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, 2016, 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 LOV_USER_MAGIC_COMP_V1: {
574 if (lump->lmm_magic !=
575 cpu_to_le32(LOV_USER_MAGIC_COMP_V1))
576 lustre_swab_lov_comp_md_v1(
577 (struct lov_comp_md_v1 *)lump);
578 lum_size = le32_to_cpu(
579 ((struct lov_comp_md_v1 *)lump)->lcm_size);
582 case LMV_USER_MAGIC: {
583 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
584 lustre_swab_lmv_user_md(
585 (struct lmv_user_md *)lump);
586 lum_size = sizeof(struct lmv_user_md);
590 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
591 " %#08x != %#08x nor %#08x\n",
592 lump->lmm_magic, LOV_USER_MAGIC_V1,
598 lum_size = sizeof(struct lov_user_md_v1);
601 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
602 LUSTRE_OPC_ANY, NULL);
604 RETURN(PTR_ERR(op_data));
606 /* swabbing is done in lov_setstripe() on server side */
607 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
608 ll_finish_md_op_data(op_data);
609 ptlrpc_req_finished(req);
613 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
615 * 2.9 server has stored filesystem default stripe in ROOT xattr,
616 * and it's stored into system config for backward compatibility.
618 * In the following we use the fact that LOV_USER_MAGIC_V1 and
619 * LOV_USER_MAGIC_V3 have the same initial fields so we do not
620 * need the make the distiction between the 2 versions
622 if (set_default && mgc->u.cli.cl_mgc_mgsexp &&
624 le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V1 ||
625 le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V3)) {
629 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
631 GOTO(end, rc = -ENOMEM);
634 /* Get fsname and assume devname to be -MDT0000. */
635 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
636 strcat(buf, "-MDT0000.lov");
639 /* Set root stripesize */
640 snprintf(buf, MGS_PARAM_MAXLEN, ".stripesize=%u",
641 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
642 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
646 /* Set root stripecount */
647 snprintf(buf, MGS_PARAM_MAXLEN, ".stripecount=%hd",
648 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
649 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
653 /* Set root stripeoffset */
654 snprintf(buf, MGS_PARAM_MAXLEN, ".stripeoffset=%hd",
655 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
656 (typeof(lump->lmm_stripe_offset))(-1));
657 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
661 OBD_FREE(param, MGS_PARAM_MAXLEN);
668 * This function will be used to get default LOV/LMV/Default LMV
669 * @valid will be used to indicate which stripe it will retrieve
670 * OBD_MD_MEA LMV stripe EA
671 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
672 * otherwise Default LOV EA.
673 * Each time, it can only retrieve 1 stripe EA
675 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
676 struct ptlrpc_request **request, u64 valid)
678 struct ll_sb_info *sbi = ll_i2sbi(inode);
679 struct mdt_body *body;
680 struct lov_mds_md *lmm = NULL;
681 struct ptlrpc_request *req = NULL;
683 struct md_op_data *op_data;
686 rc = ll_get_default_mdsize(sbi, &lmm_size);
690 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
691 0, lmm_size, LUSTRE_OPC_ANY,
694 RETURN(PTR_ERR(op_data));
696 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
697 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
698 ll_finish_md_op_data(op_data);
700 CDEBUG(D_INFO, "md_getattr failed on inode "
701 DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
705 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
706 LASSERT(body != NULL);
708 lmm_size = body->mbo_eadatasize;
710 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
712 GOTO(out, rc = -ENODATA);
715 lmm = req_capsule_server_sized_get(&req->rq_pill,
716 &RMF_MDT_MD, lmm_size);
717 LASSERT(lmm != NULL);
720 * This is coming from the MDS, so is probably in
721 * little endian. We convert it to host endian before
722 * passing it to userspace.
724 /* We don't swab objects for directories */
725 switch (le32_to_cpu(lmm->lmm_magic)) {
727 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
728 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
731 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
732 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
734 case LOV_MAGIC_COMP_V1:
735 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
736 lustre_swab_lov_comp_md_v1(
737 (struct lov_comp_md_v1 *)lmm);
740 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
741 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
744 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
745 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
748 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
753 *plmm_size = lmm_size;
758 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
760 struct md_op_data *op_data;
765 OBD_ALLOC_PTR(op_data);
769 op_data->op_flags |= MF_GET_MDT_IDX;
770 op_data->op_fid1 = *fid;
771 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
772 mdt_index = op_data->op_mds;
773 OBD_FREE_PTR(op_data);
781 * Get MDT index for the inode.
783 int ll_get_mdt_idx(struct inode *inode)
785 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
789 * Generic handler to do any pre-copy work.
791 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
792 * first information for it that real work has started.
794 * Moreover, for a ARCHIVE request, it will sample the file data version and
795 * store it in \a copy.
797 * \return 0 on success.
799 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
801 struct ll_sb_info *sbi = ll_s2sbi(sb);
802 struct hsm_progress_kernel hpk;
807 /* Forge a hsm_progress based on data from copy. */
808 hpk.hpk_fid = copy->hc_hai.hai_fid;
809 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
810 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
811 hpk.hpk_extent.length = 0;
814 hpk.hpk_data_version = 0;
817 /* For archive request, we need to read the current file version. */
818 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
820 __u64 data_version = 0;
822 /* Get inode for this fid */
823 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
825 hpk.hpk_flags |= HP_FLAG_RETRY;
826 /* hpk_errval is >= 0 */
827 hpk.hpk_errval = -PTR_ERR(inode);
828 GOTO(progress, rc = PTR_ERR(inode));
831 /* Read current file data version */
832 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
835 CDEBUG(D_HSM, "Could not read file data version of "
836 DFID" (rc = %d). Archive request ("
837 "%#llx) could not be done.\n",
838 PFID(©->hc_hai.hai_fid), rc,
839 copy->hc_hai.hai_cookie);
840 hpk.hpk_flags |= HP_FLAG_RETRY;
841 /* hpk_errval must be >= 0 */
842 hpk.hpk_errval = -rc;
846 /* Store in the hsm_copy for later copytool use.
847 * Always modified even if no lsm. */
848 copy->hc_data_version = data_version;
852 /* On error, the request should be considered as completed */
853 if (hpk.hpk_errval > 0)
854 hpk.hpk_flags |= HP_FLAG_COMPLETED;
856 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
859 /* Return first error */
860 RETURN(rc != 0 ? rc : rc2);
864 * Generic handler to do any post-copy work.
866 * It will send the last hsm_progress update to coordinator to inform it
867 * that copy is finished and whether it was successful or not.
870 * - for ARCHIVE request, it will sample the file data version and compare it
871 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
872 * will be considered as failed.
873 * - for RESTORE request, it will sample the file data version and send it to
874 * coordinator which is useful if the file was imported as 'released'.
876 * \return 0 on success.
878 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
880 struct ll_sb_info *sbi = ll_s2sbi(sb);
881 struct hsm_progress_kernel hpk;
886 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
887 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
888 * initialized if copy_end was called with copy == NULL.
891 /* Forge a hsm_progress based on data from copy. */
892 hpk.hpk_fid = copy->hc_hai.hai_fid;
893 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
894 hpk.hpk_extent = copy->hc_hai.hai_extent;
895 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
896 hpk.hpk_errval = copy->hc_errval;
897 hpk.hpk_data_version = 0;
899 /* For archive request, we need to check the file data was not changed.
901 * For restore request, we need to send the file data version, this is
902 * useful when the file was created using hsm_import.
904 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
905 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
906 (copy->hc_errval == 0)) {
908 __u64 data_version = 0;
910 /* Get lsm for this fid */
911 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
913 hpk.hpk_flags |= HP_FLAG_RETRY;
914 /* hpk_errval must be >= 0 */
915 hpk.hpk_errval = -PTR_ERR(inode);
916 GOTO(progress, rc = PTR_ERR(inode));
919 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
922 CDEBUG(D_HSM, "Could not read file data version. "
923 "Request could not be confirmed.\n");
924 if (hpk.hpk_errval == 0)
925 hpk.hpk_errval = -rc;
929 /* Store in the hsm_copy for later copytool use.
930 * Always modified even if no lsm. */
931 hpk.hpk_data_version = data_version;
933 /* File could have been stripped during archiving, so we need
934 * to check anyway. */
935 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
936 (copy->hc_data_version != data_version)) {
937 CDEBUG(D_HSM, "File data version mismatched. "
938 "File content was changed during archiving. "
939 DFID", start:%#llx current:%#llx\n",
940 PFID(©->hc_hai.hai_fid),
941 copy->hc_data_version, data_version);
942 /* File was changed, send error to cdt. Do not ask for
943 * retry because if a file is modified frequently,
944 * the cdt will loop on retried archive requests.
945 * The policy engine will ask for a new archive later
946 * when the file will not be modified for some tunable
948 hpk.hpk_flags &= ~HP_FLAG_RETRY;
950 /* hpk_errval must be >= 0 */
951 hpk.hpk_errval = -rc;
958 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
961 /* Return first error */
962 RETURN(rc != 0 ? rc : rc2);
966 static int copy_and_ioctl(int cmd, struct obd_export *exp,
967 const void __user *data, size_t size)
972 OBD_ALLOC(copy, size);
976 if (copy_from_user(copy, data, size)) {
981 rc = obd_iocontrol(cmd, exp, size, copy, NULL);
983 OBD_FREE(copy, size);
988 static int check_owner(int type, int id)
992 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
996 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1005 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1007 int cmd = qctl->qc_cmd;
1008 int type = qctl->qc_type;
1009 int id = qctl->qc_id;
1010 int valid = qctl->qc_valid;
1017 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1021 if (check_owner(type, id) &&
1022 (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1028 CERROR("unsupported quotactl op: %#x\n", cmd);
1032 if (valid != QC_GENERAL) {
1033 if (cmd == Q_GETINFO)
1034 qctl->qc_cmd = Q_GETOINFO;
1035 else if (cmd == Q_GETQUOTA)
1036 qctl->qc_cmd = Q_GETOQUOTA;
1042 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1043 sizeof(*qctl), qctl, NULL);
1046 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1047 sizeof(*qctl), qctl, NULL);
1050 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1051 sizeof(*qctl), qctl, NULL);
1053 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1055 sizeof(*qctl), qctl, NULL);
1067 struct obd_quotactl *oqctl;
1069 OBD_ALLOC_PTR(oqctl);
1073 QCTL_COPY(oqctl, qctl);
1074 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1076 OBD_FREE_PTR(oqctl);
1079 /* If QIF_SPACE is not set, client should collect the
1080 * space usage from OSSs by itself */
1081 if (cmd == Q_GETQUOTA &&
1082 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1083 !oqctl->qc_dqblk.dqb_curspace) {
1084 struct obd_quotactl *oqctl_tmp;
1086 OBD_ALLOC_PTR(oqctl_tmp);
1087 if (oqctl_tmp == NULL)
1088 GOTO(out, rc = -ENOMEM);
1090 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1091 oqctl_tmp->qc_id = oqctl->qc_id;
1092 oqctl_tmp->qc_type = oqctl->qc_type;
1094 /* collect space usage from OSTs */
1095 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1096 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1097 if (!rc || rc == -EREMOTEIO) {
1098 oqctl->qc_dqblk.dqb_curspace =
1099 oqctl_tmp->qc_dqblk.dqb_curspace;
1100 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1103 /* collect space & inode usage from MDTs */
1104 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1105 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1106 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1107 if (!rc || rc == -EREMOTEIO) {
1108 oqctl->qc_dqblk.dqb_curspace +=
1109 oqctl_tmp->qc_dqblk.dqb_curspace;
1110 oqctl->qc_dqblk.dqb_curinodes =
1111 oqctl_tmp->qc_dqblk.dqb_curinodes;
1112 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1114 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1117 OBD_FREE_PTR(oqctl_tmp);
1120 QCTL_COPY(qctl, oqctl);
1121 OBD_FREE_PTR(oqctl);
1127 /* This function tries to get a single name component,
1128 * to send to the server. No actual path traversal involved,
1129 * so we limit to NAME_MAX */
1130 static char *ll_getname(const char __user *filename)
1135 OBD_ALLOC(tmp, NAME_MAX + 1);
1138 return ERR_PTR(-ENOMEM);
1140 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1143 else if (len > NAME_MAX)
1144 ret = -ENAMETOOLONG;
1147 OBD_FREE(tmp, NAME_MAX + 1);
1153 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1155 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1157 struct dentry *dentry = file_dentry(file);
1158 struct inode *inode = file_inode(file);
1159 struct ll_sb_info *sbi = ll_i2sbi(inode);
1160 struct obd_ioctl_data *data;
1164 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1165 PFID(ll_inode2fid(inode)), inode, cmd);
1167 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1168 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1171 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1173 case FSFILT_IOC_GETFLAGS:
1174 case FSFILT_IOC_SETFLAGS:
1175 RETURN(ll_iocontrol(inode, file, cmd, arg));
1176 case FSFILT_IOC_GETVERSION_OLD:
1177 case FSFILT_IOC_GETVERSION:
1178 RETURN(put_user(inode->i_generation, (int __user *)arg));
1179 /* We need to special case any other ioctls we want to handle,
1180 * to send them to the MDS/OST as appropriate and to properly
1181 * network encode the arg field.
1182 case FSFILT_IOC_SETVERSION_OLD:
1183 case FSFILT_IOC_SETVERSION:
1185 case LL_IOC_GET_MDTIDX: {
1188 mdtidx = ll_get_mdt_idx(inode);
1192 if (put_user((int)mdtidx, (int __user *)arg))
1197 case IOC_MDC_LOOKUP: {
1198 int namelen, len = 0;
1202 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1207 filename = data->ioc_inlbuf1;
1208 namelen = strlen(filename);
1210 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1211 GOTO(out_free, rc = -EINVAL);
1214 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1216 CERROR("%s: lookup %.*s failed: rc = %d\n",
1217 ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1222 obd_ioctl_freedata(buf, len);
1225 case LL_IOC_LMV_SETSTRIPE: {
1226 struct lmv_user_md *lum;
1235 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1240 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1241 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1242 GOTO(lmv_out_free, rc = -EINVAL);
1244 filename = data->ioc_inlbuf1;
1245 namelen = data->ioc_inllen1;
1248 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1249 GOTO(lmv_out_free, rc = -EINVAL);
1251 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1252 lumlen = data->ioc_inllen2;
1254 if (lum->lum_magic != LMV_USER_MAGIC ||
1255 lumlen != sizeof(*lum)) {
1256 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1257 filename, lum->lum_magic, lumlen, -EFAULT);
1258 GOTO(lmv_out_free, rc = -EINVAL);
1261 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1262 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1264 mode = data->ioc_type;
1266 rc = ll_dir_setdirstripe(dentry, lum, filename, mode);
1268 obd_ioctl_freedata(buf, len);
1272 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1273 struct lmv_user_md lum;
1274 struct lmv_user_md __user *ulump =
1275 (struct lmv_user_md __user *)arg;
1278 if (copy_from_user(&lum, ulump, sizeof(lum)))
1281 if (lum.lum_magic != LMV_USER_MAGIC)
1284 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1288 case LL_IOC_LOV_SETSTRIPE: {
1289 struct lov_user_md_v3 lumv3;
1290 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1291 struct lov_user_md_v1 __user *lumv1p =
1292 (struct lov_user_md_v1 __user *)arg;
1293 struct lov_user_md_v3 __user *lumv3p =
1294 (struct lov_user_md_v3 __user *)arg;
1296 int set_default = 0;
1298 CLASSERT(sizeof(struct lov_user_md_v3) >
1299 sizeof(struct lov_comp_md_v1));
1300 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1301 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1302 sizeof(lumv3p->lmm_objects[0]));
1303 /* first try with v1 which is smaller than v3 */
1304 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1307 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3)
1308 if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1311 if (inode->i_sb->s_root == file_dentry(file))
1314 /* in v1 and v3 cases lumv1 points to data */
1315 rc = ll_dir_setstripe(inode, lumv1, set_default);
1319 case LL_IOC_LMV_GETSTRIPE: {
1320 struct lmv_user_md __user *ulmv =
1321 (struct lmv_user_md __user *)arg;
1322 struct lmv_user_md lum;
1323 struct ptlrpc_request *request = NULL;
1324 union lmv_mds_md *lmm = NULL;
1327 struct lmv_user_md *tmp = NULL;
1331 int max_stripe_count;
1335 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1338 max_stripe_count = lum.lum_stripe_count;
1339 /* lum_magic will indicate which stripe the ioctl will like
1340 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1341 * is for default LMV stripe */
1342 if (lum.lum_magic == LMV_MAGIC_V1)
1343 valid |= OBD_MD_MEA;
1344 else if (lum.lum_magic == LMV_USER_MAGIC)
1345 valid |= OBD_MD_DEFAULT_MEA;
1349 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1352 GOTO(finish_req, rc);
1354 /* Get default LMV EA */
1355 if (lum.lum_magic == LMV_USER_MAGIC) {
1356 if (lmmsize > sizeof(*ulmv))
1357 GOTO(finish_req, rc = -EINVAL);
1359 if (copy_to_user(ulmv, lmm, lmmsize))
1360 GOTO(finish_req, rc = -EFAULT);
1362 GOTO(finish_req, rc);
1365 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1366 if (max_stripe_count < stripe_count) {
1367 lum.lum_stripe_count = stripe_count;
1368 if (copy_to_user(ulmv, &lum, sizeof(lum)))
1369 GOTO(finish_req, rc = -EFAULT);
1370 GOTO(finish_req, rc = -E2BIG);
1373 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1374 OBD_ALLOC(tmp, lum_size);
1376 GOTO(finish_req, rc = -ENOMEM);
1378 mdt_index = ll_get_mdt_idx(inode);
1380 GOTO(out_tmp, rc = -ENOMEM);
1382 tmp->lum_magic = LMV_MAGIC_V1;
1383 tmp->lum_stripe_count = 0;
1384 tmp->lum_stripe_offset = mdt_index;
1385 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1386 for (i = 0; i < stripe_count; i++) {
1389 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1390 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1392 GOTO(out_tmp, rc = mdt_index);
1394 tmp->lum_objects[i].lum_mds = mdt_index;
1395 tmp->lum_objects[i].lum_fid = fid;
1396 tmp->lum_stripe_count++;
1399 if (copy_to_user(ulmv, tmp, lum_size))
1400 GOTO(out_tmp, rc = -EFAULT);
1402 OBD_FREE(tmp, lum_size);
1404 ptlrpc_req_finished(request);
1408 case LL_IOC_REMOVE_ENTRY: {
1409 char *filename = NULL;
1413 /* Here is a little hack to avoid sending REINT_RMENTRY to
1414 * unsupported server, which might crash the server(LU-2730),
1415 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1416 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1417 * server will support REINT_RMENTRY XXX*/
1418 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1421 filename = ll_getname((const char __user *)arg);
1422 if (IS_ERR(filename))
1423 RETURN(PTR_ERR(filename));
1425 namelen = strlen(filename);
1427 GOTO(out_rmdir, rc = -EINVAL);
1429 rc = ll_rmdir_entry(inode, filename, namelen);
1432 ll_putname(filename);
1435 case LL_IOC_LOV_SWAP_LAYOUTS:
1437 case IOC_OBD_STATFS:
1438 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1439 case LL_IOC_LOV_GETSTRIPE:
1440 case LL_IOC_MDC_GETINFO:
1441 case IOC_MDC_GETFILEINFO:
1442 case IOC_MDC_GETFILESTRIPE: {
1443 struct ptlrpc_request *request = NULL;
1444 struct lov_user_md __user *lump;
1445 struct lov_mds_md *lmm = NULL;
1446 struct mdt_body *body;
1447 char *filename = NULL;
1450 if (cmd == IOC_MDC_GETFILEINFO ||
1451 cmd == IOC_MDC_GETFILESTRIPE) {
1452 filename = ll_getname((const char __user *)arg);
1453 if (IS_ERR(filename))
1454 RETURN(PTR_ERR(filename));
1456 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1457 &lmmsize, &request);
1459 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1464 body = req_capsule_server_get(&request->rq_pill,
1466 LASSERT(body != NULL);
1472 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1473 cmd == LL_IOC_MDC_GETINFO))
1474 GOTO(skip_lmm, rc = 0);
1479 if (cmd == IOC_MDC_GETFILESTRIPE ||
1480 cmd == LL_IOC_LOV_GETSTRIPE) {
1481 lump = (struct lov_user_md __user *)arg;
1483 struct lov_user_mds_data __user *lmdp;
1484 lmdp = (struct lov_user_mds_data __user *)arg;
1485 lump = &lmdp->lmd_lmm;
1487 if (copy_to_user(lump, lmm, lmmsize)) {
1488 if (copy_to_user(lump, lmm, sizeof(*lump)))
1489 GOTO(out_req, rc = -EFAULT);
1493 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1494 struct lov_user_mds_data __user *lmdp;
1497 st.st_dev = inode->i_sb->s_dev;
1498 st.st_mode = body->mbo_mode;
1499 st.st_nlink = body->mbo_nlink;
1500 st.st_uid = body->mbo_uid;
1501 st.st_gid = body->mbo_gid;
1502 st.st_rdev = body->mbo_rdev;
1503 st.st_size = body->mbo_size;
1504 st.st_blksize = PAGE_SIZE;
1505 st.st_blocks = body->mbo_blocks;
1506 st.st_atime = body->mbo_atime;
1507 st.st_mtime = body->mbo_mtime;
1508 st.st_ctime = body->mbo_ctime;
1509 st.st_ino = cl_fid_build_ino(&body->mbo_fid1,
1513 lmdp = (struct lov_user_mds_data __user *)arg;
1514 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1515 GOTO(out_req, rc = -EFAULT);
1520 ptlrpc_req_finished(request);
1522 ll_putname(filename);
1525 case OBD_IOC_QUOTACTL: {
1526 struct if_quotactl *qctl;
1528 OBD_ALLOC_PTR(qctl);
1532 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1533 GOTO(out_quotactl, rc = -EFAULT);
1535 rc = quotactl_ioctl(sbi, qctl);
1538 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1545 case OBD_IOC_GETDTNAME:
1546 case OBD_IOC_GETMDNAME:
1547 RETURN(ll_get_obd_name(inode, cmd, arg));
1548 case LL_IOC_FLUSHCTX:
1549 RETURN(ll_flush_ctx(inode));
1550 case LL_IOC_GETOBDCOUNT: {
1552 struct obd_export *exp;
1554 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1557 /* get ost count when count is zero, get mdt count otherwise */
1558 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1559 vallen = sizeof(count);
1560 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1561 KEY_TGT_COUNT, &vallen, &count);
1563 CERROR("get target count failed: %d\n", rc);
1567 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1572 case LL_IOC_PATH2FID:
1573 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1574 sizeof(struct lu_fid)))
1577 case LL_IOC_GET_CONNECT_FLAGS: {
1578 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1579 (void __user *)arg));
1581 case OBD_IOC_FID2PATH:
1582 RETURN(ll_fid2path(inode, (void __user *)arg));
1583 case LL_IOC_GETPARENT:
1584 RETURN(ll_getparent(file, (void __user *)arg));
1585 case LL_IOC_FID2MDTIDX: {
1586 struct obd_export *exp = ll_i2mdexp(inode);
1590 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1594 /* Call mdc_iocontrol */
1595 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1596 (__u32 __user *)&index);
1602 case LL_IOC_HSM_REQUEST: {
1603 struct hsm_user_request *hur;
1610 /* We don't know the true size yet; copy the fixed-size part */
1611 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1616 /* Compute the whole struct size */
1617 totalsize = hur_len(hur);
1622 /* Final size will be more than double totalsize */
1623 if (totalsize >= MDS_MAXREQSIZE / 3)
1626 OBD_ALLOC_LARGE(hur, totalsize);
1630 /* Copy the whole struct */
1631 if (copy_from_user(hur, (void __user *)arg, totalsize))
1632 GOTO(out_hur, rc = -EFAULT);
1634 if (hur->hur_request.hr_action == HUA_RELEASE) {
1635 const struct lu_fid *fid;
1639 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1640 fid = &hur->hur_user_item[i].hui_fid;
1641 f = search_inode_for_lustre(inode->i_sb, fid);
1647 rc = ll_hsm_release(f);
1653 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1658 OBD_FREE_LARGE(hur, totalsize);
1662 case LL_IOC_HSM_PROGRESS: {
1663 struct hsm_progress_kernel hpk;
1664 struct hsm_progress hp;
1666 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1669 hpk.hpk_fid = hp.hp_fid;
1670 hpk.hpk_cookie = hp.hp_cookie;
1671 hpk.hpk_extent = hp.hp_extent;
1672 hpk.hpk_flags = hp.hp_flags;
1673 hpk.hpk_errval = hp.hp_errval;
1674 hpk.hpk_data_version = 0;
1676 /* File may not exist in Lustre; all progress
1677 * reported to Lustre root */
1678 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1682 case LL_IOC_HSM_CT_START:
1683 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1686 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1687 sizeof(struct lustre_kernelcomm));
1690 case LL_IOC_HSM_COPY_START: {
1691 struct hsm_copy *copy;
1694 OBD_ALLOC_PTR(copy);
1697 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1702 rc = ll_ioc_copy_start(inode->i_sb, copy);
1703 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1709 case LL_IOC_HSM_COPY_END: {
1710 struct hsm_copy *copy;
1713 OBD_ALLOC_PTR(copy);
1716 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1721 rc = ll_ioc_copy_end(inode->i_sb, copy);
1722 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1728 case LL_IOC_MIGRATE: {
1730 const char *filename;
1736 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1740 data = (struct obd_ioctl_data *)buf;
1741 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1742 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1743 GOTO(migrate_free, rc = -EINVAL);
1745 filename = data->ioc_inlbuf1;
1746 namelen = data->ioc_inllen1;
1747 /* \0 is packed at the end of filename */
1748 if (namelen < 1 || namelen != strlen(filename) + 1)
1749 GOTO(migrate_free, rc = -EINVAL);
1751 if (data->ioc_inllen2 != sizeof(mdtidx))
1752 GOTO(migrate_free, rc = -EINVAL);
1753 mdtidx = *(int *)data->ioc_inlbuf2;
1755 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1757 obd_ioctl_freedata(buf, len);
1761 case LL_IOC_FSGETXATTR:
1762 case LL_IOC_FSSETXATTR:
1763 RETURN(ll_ioctl_projid(inode, cmd, arg));
1765 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1766 (void __user *)arg));
1770 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1772 struct inode *inode = file->f_mapping->host;
1773 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1774 struct ll_sb_info *sbi = ll_i2sbi(inode);
1775 int api32 = ll_need_32bit_api(sbi);
1776 loff_t ret = -EINVAL;
1784 offset += file->f_pos;
1790 offset += LL_DIR_END_OFF_32BIT;
1792 offset += LL_DIR_END_OFF;
1799 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1800 (!api32 && offset <= LL_DIR_END_OFF))) {
1801 if (offset != file->f_pos) {
1802 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1803 (!api32 && offset == LL_DIR_END_OFF))
1804 fd->lfd_pos = MDS_DIR_END_OFF;
1805 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1806 fd->lfd_pos = offset << 32;
1808 fd->lfd_pos = offset;
1809 file->f_pos = offset;
1810 file->f_version = 0;
1817 inode_unlock(inode);
1821 static int ll_dir_open(struct inode *inode, struct file *file)
1824 RETURN(ll_file_open(inode, file));
1827 static int ll_dir_release(struct inode *inode, struct file *file)
1830 RETURN(ll_file_release(inode, file));
1833 const struct file_operations ll_dir_operations = {
1834 .llseek = ll_dir_seek,
1835 .open = ll_dir_open,
1836 .release = ll_dir_release,
1837 .read = generic_read_dir,
1838 #ifdef HAVE_DIR_CONTEXT
1839 .iterate = ll_iterate,
1841 .readdir = ll_readdir,
1843 .unlocked_ioctl = ll_dir_ioctl,