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 <linux/user_namespace.h>
46 #ifdef HAVE_UIDGID_HEADER
47 # include <linux/uidgid.h>
49 #include <asm/uaccess.h>
50 #include <linux/buffer_head.h> // for wait_on_buffer
51 #include <linux/pagevec.h>
53 #define DEBUG_SUBSYSTEM S_LLITE
55 #include <lustre/lustre_idl.h>
56 #include <obd_support.h>
57 #include <obd_class.h>
58 #include <lustre_ioctl.h>
59 #include <lustre_lib.h>
60 #include <lustre_dlm.h>
61 #include <lustre_fid.h>
62 #include <lustre_kernelcomm.h>
63 #include "llite_internal.h"
66 * (new) readdir implementation overview.
68 * Original lustre readdir implementation cached exact copy of raw directory
69 * pages on the client. These pages were indexed in client page cache by
70 * logical offset in the directory file. This design, while very simple and
71 * intuitive had some inherent problems:
73 * . it implies that byte offset to the directory entry serves as a
74 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
75 * ext3/htree directory entries may move due to splits, and more
78 * . it is incompatible with the design of split directories for cmd3,
79 * that assumes that names are distributed across nodes based on their
80 * hash, and so readdir should be done in hash order.
82 * New readdir implementation does readdir in hash order, and uses hash of a
83 * file name as a telldir/seekdir cookie. This led to number of complications:
85 * . hash is not unique, so it cannot be used to index cached directory
86 * pages on the client (note, that it requires a whole pageful of hash
87 * collided entries to cause two pages to have identical hashes);
89 * . hash is not unique, so it cannot, strictly speaking, be used as an
90 * entry cookie. ext3/htree has the same problem and lustre implementation
91 * mimics their solution: seekdir(hash) positions directory at the first
92 * entry with the given hash.
98 * Client caches directory pages using hash of the first entry as an index. As
99 * noted above hash is not unique, so this solution doesn't work as is:
100 * special processing is needed for "page hash chains" (i.e., sequences of
101 * pages filled with entries all having the same hash value).
103 * First, such chains have to be detected. To this end, server returns to the
104 * client the hash of the first entry on the page next to one returned. When
105 * client detects that this hash is the same as hash of the first entry on the
106 * returned page, page hash collision has to be handled. Pages in the
107 * hash chain, except first one, are termed "overflow pages".
109 * Solution to index uniqueness problem is to not cache overflow
110 * pages. Instead, when page hash collision is detected, all overflow pages
111 * from emerging chain are immediately requested from the server and placed in
112 * a special data structure (struct ll_dir_chain). This data structure is used
113 * by ll_readdir() to process entries from overflow pages. When readdir
114 * invocation finishes, overflow pages are discarded. If page hash collision
115 * chain weren't completely processed, next call to readdir will again detect
116 * page hash collision, again read overflow pages in, process next portion of
117 * entries and again discard the pages. This is not as wasteful as it looks,
118 * because, given reasonable hash, page hash collisions are extremely rare.
120 * 1. directory positioning
122 * When seekdir(hash) is called, original
133 * identification of and access to overflow pages
137 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
138 * a header lu_dirpage which describes the start/end hash, and whether this
139 * page is empty (contains no dir entry) or hash collide with next page.
140 * After client receives reply, several pages will be integrated into dir page
141 * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
142 * lu_dirpage for this integrated page will be adjusted. See
143 * mdc_adjust_dirpages().
146 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
147 __u64 offset, struct ll_dir_chain *chain)
149 struct md_callback cb_op;
153 cb_op.md_blocking_ast = ll_md_blocking_ast;
154 rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
161 void ll_release_page(struct inode *inode, struct page *page,
166 /* Always remove the page for striped dir, because the page is
167 * built from temporarily in LMV layer */
168 if (inode != NULL && S_ISDIR(inode->i_mode) &&
169 ll_i2info(inode)->lli_lsm_md != NULL) {
176 if (likely(page->mapping != NULL))
177 truncate_complete_page(page->mapping, page);
180 page_cache_release(page);
184 * return IF_* type for given lu_dirent entry.
185 * IF_* flag shld be converted to particular OS file type in
186 * platform llite module.
188 static u16 ll_dirent_type_get(struct lu_dirent *ent)
191 struct luda_type *lt;
194 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
195 const unsigned align = sizeof(struct luda_type) - 1;
197 len = le16_to_cpu(ent->lde_namelen);
198 len = (len + align) & ~align;
199 lt = (void *)ent->lde_name + len;
200 type = IFTODT(le16_to_cpu(lt->lt_type));
206 #ifdef HAVE_DIR_CONTEXT
207 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
208 struct dir_context *ctx)
211 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
212 void *cookie, filldir_t filldir)
215 struct ll_sb_info *sbi = ll_i2sbi(inode);
217 bool is_api32 = ll_need_32bit_api(sbi);
218 bool is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
220 struct ll_dir_chain chain;
225 ll_dir_chain_init(&chain);
227 page = ll_get_dir_page(inode, op_data, pos, &chain);
229 while (rc == 0 && !done) {
230 struct lu_dirpage *dp;
231 struct lu_dirent *ent;
240 hash = MDS_DIR_END_OFF;
241 dp = page_address(page);
242 for (ent = lu_dirent_start(dp); ent != NULL && !done;
243 ent = lu_dirent_next(ent)) {
250 hash = le64_to_cpu(ent->lde_hash);
253 * Skip until we find target hash
258 namelen = le16_to_cpu(ent->lde_namelen);
265 if (is_api32 && is_hash64)
269 fid_le_to_cpu(&fid, &ent->lde_fid);
270 ino = cl_fid_build_ino(&fid, is_api32);
271 type = ll_dirent_type_get(ent);
272 /* For 'll_nfs_get_name_filldir()', it will try
273 * to access the 'ent' through its 'lde_name',
274 * so the parameter 'name' for 'filldir()' must
275 * be part of the 'ent'. */
276 #ifdef HAVE_DIR_CONTEXT
278 done = !dir_emit(ctx, ent->lde_name, namelen, ino,
281 done = filldir(cookie, ent->lde_name, namelen, lhash,
288 ll_release_page(inode, page, false);
292 next = le64_to_cpu(dp->ldp_hash_end);
294 if (pos == MDS_DIR_END_OFF) {
296 * End of directory reached.
299 ll_release_page(inode, page, false);
302 * Normal case: continue to the next
305 ll_release_page(inode, page,
306 le32_to_cpu(dp->ldp_flags) &
309 page = ll_get_dir_page(inode, op_data, pos,
313 #ifdef HAVE_DIR_CONTEXT
318 ll_dir_chain_fini(&chain);
322 #ifdef HAVE_DIR_CONTEXT
323 static int ll_iterate(struct file *filp, struct dir_context *ctx)
325 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
328 struct inode *inode = filp->f_path.dentry->d_inode;
329 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
330 struct ll_sb_info *sbi = ll_i2sbi(inode);
331 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
332 int api32 = ll_need_32bit_api(sbi);
333 struct md_op_data *op_data;
343 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
344 "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
345 inode, (unsigned long)pos, i_size_read(inode), api32);
347 if (pos == MDS_DIR_END_OFF)
353 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
354 LUSTRE_OPC_ANY, inode);
356 GOTO(out, rc = PTR_ERR(op_data));
358 if (unlikely(op_data->op_mea1 != NULL)) {
359 /* This is only needed for striped dir to fill ..,
360 * see lmv_read_entry */
361 if (filp->f_path.dentry->d_parent != NULL &&
362 filp->f_path.dentry->d_parent->d_inode != NULL) {
363 __u64 ibits = MDS_INODELOCK_UPDATE;
364 struct inode *parent =
365 filp->f_path.dentry->d_parent->d_inode;
367 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
368 op_data->op_fid3 = *ll_inode2fid(parent);
371 /* If it can not find in cache, do lookup .. on the master
373 if (fid_is_zero(&op_data->op_fid3)) {
374 rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
376 ll_finish_md_op_data(op_data);
381 op_data->op_max_pages = sbi->ll_md_brw_pages;
382 #ifdef HAVE_DIR_CONTEXT
384 rc = ll_dir_read(inode, &pos, op_data, ctx);
387 rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
392 if (pos == MDS_DIR_END_OFF) {
394 pos = LL_DIR_END_OFF_32BIT;
396 pos = LL_DIR_END_OFF;
401 #ifdef HAVE_DIR_CONTEXT
406 ll_finish_md_op_data(op_data);
407 filp->f_version = inode->i_version;
411 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
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);
436 * Create striped directory with specified stripe(@lump)
438 * param[in]parent the parent of the directory.
439 * param[in]lump the specified stripes.
440 * param[in]dirname the name of the directory.
441 * param[in]mode the specified mode of the directory.
443 * retval =0 if striped directory is being created successfully.
444 * <0 if the creation is failed.
446 static int ll_dir_setdirstripe(struct inode *parent, struct lmv_user_md *lump,
447 const char *dirname, umode_t mode)
449 struct ptlrpc_request *request = NULL;
450 struct md_op_data *op_data;
451 struct ll_sb_info *sbi = ll_i2sbi(parent);
452 struct inode *inode = NULL;
453 struct dentry dentry;
457 if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
460 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
461 "stripe_offset %d, stripe_count: %u\n",
462 PFID(ll_inode2fid(parent)), parent, dirname,
463 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
465 if (lump->lum_stripe_count > 1 &&
466 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
469 if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
470 lustre_swab_lmv_user_md(lump);
472 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
473 mode &= ~current_umask();
474 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
475 op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
476 strlen(dirname), mode, LUSTRE_OPC_MKDIR,
479 GOTO(err_exit, err = PTR_ERR(op_data));
481 op_data->op_cli_flags |= CLI_SET_MEA;
482 err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
483 from_kuid(&init_user_ns, current_fsuid()),
484 from_kgid(&init_user_ns, current_fsgid()),
485 cfs_curproc_cap_pack(), 0, &request);
486 ll_finish_md_op_data(op_data);
490 err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
494 memset(&dentry, 0, sizeof(dentry));
495 dentry.d_inode = inode;
497 err = ll_init_security(&dentry, inode, parent);
503 ptlrpc_req_finished(request);
507 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
510 struct ll_sb_info *sbi = ll_i2sbi(inode);
511 struct md_op_data *op_data;
512 struct ptlrpc_request *req = NULL;
514 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
515 struct obd_device *mgc = lsi->lsi_mgc;
521 * This is coming from userspace, so should be in
522 * local endian. But the MDS would like it in little
523 * endian, so we swab it before we send it.
525 switch (lump->lmm_magic) {
526 case LOV_USER_MAGIC_V1: {
527 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
528 lustre_swab_lov_user_md_v1(lump);
529 lum_size = sizeof(struct lov_user_md_v1);
532 case LOV_USER_MAGIC_V3: {
533 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
534 lustre_swab_lov_user_md_v3(
535 (struct lov_user_md_v3 *)lump);
536 lum_size = sizeof(struct lov_user_md_v3);
539 case LMV_USER_MAGIC: {
540 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
541 lustre_swab_lmv_user_md(
542 (struct lmv_user_md *)lump);
543 lum_size = sizeof(struct lmv_user_md);
547 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
548 " %#08x != %#08x nor %#08x\n",
549 lump->lmm_magic, LOV_USER_MAGIC_V1,
555 lum_size = sizeof(struct lov_user_md_v1);
558 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
559 LUSTRE_OPC_ANY, NULL);
561 RETURN(PTR_ERR(op_data));
563 /* swabbing is done in lov_setstripe() on server side */
564 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
565 ll_finish_md_op_data(op_data);
566 ptlrpc_req_finished(req);
568 if (rc != -EPERM && rc != -EACCES)
569 CERROR("mdc_setattr fails: rc = %d\n", rc);
572 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
573 LOV_USER_MAGIC_V3 have the same initial fields so we do not
574 need the make the distiction between the 2 versions */
575 if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
579 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
581 GOTO(end, rc = -ENOMEM);
584 /* Get fsname and assume devname to be -MDT0000. */
585 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
586 strcat(buf, "-MDT0000.lov");
589 /* Set root stripesize */
590 sprintf(buf, ".stripesize=%u",
591 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
592 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
596 /* Set root stripecount */
597 sprintf(buf, ".stripecount=%hd",
598 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
599 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
603 /* Set root stripeoffset */
604 sprintf(buf, ".stripeoffset=%hd",
605 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
606 (typeof(lump->lmm_stripe_offset))(-1));
607 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
611 OBD_FREE(param, MGS_PARAM_MAXLEN);
617 * This function will be used to get default LOV/LMV/Default LMV
618 * @valid will be used to indicate which stripe it will retrieve
619 * OBD_MD_MEA LMV stripe EA
620 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
621 * otherwise Default LOV EA.
622 * Each time, it can only retrieve 1 stripe EA
624 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
625 struct ptlrpc_request **request, u64 valid)
627 struct ll_sb_info *sbi = ll_i2sbi(inode);
628 struct mdt_body *body;
629 struct lov_mds_md *lmm = NULL;
630 struct ptlrpc_request *req = NULL;
632 struct md_op_data *op_data;
635 rc = ll_get_default_mdsize(sbi, &lmm_size);
639 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
640 0, lmm_size, LUSTRE_OPC_ANY,
643 RETURN(PTR_ERR(op_data));
645 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
646 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
647 ll_finish_md_op_data(op_data);
649 CDEBUG(D_INFO, "md_getattr failed on inode "
650 DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
654 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
655 LASSERT(body != NULL);
657 lmm_size = body->mbo_eadatasize;
659 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
661 GOTO(out, rc = -ENODATA);
664 lmm = req_capsule_server_sized_get(&req->rq_pill,
665 &RMF_MDT_MD, lmm_size);
666 LASSERT(lmm != NULL);
669 * This is coming from the MDS, so is probably in
670 * little endian. We convert it to host endian before
671 * passing it to userspace.
673 /* We don't swab objects for directories */
674 switch (le32_to_cpu(lmm->lmm_magic)) {
676 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
677 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
680 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
681 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
684 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
685 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
688 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
689 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
692 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
697 *plmm_size = lmm_size;
702 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
704 struct md_op_data *op_data;
709 OBD_ALLOC_PTR(op_data);
713 op_data->op_flags |= MF_GET_MDT_IDX;
714 op_data->op_fid1 = *fid;
715 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
716 mdt_index = op_data->op_mds;
717 OBD_FREE_PTR(op_data);
725 * Get MDT index for the inode.
727 int ll_get_mdt_idx(struct inode *inode)
729 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
733 * Generic handler to do any pre-copy work.
735 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
736 * first information for it that real work has started.
738 * Moreover, for a ARCHIVE request, it will sample the file data version and
739 * store it in \a copy.
741 * \return 0 on success.
743 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
745 struct ll_sb_info *sbi = ll_s2sbi(sb);
746 struct hsm_progress_kernel hpk;
751 /* Forge a hsm_progress based on data from copy. */
752 hpk.hpk_fid = copy->hc_hai.hai_fid;
753 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
754 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
755 hpk.hpk_extent.length = 0;
758 hpk.hpk_data_version = 0;
761 /* For archive request, we need to read the current file version. */
762 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
764 __u64 data_version = 0;
766 /* Get inode for this fid */
767 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
769 hpk.hpk_flags |= HP_FLAG_RETRY;
770 /* hpk_errval is >= 0 */
771 hpk.hpk_errval = -PTR_ERR(inode);
772 GOTO(progress, rc = PTR_ERR(inode));
775 /* Read current file data version */
776 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
779 CDEBUG(D_HSM, "Could not read file data version of "
780 DFID" (rc = %d). Archive request ("
781 LPX64") could not be done.\n",
782 PFID(©->hc_hai.hai_fid), rc,
783 copy->hc_hai.hai_cookie);
784 hpk.hpk_flags |= HP_FLAG_RETRY;
785 /* hpk_errval must be >= 0 */
786 hpk.hpk_errval = -rc;
790 /* Store in the hsm_copy for later copytool use.
791 * Always modified even if no lsm. */
792 copy->hc_data_version = data_version;
796 /* On error, the request should be considered as completed */
797 if (hpk.hpk_errval > 0)
798 hpk.hpk_flags |= HP_FLAG_COMPLETED;
800 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
803 /* Return first error */
804 RETURN(rc != 0 ? rc : rc2);
808 * Generic handler to do any post-copy work.
810 * It will send the last hsm_progress update to coordinator to inform it
811 * that copy is finished and whether it was successful or not.
814 * - for ARCHIVE request, it will sample the file data version and compare it
815 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
816 * will be considered as failed.
817 * - for RESTORE request, it will sample the file data version and send it to
818 * coordinator which is useful if the file was imported as 'released'.
820 * \return 0 on success.
822 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
824 struct ll_sb_info *sbi = ll_s2sbi(sb);
825 struct hsm_progress_kernel hpk;
830 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
831 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
832 * initialized if copy_end was called with copy == NULL.
835 /* Forge a hsm_progress based on data from copy. */
836 hpk.hpk_fid = copy->hc_hai.hai_fid;
837 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
838 hpk.hpk_extent = copy->hc_hai.hai_extent;
839 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
840 hpk.hpk_errval = copy->hc_errval;
841 hpk.hpk_data_version = 0;
843 /* For archive request, we need to check the file data was not changed.
845 * For restore request, we need to send the file data version, this is
846 * useful when the file was created using hsm_import.
848 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
849 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
850 (copy->hc_errval == 0)) {
852 __u64 data_version = 0;
854 /* Get lsm for this fid */
855 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
857 hpk.hpk_flags |= HP_FLAG_RETRY;
858 /* hpk_errval must be >= 0 */
859 hpk.hpk_errval = -PTR_ERR(inode);
860 GOTO(progress, rc = PTR_ERR(inode));
863 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
866 CDEBUG(D_HSM, "Could not read file data version. "
867 "Request could not be confirmed.\n");
868 if (hpk.hpk_errval == 0)
869 hpk.hpk_errval = -rc;
873 /* Store in the hsm_copy for later copytool use.
874 * Always modified even if no lsm. */
875 hpk.hpk_data_version = data_version;
877 /* File could have been stripped during archiving, so we need
878 * to check anyway. */
879 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
880 (copy->hc_data_version != data_version)) {
881 CDEBUG(D_HSM, "File data version mismatched. "
882 "File content was changed during archiving. "
883 DFID", start:"LPX64" current:"LPX64"\n",
884 PFID(©->hc_hai.hai_fid),
885 copy->hc_data_version, data_version);
886 /* File was changed, send error to cdt. Do not ask for
887 * retry because if a file is modified frequently,
888 * the cdt will loop on retried archive requests.
889 * The policy engine will ask for a new archive later
890 * when the file will not be modified for some tunable
892 hpk.hpk_flags &= ~HP_FLAG_RETRY;
894 /* hpk_errval must be >= 0 */
895 hpk.hpk_errval = -rc;
902 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
905 /* Return first error */
906 RETURN(rc != 0 ? rc : rc2);
910 static int copy_and_ioctl(int cmd, struct obd_export *exp,
911 const void __user *data, size_t size)
916 OBD_ALLOC(copy, size);
920 if (copy_from_user(copy, data, size)) {
925 rc = obd_iocontrol(cmd, exp, size, copy, NULL);
927 OBD_FREE(copy, size);
932 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
934 int cmd = qctl->qc_cmd;
935 int type = qctl->qc_type;
936 int id = qctl->qc_id;
937 int valid = qctl->qc_valid;
944 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
945 sbi->ll_flags & LL_SBI_RMT_CLIENT)
949 if (((type == USRQUOTA &&
950 !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
952 !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
953 (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
954 sbi->ll_flags & LL_SBI_RMT_CLIENT))
960 CERROR("unsupported quotactl op: %#x\n", cmd);
964 if (valid != QC_GENERAL) {
965 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
968 if (cmd == Q_GETINFO)
969 qctl->qc_cmd = Q_GETOINFO;
970 else if (cmd == Q_GETQUOTA)
971 qctl->qc_cmd = Q_GETOQUOTA;
977 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
978 sizeof(*qctl), qctl, NULL);
981 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
982 sizeof(*qctl), qctl, NULL);
985 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
986 sizeof(*qctl), qctl, NULL);
988 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
990 sizeof(*qctl), qctl, NULL);
1002 struct obd_quotactl *oqctl;
1004 OBD_ALLOC_PTR(oqctl);
1008 QCTL_COPY(oqctl, qctl);
1009 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1011 OBD_FREE_PTR(oqctl);
1014 /* If QIF_SPACE is not set, client should collect the
1015 * space usage from OSSs by itself */
1016 if (cmd == Q_GETQUOTA &&
1017 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1018 !oqctl->qc_dqblk.dqb_curspace) {
1019 struct obd_quotactl *oqctl_tmp;
1021 OBD_ALLOC_PTR(oqctl_tmp);
1022 if (oqctl_tmp == NULL)
1023 GOTO(out, rc = -ENOMEM);
1025 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1026 oqctl_tmp->qc_id = oqctl->qc_id;
1027 oqctl_tmp->qc_type = oqctl->qc_type;
1029 /* collect space usage from OSTs */
1030 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1031 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1032 if (!rc || rc == -EREMOTEIO) {
1033 oqctl->qc_dqblk.dqb_curspace =
1034 oqctl_tmp->qc_dqblk.dqb_curspace;
1035 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1038 /* collect space & inode usage from MDTs */
1039 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1040 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1041 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1042 if (!rc || rc == -EREMOTEIO) {
1043 oqctl->qc_dqblk.dqb_curspace +=
1044 oqctl_tmp->qc_dqblk.dqb_curspace;
1045 oqctl->qc_dqblk.dqb_curinodes =
1046 oqctl_tmp->qc_dqblk.dqb_curinodes;
1047 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1049 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1052 OBD_FREE_PTR(oqctl_tmp);
1055 QCTL_COPY(qctl, oqctl);
1056 OBD_FREE_PTR(oqctl);
1062 /* This function tries to get a single name component,
1063 * to send to the server. No actual path traversal involved,
1064 * so we limit to NAME_MAX */
1065 static char *ll_getname(const char __user *filename)
1070 OBD_ALLOC(tmp, NAME_MAX + 1);
1073 return ERR_PTR(-ENOMEM);
1075 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1078 else if (len > NAME_MAX)
1079 ret = -ENAMETOOLONG;
1082 OBD_FREE(tmp, NAME_MAX + 1);
1088 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1090 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1092 struct inode *inode = file->f_path.dentry->d_inode;
1093 struct ll_sb_info *sbi = ll_i2sbi(inode);
1094 struct obd_ioctl_data *data;
1098 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1099 PFID(ll_inode2fid(inode)), inode, cmd);
1101 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1102 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1105 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1107 case FSFILT_IOC_GETFLAGS:
1108 case FSFILT_IOC_SETFLAGS:
1109 RETURN(ll_iocontrol(inode, file, cmd, arg));
1110 case FSFILT_IOC_GETVERSION_OLD:
1111 case FSFILT_IOC_GETVERSION:
1112 RETURN(put_user(inode->i_generation, (int __user *)arg));
1113 /* We need to special case any other ioctls we want to handle,
1114 * to send them to the MDS/OST as appropriate and to properly
1115 * network encode the arg field.
1116 case FSFILT_IOC_SETVERSION_OLD:
1117 case FSFILT_IOC_SETVERSION:
1119 case LL_IOC_GET_MDTIDX: {
1122 mdtidx = ll_get_mdt_idx(inode);
1126 if (put_user((int)mdtidx, (int __user *)arg))
1131 case IOC_MDC_LOOKUP: {
1132 int namelen, len = 0;
1136 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1141 filename = data->ioc_inlbuf1;
1142 namelen = strlen(filename);
1144 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1145 GOTO(out_free, rc = -EINVAL);
1148 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1150 CERROR("%s: lookup %.*s failed: rc = %d\n",
1151 ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1156 obd_ioctl_freedata(buf, len);
1159 case LL_IOC_LMV_SETSTRIPE: {
1160 struct lmv_user_md *lum;
1169 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1174 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1175 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1176 GOTO(lmv_out_free, rc = -EINVAL);
1178 filename = data->ioc_inlbuf1;
1179 namelen = data->ioc_inllen1;
1182 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1183 GOTO(lmv_out_free, rc = -EINVAL);
1185 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1186 lumlen = data->ioc_inllen2;
1188 if (lum->lum_magic != LMV_USER_MAGIC ||
1189 lumlen != sizeof(*lum)) {
1190 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1191 filename, lum->lum_magic, lumlen, -EFAULT);
1192 GOTO(lmv_out_free, rc = -EINVAL);
1195 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1196 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1198 mode = data->ioc_type;
1200 rc = ll_dir_setdirstripe(inode, lum, filename, mode);
1202 obd_ioctl_freedata(buf, len);
1206 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1207 struct lmv_user_md lum;
1208 struct lmv_user_md __user *ulump =
1209 (struct lmv_user_md __user *)arg;
1212 if (copy_from_user(&lum, ulump, sizeof(lum)))
1215 if (lum.lum_magic != LMV_USER_MAGIC)
1218 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1222 case LL_IOC_LOV_SETSTRIPE: {
1223 struct lov_user_md_v3 lumv3;
1224 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1225 struct lov_user_md_v1 __user *lumv1p =
1226 (struct lov_user_md_v1 __user *)arg;
1227 struct lov_user_md_v3 __user *lumv3p =
1228 (struct lov_user_md_v3 __user *)arg;
1230 int set_default = 0;
1232 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1233 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1234 sizeof(lumv3p->lmm_objects[0]));
1235 /* first try with v1 which is smaller than v3 */
1236 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1239 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1240 if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1244 if (inode->i_sb->s_root == file->f_path.dentry)
1247 /* in v1 and v3 cases lumv1 points to data */
1248 rc = ll_dir_setstripe(inode, lumv1, set_default);
1252 case LL_IOC_LMV_GETSTRIPE: {
1253 struct lmv_user_md __user *ulmv =
1254 (struct lmv_user_md __user *)arg;
1255 struct lmv_user_md lum;
1256 struct ptlrpc_request *request = NULL;
1257 union lmv_mds_md *lmm = NULL;
1260 struct lmv_user_md *tmp = NULL;
1264 int max_stripe_count;
1268 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1271 max_stripe_count = lum.lum_stripe_count;
1272 /* lum_magic will indicate which stripe the ioctl will like
1273 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1274 * is for default LMV stripe */
1275 if (lum.lum_magic == LMV_MAGIC_V1)
1276 valid |= OBD_MD_MEA;
1277 else if (lum.lum_magic == LMV_USER_MAGIC)
1278 valid |= OBD_MD_DEFAULT_MEA;
1282 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1285 GOTO(finish_req, rc);
1287 /* Get default LMV EA */
1288 if (lum.lum_magic == LMV_USER_MAGIC) {
1290 GOTO(finish_req, rc);
1292 if (lmmsize > sizeof(*ulmv))
1293 GOTO(finish_req, rc = -EINVAL);
1295 if (copy_to_user(ulmv, lmm, lmmsize))
1296 GOTO(finish_req, rc = -EFAULT);
1298 GOTO(finish_req, rc);
1301 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1302 if (max_stripe_count < stripe_count) {
1303 lum.lum_stripe_count = stripe_count;
1304 if (copy_to_user(ulmv, &lum, sizeof(lum)))
1305 GOTO(finish_req, rc = -EFAULT);
1306 GOTO(finish_req, rc = -E2BIG);
1309 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1310 OBD_ALLOC(tmp, lum_size);
1312 GOTO(finish_req, rc = -ENOMEM);
1314 mdt_index = ll_get_mdt_idx(inode);
1316 GOTO(out_tmp, rc = -ENOMEM);
1318 tmp->lum_magic = LMV_MAGIC_V1;
1319 tmp->lum_stripe_count = 0;
1320 tmp->lum_stripe_offset = mdt_index;
1321 for (i = 0; i < stripe_count; i++) {
1324 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1325 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1327 GOTO(out_tmp, rc = mdt_index);
1329 tmp->lum_objects[i].lum_mds = mdt_index;
1330 tmp->lum_objects[i].lum_fid = fid;
1331 tmp->lum_stripe_count++;
1334 if (copy_to_user(ulmv, tmp, lum_size))
1335 GOTO(out_tmp, rc = -EFAULT);
1337 OBD_FREE(tmp, lum_size);
1339 ptlrpc_req_finished(request);
1343 case LL_IOC_REMOVE_ENTRY: {
1344 char *filename = NULL;
1348 /* Here is a little hack to avoid sending REINT_RMENTRY to
1349 * unsupported server, which might crash the server(LU-2730),
1350 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1351 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1352 * server will support REINT_RMENTRY XXX*/
1353 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1356 filename = ll_getname((const char __user *)arg);
1357 if (IS_ERR(filename))
1358 RETURN(PTR_ERR(filename));
1360 namelen = strlen(filename);
1362 GOTO(out_rmdir, rc = -EINVAL);
1364 rc = ll_rmdir_entry(inode, filename, namelen);
1367 ll_putname(filename);
1370 case LL_IOC_LOV_SWAP_LAYOUTS:
1372 case IOC_OBD_STATFS:
1373 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1374 case LL_IOC_LOV_GETSTRIPE:
1375 case LL_IOC_MDC_GETINFO:
1376 case IOC_MDC_GETFILEINFO:
1377 case IOC_MDC_GETFILESTRIPE: {
1378 struct ptlrpc_request *request = NULL;
1379 struct lov_user_md __user *lump;
1380 struct lov_mds_md *lmm = NULL;
1381 struct mdt_body *body;
1382 char *filename = NULL;
1385 if (cmd == IOC_MDC_GETFILEINFO ||
1386 cmd == IOC_MDC_GETFILESTRIPE) {
1387 filename = ll_getname((const char __user *)arg);
1388 if (IS_ERR(filename))
1389 RETURN(PTR_ERR(filename));
1391 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1392 &lmmsize, &request);
1394 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1399 body = req_capsule_server_get(&request->rq_pill,
1401 LASSERT(body != NULL);
1407 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1408 cmd == LL_IOC_MDC_GETINFO))
1409 GOTO(skip_lmm, rc = 0);
1414 if (cmd == IOC_MDC_GETFILESTRIPE ||
1415 cmd == LL_IOC_LOV_GETSTRIPE) {
1416 lump = (struct lov_user_md __user *)arg;
1418 struct lov_user_mds_data __user *lmdp;
1419 lmdp = (struct lov_user_mds_data __user *)arg;
1420 lump = &lmdp->lmd_lmm;
1422 if (copy_to_user(lump, lmm, lmmsize)) {
1423 if (copy_to_user(lump, lmm, sizeof(*lump)))
1424 GOTO(out_req, rc = -EFAULT);
1428 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1429 struct lov_user_mds_data __user *lmdp;
1432 st.st_dev = inode->i_sb->s_dev;
1433 st.st_mode = body->mbo_mode;
1434 st.st_nlink = body->mbo_nlink;
1435 st.st_uid = body->mbo_uid;
1436 st.st_gid = body->mbo_gid;
1437 st.st_rdev = body->mbo_rdev;
1438 st.st_size = body->mbo_size;
1439 st.st_blksize = PAGE_CACHE_SIZE;
1440 st.st_blocks = body->mbo_blocks;
1441 st.st_atime = body->mbo_atime;
1442 st.st_mtime = body->mbo_mtime;
1443 st.st_ctime = body->mbo_ctime;
1444 st.st_ino = inode->i_ino;
1446 lmdp = (struct lov_user_mds_data __user *)arg;
1447 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1448 GOTO(out_req, rc = -EFAULT);
1453 ptlrpc_req_finished(request);
1455 ll_putname(filename);
1458 case OBD_IOC_QUOTACTL: {
1459 struct if_quotactl *qctl;
1461 OBD_ALLOC_PTR(qctl);
1465 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1466 GOTO(out_quotactl, rc = -EFAULT);
1468 rc = quotactl_ioctl(sbi, qctl);
1471 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1478 case OBD_IOC_GETDTNAME:
1479 case OBD_IOC_GETMDNAME:
1480 RETURN(ll_get_obd_name(inode, cmd, arg));
1481 case LL_IOC_FLUSHCTX:
1482 RETURN(ll_flush_ctx(inode));
1483 #ifdef CONFIG_FS_POSIX_ACL
1484 case LL_IOC_RMTACL: {
1485 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1486 inode == inode->i_sb->s_root->d_inode) {
1487 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1489 LASSERT(fd != NULL);
1490 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1492 fd->fd_flags |= LL_FILE_RMTACL;
1498 case LL_IOC_GETOBDCOUNT: {
1500 struct obd_export *exp;
1502 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1505 /* get ost count when count is zero, get mdt count otherwise */
1506 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1507 vallen = sizeof(count);
1508 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1509 KEY_TGT_COUNT, &vallen, &count);
1511 CERROR("get target count failed: %d\n", rc);
1515 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1520 case LL_IOC_PATH2FID:
1521 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1522 sizeof(struct lu_fid)))
1525 case LL_IOC_GET_CONNECT_FLAGS: {
1526 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1527 (void __user *)arg));
1529 case OBD_IOC_CHANGELOG_SEND:
1530 case OBD_IOC_CHANGELOG_CLEAR:
1531 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1534 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1535 sizeof(struct ioc_changelog));
1537 case OBD_IOC_FID2PATH:
1538 RETURN(ll_fid2path(inode, (void __user *)arg));
1539 case LL_IOC_GETPARENT:
1540 RETURN(ll_getparent(file, (void __user *)arg));
1541 case LL_IOC_FID2MDTIDX: {
1542 struct obd_export *exp = ll_i2mdexp(inode);
1546 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1550 /* Call mdc_iocontrol */
1551 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1552 (__u32 __user *)&index);
1558 case LL_IOC_HSM_REQUEST: {
1559 struct hsm_user_request *hur;
1566 /* We don't know the true size yet; copy the fixed-size part */
1567 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1572 /* Compute the whole struct size */
1573 totalsize = hur_len(hur);
1578 /* Final size will be more than double totalsize */
1579 if (totalsize >= MDS_MAXREQSIZE / 3)
1582 OBD_ALLOC_LARGE(hur, totalsize);
1586 /* Copy the whole struct */
1587 if (copy_from_user(hur, (void __user *)arg, totalsize)) {
1588 OBD_FREE_LARGE(hur, totalsize);
1592 if (hur->hur_request.hr_action == HUA_RELEASE) {
1593 const struct lu_fid *fid;
1597 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1598 fid = &hur->hur_user_item[i].hui_fid;
1599 f = search_inode_for_lustre(inode->i_sb, fid);
1605 rc = ll_hsm_release(f);
1611 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1615 OBD_FREE_LARGE(hur, totalsize);
1619 case LL_IOC_HSM_PROGRESS: {
1620 struct hsm_progress_kernel hpk;
1621 struct hsm_progress hp;
1623 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1626 hpk.hpk_fid = hp.hp_fid;
1627 hpk.hpk_cookie = hp.hp_cookie;
1628 hpk.hpk_extent = hp.hp_extent;
1629 hpk.hpk_flags = hp.hp_flags;
1630 hpk.hpk_errval = hp.hp_errval;
1631 hpk.hpk_data_version = 0;
1633 /* File may not exist in Lustre; all progress
1634 * reported to Lustre root */
1635 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1639 case LL_IOC_HSM_CT_START:
1640 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1643 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1644 sizeof(struct lustre_kernelcomm));
1647 case LL_IOC_HSM_COPY_START: {
1648 struct hsm_copy *copy;
1651 OBD_ALLOC_PTR(copy);
1654 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1659 rc = ll_ioc_copy_start(inode->i_sb, copy);
1660 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1666 case LL_IOC_HSM_COPY_END: {
1667 struct hsm_copy *copy;
1670 OBD_ALLOC_PTR(copy);
1673 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1678 rc = ll_ioc_copy_end(inode->i_sb, copy);
1679 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1685 case LL_IOC_MIGRATE: {
1687 const char *filename;
1693 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1697 data = (struct obd_ioctl_data *)buf;
1698 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1699 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1700 GOTO(migrate_free, rc = -EINVAL);
1702 filename = data->ioc_inlbuf1;
1703 namelen = data->ioc_inllen1;
1704 /* \0 is packed at the end of filename */
1705 if (namelen < 1 || namelen != strlen(filename) + 1)
1706 GOTO(migrate_free, rc = -EINVAL);
1708 if (data->ioc_inllen2 != sizeof(mdtidx))
1709 GOTO(migrate_free, rc = -EINVAL);
1710 mdtidx = *(int *)data->ioc_inlbuf2;
1712 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1714 obd_ioctl_freedata(buf, len);
1719 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1720 (void __user *)arg));
1724 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1726 struct inode *inode = file->f_mapping->host;
1727 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1728 struct ll_sb_info *sbi = ll_i2sbi(inode);
1729 int api32 = ll_need_32bit_api(sbi);
1730 loff_t ret = -EINVAL;
1733 mutex_lock(&inode->i_mutex);
1738 offset += file->f_pos;
1744 offset += LL_DIR_END_OFF_32BIT;
1746 offset += LL_DIR_END_OFF;
1753 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1754 (!api32 && offset <= LL_DIR_END_OFF))) {
1755 if (offset != file->f_pos) {
1756 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1757 (!api32 && offset == LL_DIR_END_OFF))
1758 fd->lfd_pos = MDS_DIR_END_OFF;
1759 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1760 fd->lfd_pos = offset << 32;
1762 fd->lfd_pos = offset;
1763 file->f_pos = offset;
1764 file->f_version = 0;
1771 mutex_unlock(&inode->i_mutex);
1775 static int ll_dir_open(struct inode *inode, struct file *file)
1778 RETURN(ll_file_open(inode, file));
1781 static int ll_dir_release(struct inode *inode, struct file *file)
1784 RETURN(ll_file_release(inode, file));
1787 const struct file_operations ll_dir_operations = {
1788 .llseek = ll_dir_seek,
1789 .open = ll_dir_open,
1790 .release = ll_dir_release,
1791 .read = generic_read_dir,
1792 #ifdef HAVE_DIR_CONTEXT
1793 .iterate = ll_iterate,
1795 .readdir = ll_readdir,
1797 .unlocked_ioctl = ll_dir_ioctl,