
665 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
666 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
669 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
674 *plmm_size = lmm_size;
679 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
681 struct md_op_data *op_data;
686 OBD_ALLOC_PTR(op_data);
690 op_data->op_flags |= MF_GET_MDT_IDX;
691 op_data->op_fid1 = *fid;
692 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
693 mdt_index = op_data->op_mds;
694 OBD_FREE_PTR(op_data);
702 * Get MDT index for the inode.
704 int ll_get_mdt_idx(struct inode *inode)
706 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
710 * Generic handler to do any pre-copy work.
712 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
713 * first information for it that real work has started.
715 * Moreover, for a ARCHIVE request, it will sample the file data version and
716 * store it in \a copy.
718 * \return 0 on success.
720 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
722 struct ll_sb_info *sbi = ll_s2sbi(sb);
723 struct hsm_progress_kernel hpk;
727 /* Forge a hsm_progress based on data from copy. */
728 hpk.hpk_fid = copy->hc_hai.hai_fid;
729 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
730 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
731 hpk.hpk_extent.length = 0;
734 hpk.hpk_data_version = 0;
737 /* For archive request, we need to read the current file version. */
738 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
740 __u64 data_version = 0;
742 /* Get inode for this fid */
743 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
745 hpk.hpk_flags |= HP_FLAG_RETRY;
746 /* hpk_errval is >= 0 */
747 hpk.hpk_errval = -PTR_ERR(inode);
748 GOTO(progress, rc = PTR_ERR(inode));
751 /* Read current file data version */
752 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
755 CDEBUG(D_HSM, "Could not read file data version of "
756 DFID" (rc = %d). Archive request ("
757 LPX64") could not be done.\n",
758 PFID(©->hc_hai.hai_fid), rc,
759 copy->hc_hai.hai_cookie);
760 hpk.hpk_flags |= HP_FLAG_RETRY;
761 /* hpk_errval must be >= 0 */
762 hpk.hpk_errval = -rc;
766 /* Store in the hsm_copy for later copytool use.
767 * Always modified even if no lsm. */
768 copy->hc_data_version = data_version;
772 /* On error, the request should be considered as completed */
773 if (hpk.hpk_errval > 0)
774 hpk.hpk_flags |= HP_FLAG_COMPLETED;
775 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
782 * Generic handler to do any post-copy work.
784 * It will send the last hsm_progress update to coordinator to inform it
785 * that copy is finished and whether it was successful or not.
788 * - for ARCHIVE request, it will sample the file data version and compare it
789 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
790 * will be considered as failed.
791 * - for RESTORE request, it will sample the file data version and send it to
792 * coordinator which is useful if the file was imported as 'released'.
794 * \return 0 on success.
796 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
798 struct ll_sb_info *sbi = ll_s2sbi(sb);
799 struct hsm_progress_kernel hpk;
803 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
804 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
805 * initialized if copy_end was called with copy == NULL.
808 /* Forge a hsm_progress based on data from copy. */
809 hpk.hpk_fid = copy->hc_hai.hai_fid;
810 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
811 hpk.hpk_extent = copy->hc_hai.hai_extent;
812 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
813 hpk.hpk_errval = copy->hc_errval;
814 hpk.hpk_data_version = 0;
816 /* For archive request, we need to check the file data was not changed.
818 * For restore request, we need to send the file data version, this is
819 * useful when the file was created using hsm_import.
821 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
822 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
823 (copy->hc_errval == 0)) {
825 __u64 data_version = 0;
827 /* Get lsm for this fid */
828 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
830 hpk.hpk_flags |= HP_FLAG_RETRY;
831 /* hpk_errval must be >= 0 */
832 hpk.hpk_errval = -PTR_ERR(inode);
833 GOTO(progress, rc = PTR_ERR(inode));
836 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
839 CDEBUG(D_HSM, "Could not read file data version. "
840 "Request could not be confirmed.\n");
841 if (hpk.hpk_errval == 0)
842 hpk.hpk_errval = -rc;
846 /* Store in the hsm_copy for later copytool use.
847 * Always modified even if no lsm. */
848 hpk.hpk_data_version = data_version;
850 /* File could have been stripped during archiving, so we need
851 * to check anyway. */
852 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
853 (copy->hc_data_version != data_version)) {
854 CDEBUG(D_HSM, "File data version mismatched. "
855 "File content was changed during archiving. "
856 DFID", start:"LPX64" current:"LPX64"\n",
857 PFID(©->hc_hai.hai_fid),
858 copy->hc_data_version, data_version);
859 /* File was changed, send error to cdt. Do not ask for
860 * retry because if a file is modified frequently,
861 * the cdt will loop on retried archive requests.
862 * The policy engine will ask for a new archive later
863 * when the file will not be modified for some tunable
865 /* we do not notify caller */
866 hpk.hpk_flags &= ~HP_FLAG_RETRY;
867 /* hpk_errval must be >= 0 */
868 hpk.hpk_errval = EBUSY;
874 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
881 static int copy_and_ioctl(int cmd, struct obd_export *exp,
882 const void __user *data, size_t size)
887 OBD_ALLOC(copy, size);
891 if (copy_from_user(copy, data, size)) {
896 rc = obd_iocontrol(cmd, exp, size, copy, NULL);
898 OBD_FREE(copy, size);
903 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
905 int cmd = qctl->qc_cmd;
906 int type = qctl->qc_type;
907 int id = qctl->qc_id;
908 int valid = qctl->qc_valid;
913 case LUSTRE_Q_INVALIDATE:
914 case LUSTRE_Q_FINVALIDATE:
919 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
920 sbi->ll_flags & LL_SBI_RMT_CLIENT)
924 if (((type == USRQUOTA &&
925 !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
927 !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
928 (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
929 sbi->ll_flags & LL_SBI_RMT_CLIENT))
935 CERROR("unsupported quotactl op: %#x\n", cmd);
939 if (valid != QC_GENERAL) {
940 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
943 if (cmd == Q_GETINFO)
944 qctl->qc_cmd = Q_GETOINFO;
945 else if (cmd == Q_GETQUOTA)
946 qctl->qc_cmd = Q_GETOQUOTA;
952 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
953 sizeof(*qctl), qctl, NULL);
956 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
957 sizeof(*qctl), qctl, NULL);
960 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
961 sizeof(*qctl), qctl, NULL);
963 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
965 sizeof(*qctl), qctl, NULL);
977 struct obd_quotactl *oqctl;
979 OBD_ALLOC_PTR(oqctl);
983 QCTL_COPY(oqctl, qctl);
984 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
986 if (rc != -EALREADY && cmd == Q_QUOTAON) {
987 oqctl->qc_cmd = Q_QUOTAOFF;
988 obd_quotactl(sbi->ll_md_exp, oqctl);
993 /* If QIF_SPACE is not set, client should collect the
994 * space usage from OSSs by itself */
995 if (cmd == Q_GETQUOTA &&
996 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
997 !oqctl->qc_dqblk.dqb_curspace) {
998 struct obd_quotactl *oqctl_tmp;
1000 OBD_ALLOC_PTR(oqctl_tmp);
1001 if (oqctl_tmp == NULL)
1002 GOTO(out, rc = -ENOMEM);
1004 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1005 oqctl_tmp->qc_id = oqctl->qc_id;
1006 oqctl_tmp->qc_type = oqctl->qc_type;
1008 /* collect space usage from OSTs */
1009 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1010 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1011 if (!rc || rc == -EREMOTEIO) {
1012 oqctl->qc_dqblk.dqb_curspace =
1013 oqctl_tmp->qc_dqblk.dqb_curspace;
1014 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1017 /* collect space & inode usage from MDTs */
1018 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1019 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1020 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1021 if (!rc || rc == -EREMOTEIO) {
1022 oqctl->qc_dqblk.dqb_curspace +=
1023 oqctl_tmp->qc_dqblk.dqb_curspace;
1024 oqctl->qc_dqblk.dqb_curinodes =
1025 oqctl_tmp->qc_dqblk.dqb_curinodes;
1026 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1028 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1031 OBD_FREE_PTR(oqctl_tmp);
1034 QCTL_COPY(qctl, oqctl);
1035 OBD_FREE_PTR(oqctl);
1042 ll_getname(const char __user *filename)
1045 char *tmp = __getname();
1048 return ERR_PTR(-ENOMEM);
1050 len = strncpy_from_user(tmp, filename, PATH_MAX);
1053 else if (len > PATH_MAX)
1054 ret = -ENAMETOOLONG;
1063 #define ll_putname(filename) __putname(filename)
1065 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1067 struct inode *inode = file->f_dentry->d_inode;
1068 struct ll_sb_info *sbi = ll_i2sbi(inode);
1069 struct obd_ioctl_data *data;
1073 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1074 PFID(ll_inode2fid(inode)), inode, cmd);
1076 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1077 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1080 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1082 case FSFILT_IOC_GETFLAGS:
1083 case FSFILT_IOC_SETFLAGS:
1084 RETURN(ll_iocontrol(inode, file, cmd, arg));
1085 case FSFILT_IOC_GETVERSION_OLD:
1086 case FSFILT_IOC_GETVERSION:
1087 RETURN(put_user(inode->i_generation, (int __user *)arg));
1088 /* We need to special case any other ioctls we want to handle,
1089 * to send them to the MDS/OST as appropriate and to properly
1090 * network encode the arg field.
1091 case FSFILT_IOC_SETVERSION_OLD:
1092 case FSFILT_IOC_SETVERSION:
1094 case LL_IOC_GET_MDTIDX: {
1097 mdtidx = ll_get_mdt_idx(inode);
1101 if (put_user((int)mdtidx, (int __user *)arg))
1106 case IOC_MDC_LOOKUP: {
1107 int namelen, len = 0;
1111 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1116 filename = data->ioc_inlbuf1;
1117 namelen = strlen(filename);
1119 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1120 GOTO(out_free, rc = -EINVAL);
1123 rc = ll_get_fid_by_name(inode, filename, namelen, NULL);
1125 CERROR("%s: lookup %.*s failed: rc = %d\n",
1126 ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1131 obd_ioctl_freedata(buf, len);
1134 case LL_IOC_LMV_SETSTRIPE: {
1135 struct lmv_user_md *lum;
1144 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1149 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1150 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1151 GOTO(lmv_out_free, rc = -EINVAL);
1153 filename = data->ioc_inlbuf1;
1154 namelen = data->ioc_inllen1;
1157 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1158 GOTO(lmv_out_free, rc = -EINVAL);
1160 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1161 lumlen = data->ioc_inllen2;
1163 if (lum->lum_magic != LMV_USER_MAGIC ||
1164 lumlen != sizeof(*lum)) {
1165 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1166 filename, lum->lum_magic, lumlen, -EFAULT);
1167 GOTO(lmv_out_free, rc = -EINVAL);
1170 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1171 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1173 mode = data->ioc_type;
1175 rc = ll_dir_setdirstripe(inode, lum, filename, mode);
1177 obd_ioctl_freedata(buf, len);
1181 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1182 struct lmv_user_md lum;
1183 struct lmv_user_md __user *ulump =
1184 (struct lmv_user_md __user *)arg;
1187 if (copy_from_user(&lum, ulump, sizeof(lum)))
1190 if (lum.lum_magic != LMV_USER_MAGIC)
1193 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1197 case LL_IOC_LOV_SETSTRIPE: {
1198 struct lov_user_md_v3 lumv3;
1199 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1200 struct lov_user_md_v1 __user *lumv1p =
1201 (struct lov_user_md_v1 __user *)arg;
1202 struct lov_user_md_v3 __user *lumv3p =
1203 (struct lov_user_md_v3 __user *)arg;
1205 int set_default = 0;
1207 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1208 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1209 sizeof(lumv3p->lmm_objects[0]));
1210 /* first try with v1 which is smaller than v3 */
1211 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1214 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1215 if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1219 if (inode->i_sb->s_root == file->f_dentry)
1222 /* in v1 and v3 cases lumv1 points to data */
1223 rc = ll_dir_setstripe(inode, lumv1, set_default);
1227 case LL_IOC_LMV_GETSTRIPE: {
1228 struct lmv_user_md __user *ulmv =
1229 (struct lmv_user_md __user *)arg;
1230 struct lmv_user_md lum;
1231 struct ptlrpc_request *request = NULL;
1232 union lmv_mds_md *lmm = NULL;
1235 struct lmv_user_md *tmp = NULL;
1242 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1245 /* lum_magic will indicate which stripe the ioctl will like
1246 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1247 * is for default LMV stripe */
1248 if (lum.lum_magic == LMV_MAGIC_V1)
1249 valid |= OBD_MD_MEA;
1250 else if (lum.lum_magic == LMV_USER_MAGIC)
1251 valid |= OBD_MD_DEFAULT_MEA;
1255 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1258 GOTO(finish_req, rc);
1260 /* Get default LMV EA */
1261 if (lum.lum_magic == LMV_USER_MAGIC) {
1263 GOTO(finish_req, rc);
1265 if (lmmsize > sizeof(*ulmv))
1266 GOTO(finish_req, rc = -EINVAL);
1268 if (copy_to_user(ulmv, lmm, lmmsize))
1269 GOTO(finish_req, rc = -EFAULT);
1271 GOTO(finish_req, rc);
1274 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1275 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1276 OBD_ALLOC(tmp, lum_size);
1278 GOTO(finish_req, rc = -ENOMEM);
1280 mdt_index = ll_get_mdt_idx(inode);
1282 GOTO(out_tmp, rc = -ENOMEM);
1284 tmp->lum_magic = LMV_MAGIC_V1;
1285 tmp->lum_stripe_count = 0;
1286 tmp->lum_stripe_offset = mdt_index;
1287 for (i = 0; i < stripe_count; i++) {
1290 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1291 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1293 GOTO(out_tmp, rc = mdt_index);
1295 tmp->lum_objects[i].lum_mds = mdt_index;
1296 tmp->lum_objects[i].lum_fid = fid;
1297 tmp->lum_stripe_count++;
1300 if (copy_to_user(ulmv, tmp, lum_size))
1301 GOTO(out_tmp, rc = -EFAULT);
1303 OBD_FREE(tmp, lum_size);
1305 ptlrpc_req_finished(request);
1309 case LL_IOC_REMOVE_ENTRY: {
1310 char *filename = NULL;
1314 /* Here is a little hack to avoid sending REINT_RMENTRY to
1315 * unsupported server, which might crash the server(LU-2730),
1316 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1317 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1318 * server will support REINT_RMENTRY XXX*/
1319 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1322 filename = ll_getname((const char __user *)arg);
1323 if (IS_ERR(filename))
1324 RETURN(PTR_ERR(filename));
1326 namelen = strlen(filename);
1328 GOTO(out_rmdir, rc = -EINVAL);
1330 rc = ll_rmdir_entry(inode, filename, namelen);
1333 ll_putname(filename);
1336 case LL_IOC_LOV_SWAP_LAYOUTS:
1338 case IOC_OBD_STATFS:
1339 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1340 case LL_IOC_LOV_GETSTRIPE:
1341 case LL_IOC_MDC_GETINFO:
1342 case IOC_MDC_GETFILEINFO:
1343 case IOC_MDC_GETFILESTRIPE: {
1344 struct ptlrpc_request *request = NULL;
1345 struct lov_user_md __user *lump;
1346 struct lov_mds_md *lmm = NULL;
1347 struct mdt_body *body;
1348 char *filename = NULL;
1351 if (cmd == IOC_MDC_GETFILEINFO ||
1352 cmd == IOC_MDC_GETFILESTRIPE) {
1353 filename = ll_getname((const char __user *)arg);
1354 if (IS_ERR(filename))
1355 RETURN(PTR_ERR(filename));
1357 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1358 &lmmsize, &request);
1360 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1365 body = req_capsule_server_get(&request->rq_pill,
1367 LASSERT(body != NULL);
1373 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1374 cmd == LL_IOC_MDC_GETINFO))
1375 GOTO(skip_lmm, rc = 0);
1380 if (cmd == IOC_MDC_GETFILESTRIPE ||
1381 cmd == LL_IOC_LOV_GETSTRIPE) {
1382 lump = (struct lov_user_md __user *)arg;
1384 struct lov_user_mds_data __user *lmdp;
1385 lmdp = (struct lov_user_mds_data __user *)arg;
1386 lump = &lmdp->lmd_lmm;
1388 if (copy_to_user(lump, lmm, lmmsize)) {
1389 if (copy_to_user(lump, lmm, sizeof(*lump)))
1390 GOTO(out_req, rc = -EFAULT);
1394 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1395 struct lov_user_mds_data __user *lmdp;
1398 st.st_dev = inode->i_sb->s_dev;
1399 st.st_mode = body->mbo_mode;
1400 st.st_nlink = body->mbo_nlink;
1401 st.st_uid = body->mbo_uid;
1402 st.st_gid = body->mbo_gid;
1403 st.st_rdev = body->mbo_rdev;
1404 st.st_size = body->mbo_size;
1405 st.st_blksize = PAGE_CACHE_SIZE;
1406 st.st_blocks = body->mbo_blocks;
1407 st.st_atime = body->mbo_atime;
1408 st.st_mtime = body->mbo_mtime;
1409 st.st_ctime = body->mbo_ctime;
1410 st.st_ino = inode->i_ino;
1412 lmdp = (struct lov_user_mds_data __user *)arg;
1413 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1414 GOTO(out_req, rc = -EFAULT);
1419 ptlrpc_req_finished(request);
1421 ll_putname(filename);
1424 case OBD_IOC_QUOTACHECK: {
1425 struct obd_quotactl *oqctl;
1428 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1429 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1432 OBD_ALLOC_PTR(oqctl);
1435 oqctl->qc_type = arg;
1436 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1438 CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1442 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1444 CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1446 OBD_FREE_PTR(oqctl);
1449 case OBD_IOC_POLL_QUOTACHECK: {
1450 struct if_quotacheck *check;
1452 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1453 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1456 OBD_ALLOC_PTR(check);
1460 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1463 CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1464 if (copy_to_user((void __user *)arg, check,
1466 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1470 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1473 CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1474 if (copy_to_user((void __user *)arg, check,
1476 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1480 OBD_FREE_PTR(check);
1483 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 53, 0)
1484 case LL_IOC_QUOTACTL_18: {
1485 /* copy the old 1.x quota struct for internal use, then copy
1486 * back into old format struct. For 1.8 compatibility. */
1487 struct if_quotactl_18 *qctl_18;
1488 struct if_quotactl *qctl_20;
1490 OBD_ALLOC_PTR(qctl_18);
1494 OBD_ALLOC_PTR(qctl_20);
1496 GOTO(out_quotactl_18, rc = -ENOMEM);
1498 if (copy_from_user(qctl_18, (void __user *)arg,
1500 GOTO(out_quotactl_20, rc = -ENOMEM);
1502 QCTL_COPY(qctl_20, qctl_18);
1503 qctl_20->qc_idx = 0;
1505 /* XXX: dqb_valid was borrowed as a flag to mark that
1506 * only mds quota is wanted */
1507 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1508 qctl_18->qc_dqblk.dqb_valid) {
1509 qctl_20->qc_valid = QC_MDTIDX;
1510 qctl_20->qc_dqblk.dqb_valid = 0;
1511 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1512 qctl_20->qc_valid = QC_UUID;
1513 qctl_20->obd_uuid = qctl_18->obd_uuid;
1515 qctl_20->qc_valid = QC_GENERAL;
1518 rc = quotactl_ioctl(sbi, qctl_20);
1521 QCTL_COPY(qctl_18, qctl_20);
1522 qctl_18->obd_uuid = qctl_20->obd_uuid;
1524 if (copy_to_user((void __user *)arg, qctl_18,
1530 OBD_FREE_PTR(qctl_20);
1532 OBD_FREE_PTR(qctl_18);
1535 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 53, 0) */
1536 case OBD_IOC_QUOTACTL: {
1537 struct if_quotactl *qctl;
1539 OBD_ALLOC_PTR(qctl);
1543 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1544 GOTO(out_quotactl, rc = -EFAULT);
1546 rc = quotactl_ioctl(sbi, qctl);
1549 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1556 case OBD_IOC_GETDTNAME:
1557 case OBD_IOC_GETMDNAME:
1558 RETURN(ll_get_obd_name(inode, cmd, arg));
1559 case LL_IOC_FLUSHCTX:
1560 RETURN(ll_flush_ctx(inode));
1561 #ifdef CONFIG_FS_POSIX_ACL
1562 case LL_IOC_RMTACL: {
1563 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1564 inode == inode->i_sb->s_root->d_inode) {
1565 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1567 LASSERT(fd != NULL);
1568 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1570 fd->fd_flags |= LL_FILE_RMTACL;
1576 case LL_IOC_GETOBDCOUNT: {
1578 struct obd_export *exp;
1580 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1583 /* get ost count when count is zero, get mdt count otherwise */
1584 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1585 vallen = sizeof(count);
1586 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1587 KEY_TGT_COUNT, &vallen, &count, NULL);
1589 CERROR("get target count failed: %d\n", rc);
1593 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1598 case LL_IOC_PATH2FID:
1599 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1600 sizeof(struct lu_fid)))
1603 case LL_IOC_GET_CONNECT_FLAGS: {
1604 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1605 (void __user *)arg));
1607 case OBD_IOC_CHANGELOG_SEND:
1608 case OBD_IOC_CHANGELOG_CLEAR:
1609 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1610 sizeof(struct ioc_changelog));
1612 case OBD_IOC_FID2PATH:
1613 RETURN(ll_fid2path(inode, (void __user *)arg));
1614 case LL_IOC_GETPARENT:
1615 RETURN(ll_getparent(file, (void __user *)arg));
1616 case LL_IOC_FID2MDTIDX: {
1617 struct obd_export *exp = ll_i2mdexp(inode);
1621 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1625 /* Call mdc_iocontrol */
1626 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1627 (__u32 __user *)&index);
1633 case LL_IOC_HSM_REQUEST: {
1634 struct hsm_user_request *hur;
1641 /* We don't know the true size yet; copy the fixed-size part */
1642 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1647 /* Compute the whole struct size */
1648 totalsize = hur_len(hur);
1653 /* Final size will be more than double totalsize */
1654 if (totalsize >= MDS_MAXREQSIZE / 3)
1657 OBD_ALLOC_LARGE(hur, totalsize);
1661 /* Copy the whole struct */
1662 if (copy_from_user(hur, (void __user *)arg, totalsize)) {
1663 OBD_FREE_LARGE(hur, totalsize);
1667 if (hur->hur_request.hr_action == HUA_RELEASE) {
1668 const struct lu_fid *fid;
1672 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1673 fid = &hur->hur_user_item[i].hui_fid;
1674 f = search_inode_for_lustre(inode->i_sb, fid);
1680 rc = ll_hsm_release(f);
1686 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1690 OBD_FREE_LARGE(hur, totalsize);
1694 case LL_IOC_HSM_PROGRESS: {
1695 struct hsm_progress_kernel hpk;
1696 struct hsm_progress hp;
1698 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1701 hpk.hpk_fid = hp.hp_fid;
1702 hpk.hpk_cookie = hp.hp_cookie;
1703 hpk.hpk_extent = hp.hp_extent;
1704 hpk.hpk_flags = hp.hp_flags;
1705 hpk.hpk_errval = hp.hp_errval;
1706 hpk.hpk_data_version = 0;
1708 /* File may not exist in Lustre; all progress
1709 * reported to Lustre root */
1710 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1714 case LL_IOC_HSM_CT_START:
1715 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1718 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1719 sizeof(struct lustre_kernelcomm));
1722 case LL_IOC_HSM_COPY_START: {
1723 struct hsm_copy *copy;
1726 OBD_ALLOC_PTR(copy);
1729 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1734 rc = ll_ioc_copy_start(inode->i_sb, copy);
1735 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1741 case LL_IOC_HSM_COPY_END: {
1742 struct hsm_copy *copy;
1745 OBD_ALLOC_PTR(copy);
1748 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1753 rc = ll_ioc_copy_end(inode->i_sb, copy);
1754 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1760 case LL_IOC_MIGRATE: {
1762 const char *filename;
1768 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1772 data = (struct obd_ioctl_data *)buf;
1773 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1774 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1775 GOTO(migrate_free, rc = -EINVAL);
1777 filename = data->ioc_inlbuf1;
1778 namelen = data->ioc_inllen1;
1779 /* \0 is packed at the end of filename */
1780 if (namelen < 1 || namelen != strlen(filename) + 1)
1781 GOTO(migrate_free, rc = -EINVAL);
1783 if (data->ioc_inllen2 != sizeof(mdtidx))
1784 GOTO(migrate_free, rc = -EINVAL);
1785 mdtidx = *(int *)data->ioc_inlbuf2;
1787 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1789 obd_ioctl_freedata(buf, len);
1794 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1795 (void __user *)arg));
1799 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1801 struct inode *inode = file->f_mapping->host;
1802 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1803 struct ll_sb_info *sbi = ll_i2sbi(inode);
1804 int api32 = ll_need_32bit_api(sbi);
1805 loff_t ret = -EINVAL;
1808 mutex_lock(&inode->i_mutex);
1813 offset += file->f_pos;
1819 offset += LL_DIR_END_OFF_32BIT;
1821 offset += LL_DIR_END_OFF;
1828 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1829 (!api32 && offset <= LL_DIR_END_OFF))) {
1830 if (offset != file->f_pos) {
1831 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1832 (!api32 && offset == LL_DIR_END_OFF))
1833 fd->lfd_pos = MDS_DIR_END_OFF;
1834 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1835 fd->lfd_pos = offset << 32;
1837 fd->lfd_pos = offset;
1838 file->f_pos = offset;
1839 file->f_version = 0;
1846 mutex_unlock(&inode->i_mutex);
1850 static int ll_dir_open(struct inode *inode, struct file *file)
1853 RETURN(ll_file_open(inode, file));
1856 static int ll_dir_release(struct inode *inode, struct file *file)
1859 RETURN(ll_file_release(inode, file));
1862 const struct file_operations ll_dir_operations = {
1863 .llseek = ll_dir_seek,
1864 .open = ll_dir_open,
1865 .release = ll_dir_release,
1866 .read = generic_read_dir,
1867 #ifdef HAVE_DIR_CONTEXT
1868 .iterate = ll_iterate,
1870 .readdir = ll_readdir,
1872 .unlocked_ioctl = ll_dir_ioctl,