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 <obd_support.h>
52 #include <obd_class.h>
53 #include <uapi/linux/lustre/lustre_ioctl.h>
54 #include <lustre_lib.h>
55 #include <lustre_dlm.h>
56 #include <lustre_fid.h>
57 #include <lustre_kernelcomm.h>
58 #include <lustre_swab.h>
60 #include "llite_internal.h"
63 * (new) readdir implementation overview.
65 * Original lustre readdir implementation cached exact copy of raw directory
66 * pages on the client. These pages were indexed in client page cache by
67 * logical offset in the directory file. This design, while very simple and
68 * intuitive had some inherent problems:
70 * . it implies that byte offset to the directory entry serves as a
71 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
72 * ext3/htree directory entries may move due to splits, and more
75 * . it is incompatible with the design of split directories for cmd3,
76 * that assumes that names are distributed across nodes based on their
77 * hash, and so readdir should be done in hash order.
79 * New readdir implementation does readdir in hash order, and uses hash of a
80 * file name as a telldir/seekdir cookie. This led to number of complications:
82 * . hash is not unique, so it cannot be used to index cached directory
83 * pages on the client (note, that it requires a whole pageful of hash
84 * collided entries to cause two pages to have identical hashes);
86 * . hash is not unique, so it cannot, strictly speaking, be used as an
87 * entry cookie. ext3/htree has the same problem and lustre implementation
88 * mimics their solution: seekdir(hash) positions directory at the first
89 * entry with the given hash.
95 * Client caches directory pages using hash of the first entry as an index. As
96 * noted above hash is not unique, so this solution doesn't work as is:
97 * special processing is needed for "page hash chains" (i.e., sequences of
98 * pages filled with entries all having the same hash value).
100 * First, such chains have to be detected. To this end, server returns to the
101 * client the hash of the first entry on the page next to one returned. When
102 * client detects that this hash is the same as hash of the first entry on the
103 * returned page, page hash collision has to be handled. Pages in the
104 * hash chain, except first one, are termed "overflow pages".
106 * Solution to index uniqueness problem is to not cache overflow
107 * pages. Instead, when page hash collision is detected, all overflow pages
108 * from emerging chain are immediately requested from the server and placed in
109 * a special data structure (struct ll_dir_chain). This data structure is used
110 * by ll_readdir() to process entries from overflow pages. When readdir
111 * invocation finishes, overflow pages are discarded. If page hash collision
112 * chain weren't completely processed, next call to readdir will again detect
113 * page hash collision, again read overflow pages in, process next portion of
114 * entries and again discard the pages. This is not as wasteful as it looks,
115 * because, given reasonable hash, page hash collisions are extremely rare.
117 * 1. directory positioning
119 * When seekdir(hash) is called, original
130 * identification of and access to overflow pages
134 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
135 * a header lu_dirpage which describes the start/end hash, and whether this
136 * page is empty (contains no dir entry) or hash collide with next page.
137 * After client receives reply, several pages will be integrated into dir page
138 * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
139 * lu_dirpage for this integrated page will be adjusted. See
140 * mdc_adjust_dirpages().
143 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
144 __u64 offset, struct ll_dir_chain *chain)
146 struct md_callback cb_op;
150 cb_op.md_blocking_ast = ll_md_blocking_ast;
151 rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
158 void ll_release_page(struct inode *inode, struct page *page,
163 /* Always remove the page for striped dir, because the page is
164 * built from temporarily in LMV layer */
165 if (inode != NULL && S_ISDIR(inode->i_mode) &&
166 ll_i2info(inode)->lli_lsm_md != NULL) {
173 if (likely(page->mapping != NULL))
174 truncate_complete_page(page->mapping, page);
181 * return IF_* type for given lu_dirent entry.
182 * IF_* flag shld be converted to particular OS file type in
183 * platform llite module.
185 static u16 ll_dirent_type_get(struct lu_dirent *ent)
188 struct luda_type *lt;
191 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
192 const unsigned align = sizeof(struct luda_type) - 1;
194 len = le16_to_cpu(ent->lde_namelen);
195 len = (len + align) & ~align;
196 lt = (void *)ent->lde_name + len;
197 type = IFTODT(le16_to_cpu(lt->lt_type));
203 #ifdef HAVE_DIR_CONTEXT
204 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
205 struct dir_context *ctx)
208 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
209 void *cookie, filldir_t filldir)
212 struct ll_sb_info *sbi = ll_i2sbi(inode);
214 bool is_api32 = ll_need_32bit_api(sbi);
215 bool is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
217 struct ll_dir_chain chain;
222 ll_dir_chain_init(&chain);
224 page = ll_get_dir_page(inode, op_data, pos, &chain);
226 while (rc == 0 && !done) {
227 struct lu_dirpage *dp;
228 struct lu_dirent *ent;
237 hash = MDS_DIR_END_OFF;
238 dp = page_address(page);
239 for (ent = lu_dirent_start(dp); ent != NULL && !done;
240 ent = lu_dirent_next(ent)) {
247 hash = le64_to_cpu(ent->lde_hash);
248 if (hash < pos) /* Skip until we find target hash */
251 namelen = le16_to_cpu(ent->lde_namelen);
252 if (namelen == 0) /* Skip dummy record */
255 if (is_api32 && is_hash64)
259 fid_le_to_cpu(&fid, &ent->lde_fid);
260 ino = cl_fid_build_ino(&fid, is_api32);
261 type = ll_dirent_type_get(ent);
262 /* For ll_nfs_get_name_filldir(), it will try to access
263 * 'ent' through 'lde_name', so the parameter 'name'
264 * for 'filldir()' must be part of the 'ent'. */
265 #ifdef HAVE_DIR_CONTEXT
267 done = !dir_emit(ctx, ent->lde_name, namelen, ino,
270 done = filldir(cookie, ent->lde_name, namelen, lhash,
277 ll_release_page(inode, page, false);
281 next = le64_to_cpu(dp->ldp_hash_end);
283 if (pos == MDS_DIR_END_OFF) {
285 * End of directory reached.
288 ll_release_page(inode, page, false);
291 * Normal case: continue to the next
294 ll_release_page(inode, page,
295 le32_to_cpu(dp->ldp_flags) &
298 page = ll_get_dir_page(inode, op_data, pos,
302 #ifdef HAVE_DIR_CONTEXT
307 ll_dir_chain_fini(&chain);
311 #ifdef HAVE_DIR_CONTEXT
312 static int ll_iterate(struct file *filp, struct dir_context *ctx)
314 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
317 struct inode *inode = file_inode(filp);
318 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
319 struct ll_sb_info *sbi = ll_i2sbi(inode);
320 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
321 int api32 = ll_need_32bit_api(sbi);
322 struct md_op_data *op_data;
332 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
333 "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
334 inode, (unsigned long)pos, i_size_read(inode), api32);
336 if (pos == MDS_DIR_END_OFF)
342 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
343 LUSTRE_OPC_ANY, inode);
345 GOTO(out, rc = PTR_ERR(op_data));
347 if (unlikely(op_data->op_mea1 != NULL)) {
348 /* This is only needed for striped dir to fill ..,
349 * see lmv_read_entry */
350 if (file_dentry(filp)->d_parent != NULL &&
351 file_dentry(filp)->d_parent->d_inode != NULL) {
352 __u64 ibits = MDS_INODELOCK_UPDATE;
353 struct inode *parent =
354 file_dentry(filp)->d_parent->d_inode;
356 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
357 op_data->op_fid3 = *ll_inode2fid(parent);
360 /* If it can not find in cache, do lookup .. on the master
362 if (fid_is_zero(&op_data->op_fid3)) {
363 rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
365 ll_finish_md_op_data(op_data);
370 #ifdef HAVE_DIR_CONTEXT
372 rc = ll_dir_read(inode, &pos, op_data, ctx);
375 rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
380 if (pos == MDS_DIR_END_OFF) {
382 pos = LL_DIR_END_OFF_32BIT;
384 pos = LL_DIR_END_OFF;
389 #ifdef HAVE_DIR_CONTEXT
394 ll_finish_md_op_data(op_data);
395 filp->f_version = inode->i_version;
399 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
404 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
405 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
407 struct mgs_send_param *msp;
414 strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
415 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
416 sizeof(struct mgs_send_param), msp, NULL);
418 CERROR("Failed to set parameter: %d\n", rc);
426 * Create striped directory with specified stripe(@lump)
428 * \param[in] dparent the parent of the directory.
429 * \param[in] lump the specified stripes.
430 * \param[in] dirname the name of the directory.
431 * \param[in] mode the specified mode of the directory.
433 * \retval =0 if striped directory is being created successfully.
434 * <0 if the creation is failed.
436 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
437 size_t len, const char *dirname, umode_t mode)
439 struct inode *parent = dparent->d_inode;
440 struct ptlrpc_request *request = NULL;
441 struct md_op_data *op_data;
442 struct ll_sb_info *sbi = ll_i2sbi(parent);
443 struct inode *inode = NULL;
444 struct dentry dentry = {
448 .len = strlen(dirname),
449 .hash = ll_full_name_hash(dparent, dirname,
456 if (unlikely(lump->lum_magic != LMV_USER_MAGIC &&
457 lump->lum_magic != LMV_USER_MAGIC_SPECIFIC))
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 (IS_DEADDIR(parent) &&
470 !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
473 if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
474 lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
475 lustre_swab_lmv_user_md(lump);
477 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
478 mode &= ~current_umask();
479 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
480 op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
481 strlen(dirname), mode, LUSTRE_OPC_MKDIR,
484 RETURN(PTR_ERR(op_data));
486 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
487 /* selinux_dentry_init_security() uses dentry->d_parent and name
488 * to determine the security context for the file. So our fake
489 * dentry should be real enough for this purpose. */
490 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
491 &op_data->op_file_secctx_name,
492 &op_data->op_file_secctx,
493 &op_data->op_file_secctx_size);
495 GOTO(out_op_data, err);
498 op_data->op_cli_flags |= CLI_SET_MEA;
499 err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
500 from_kuid(&init_user_ns, current_fsuid()),
501 from_kgid(&init_user_ns, current_fsgid()),
502 cfs_curproc_cap_pack(), 0, &request);
504 GOTO(out_request, err);
506 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
508 err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
510 GOTO(out_inode, err);
512 dentry.d_inode = inode;
514 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
516 err = security_inode_notifysecctx(inode,
517 op_data->op_file_secctx,
518 op_data->op_file_secctx_size);
521 err = ll_inode_init_security(&dentry, inode, parent);
524 GOTO(out_inode, err);
530 ptlrpc_req_finished(request);
532 ll_finish_md_op_data(op_data);
537 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
540 struct ll_sb_info *sbi = ll_i2sbi(inode);
541 struct md_op_data *op_data;
542 struct ptlrpc_request *req = NULL;
544 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
545 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
546 struct obd_device *mgc = lsi->lsi_mgc;
553 * This is coming from userspace, so should be in
554 * local endian. But the MDS would like it in little
555 * endian, so we swab it before we send it.
557 switch (lump->lmm_magic) {
558 case LOV_USER_MAGIC_V1: {
559 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
560 lustre_swab_lov_user_md_v1(lump);
561 lum_size = sizeof(struct lov_user_md_v1);
564 case LOV_USER_MAGIC_V3: {
565 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
566 lustre_swab_lov_user_md_v3(
567 (struct lov_user_md_v3 *)lump);
568 lum_size = sizeof(struct lov_user_md_v3);
571 case LOV_USER_MAGIC_COMP_V1: {
572 if (lump->lmm_magic !=
573 cpu_to_le32(LOV_USER_MAGIC_COMP_V1))
574 lustre_swab_lov_comp_md_v1(
575 (struct lov_comp_md_v1 *)lump);
576 lum_size = le32_to_cpu(
577 ((struct lov_comp_md_v1 *)lump)->lcm_size);
580 case LMV_USER_MAGIC: {
581 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
582 lustre_swab_lmv_user_md(
583 (struct lmv_user_md *)lump);
584 lum_size = sizeof(struct lmv_user_md);
588 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
589 " %#08x != %#08x nor %#08x\n",
590 lump->lmm_magic, LOV_USER_MAGIC_V1,
596 lum_size = sizeof(struct lov_user_md_v1);
599 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
600 LUSTRE_OPC_ANY, NULL);
602 RETURN(PTR_ERR(op_data));
604 /* swabbing is done in lov_setstripe() on server side */
605 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
606 ll_finish_md_op_data(op_data);
607 ptlrpc_req_finished(req);
611 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
613 * 2.9 server has stored filesystem default stripe in ROOT xattr,
614 * and it's stored into system config for backward compatibility.
616 * In the following we use the fact that LOV_USER_MAGIC_V1 and
617 * LOV_USER_MAGIC_V3 have the same initial fields so we do not
618 * need the make the distiction between the 2 versions
620 if (set_default && mgc->u.cli.cl_mgc_mgsexp &&
622 le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V1 ||
623 le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V3)) {
627 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
629 GOTO(end, rc = -ENOMEM);
632 /* Get fsname and assume devname to be -MDT0000. */
633 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
634 strcat(buf, "-MDT0000.lov");
637 /* Set root stripesize */
638 snprintf(buf, MGS_PARAM_MAXLEN, ".stripesize=%u",
639 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
640 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
644 /* Set root stripecount */
645 snprintf(buf, MGS_PARAM_MAXLEN, ".stripecount=%hd",
646 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
647 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
651 /* Set root stripeoffset */
652 snprintf(buf, MGS_PARAM_MAXLEN, ".stripeoffset=%hd",
653 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
654 (typeof(lump->lmm_stripe_offset))(-1));
655 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
659 OBD_FREE(param, MGS_PARAM_MAXLEN);
666 * This function will be used to get default LOV/LMV/Default LMV
667 * @valid will be used to indicate which stripe it will retrieve
668 * OBD_MD_MEA LMV stripe EA
669 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
670 * otherwise Default LOV EA.
671 * Each time, it can only retrieve 1 stripe EA
673 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
674 struct ptlrpc_request **request, u64 valid)
676 struct ll_sb_info *sbi = ll_i2sbi(inode);
677 struct mdt_body *body;
678 struct lov_mds_md *lmm = NULL;
679 struct ptlrpc_request *req = NULL;
681 struct md_op_data *op_data;
684 rc = ll_get_default_mdsize(sbi, &lmm_size);
688 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
689 0, lmm_size, LUSTRE_OPC_ANY,
692 RETURN(PTR_ERR(op_data));
694 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
695 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
696 ll_finish_md_op_data(op_data);
698 CDEBUG(D_INFO, "md_getattr failed on inode "
699 DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
703 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
704 LASSERT(body != NULL);
706 lmm_size = body->mbo_eadatasize;
708 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
710 GOTO(out, rc = -ENODATA);
713 lmm = req_capsule_server_sized_get(&req->rq_pill,
714 &RMF_MDT_MD, lmm_size);
715 LASSERT(lmm != NULL);
718 * This is coming from the MDS, so is probably in
719 * little endian. We convert it to host endian before
720 * passing it to userspace.
722 /* We don't swab objects for directories */
723 switch (le32_to_cpu(lmm->lmm_magic)) {
725 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
726 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
729 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
730 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
732 case LOV_MAGIC_COMP_V1:
733 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
734 lustre_swab_lov_comp_md_v1(
735 (struct lov_comp_md_v1 *)lmm);
738 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
739 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
742 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
743 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
746 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
751 *plmm_size = lmm_size;
756 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
758 struct md_op_data *op_data;
763 OBD_ALLOC_PTR(op_data);
767 op_data->op_flags |= MF_GET_MDT_IDX;
768 op_data->op_fid1 = *fid;
769 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
770 mdt_index = op_data->op_mds;
771 OBD_FREE_PTR(op_data);
779 * Get MDT index for the inode.
781 int ll_get_mdt_idx(struct inode *inode)
783 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
787 * Generic handler to do any pre-copy work.
789 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
790 * first information for it that real work has started.
792 * Moreover, for a ARCHIVE request, it will sample the file data version and
793 * store it in \a copy.
795 * \return 0 on success.
797 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
799 struct ll_sb_info *sbi = ll_s2sbi(sb);
800 struct hsm_progress_kernel hpk;
805 /* Forge a hsm_progress based on data from copy. */
806 hpk.hpk_fid = copy->hc_hai.hai_fid;
807 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
808 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
809 hpk.hpk_extent.length = 0;
812 hpk.hpk_data_version = 0;
815 /* For archive request, we need to read the current file version. */
816 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
818 __u64 data_version = 0;
820 /* Get inode for this fid */
821 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
823 hpk.hpk_flags |= HP_FLAG_RETRY;
824 /* hpk_errval is >= 0 */
825 hpk.hpk_errval = -PTR_ERR(inode);
826 GOTO(progress, rc = PTR_ERR(inode));
829 /* Read current file data version */
830 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
833 CDEBUG(D_HSM, "Could not read file data version of "
834 DFID" (rc = %d). Archive request ("
835 "%#llx) could not be done.\n",
836 PFID(©->hc_hai.hai_fid), rc,
837 copy->hc_hai.hai_cookie);
838 hpk.hpk_flags |= HP_FLAG_RETRY;
839 /* hpk_errval must be >= 0 */
840 hpk.hpk_errval = -rc;
844 /* Store in the hsm_copy for later copytool use.
845 * Always modified even if no lsm. */
846 copy->hc_data_version = data_version;
850 /* On error, the request should be considered as completed */
851 if (hpk.hpk_errval > 0)
852 hpk.hpk_flags |= HP_FLAG_COMPLETED;
854 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
857 /* Return first error */
858 RETURN(rc != 0 ? rc : rc2);
862 * Generic handler to do any post-copy work.
864 * It will send the last hsm_progress update to coordinator to inform it
865 * that copy is finished and whether it was successful or not.
868 * - for ARCHIVE request, it will sample the file data version and compare it
869 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
870 * will be considered as failed.
871 * - for RESTORE request, it will sample the file data version and send it to
872 * coordinator which is useful if the file was imported as 'released'.
874 * \return 0 on success.
876 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
878 struct ll_sb_info *sbi = ll_s2sbi(sb);
879 struct hsm_progress_kernel hpk;
884 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
885 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
886 * initialized if copy_end was called with copy == NULL.
889 /* Forge a hsm_progress based on data from copy. */
890 hpk.hpk_fid = copy->hc_hai.hai_fid;
891 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
892 hpk.hpk_extent = copy->hc_hai.hai_extent;
893 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
894 hpk.hpk_errval = copy->hc_errval;
895 hpk.hpk_data_version = 0;
897 /* For archive request, we need to check the file data was not changed.
899 * For restore request, we need to send the file data version, this is
900 * useful when the file was created using hsm_import.
902 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
903 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
904 (copy->hc_errval == 0)) {
906 __u64 data_version = 0;
908 /* Get lsm for this fid */
909 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
911 hpk.hpk_flags |= HP_FLAG_RETRY;
912 /* hpk_errval must be >= 0 */
913 hpk.hpk_errval = -PTR_ERR(inode);
914 GOTO(progress, rc = PTR_ERR(inode));
917 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
920 CDEBUG(D_HSM, "Could not read file data version. "
921 "Request could not be confirmed.\n");
922 if (hpk.hpk_errval == 0)
923 hpk.hpk_errval = -rc;
927 /* Store in the hsm_copy for later copytool use.
928 * Always modified even if no lsm. */
929 hpk.hpk_data_version = data_version;
931 /* File could have been stripped during archiving, so we need
932 * to check anyway. */
933 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
934 (copy->hc_data_version != data_version)) {
935 CDEBUG(D_HSM, "File data version mismatched. "
936 "File content was changed during archiving. "
937 DFID", start:%#llx current:%#llx\n",
938 PFID(©->hc_hai.hai_fid),
939 copy->hc_data_version, data_version);
940 /* File was changed, send error to cdt. Do not ask for
941 * retry because if a file is modified frequently,
942 * the cdt will loop on retried archive requests.
943 * The policy engine will ask for a new archive later
944 * when the file will not be modified for some tunable
946 hpk.hpk_flags &= ~HP_FLAG_RETRY;
948 /* hpk_errval must be >= 0 */
949 hpk.hpk_errval = -rc;
956 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
959 /* Return first error */
960 RETURN(rc != 0 ? rc : rc2);
964 static int copy_and_ioctl(int cmd, struct obd_export *exp,
965 const void __user *data, size_t size)
970 OBD_ALLOC(copy, size);
974 if (copy_from_user(copy, data, size)) {
979 rc = obd_iocontrol(cmd, exp, size, copy, NULL);
981 OBD_FREE(copy, size);
986 static int check_owner(int type, int id)
990 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
994 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1003 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1005 int cmd = qctl->qc_cmd;
1006 int type = qctl->qc_type;
1007 int id = qctl->qc_id;
1008 int valid = qctl->qc_valid;
1015 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1019 if (check_owner(type, id) &&
1020 (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1026 CERROR("unsupported quotactl op: %#x\n", cmd);
1030 if (valid != QC_GENERAL) {
1031 if (cmd == Q_GETINFO)
1032 qctl->qc_cmd = Q_GETOINFO;
1033 else if (cmd == Q_GETQUOTA)
1034 qctl->qc_cmd = Q_GETOQUOTA;
1040 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1041 sizeof(*qctl), qctl, NULL);
1044 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1045 sizeof(*qctl), qctl, NULL);
1048 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1049 sizeof(*qctl), qctl, NULL);
1051 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1053 sizeof(*qctl), qctl, NULL);
1065 struct obd_quotactl *oqctl;
1067 OBD_ALLOC_PTR(oqctl);
1071 QCTL_COPY(oqctl, qctl);
1072 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1074 OBD_FREE_PTR(oqctl);
1077 /* If QIF_SPACE is not set, client should collect the
1078 * space usage from OSSs by itself */
1079 if (cmd == Q_GETQUOTA &&
1080 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1081 !oqctl->qc_dqblk.dqb_curspace) {
1082 struct obd_quotactl *oqctl_tmp;
1084 OBD_ALLOC_PTR(oqctl_tmp);
1085 if (oqctl_tmp == NULL)
1086 GOTO(out, rc = -ENOMEM);
1088 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1089 oqctl_tmp->qc_id = oqctl->qc_id;
1090 oqctl_tmp->qc_type = oqctl->qc_type;
1092 /* collect space usage from OSTs */
1093 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1094 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1095 if (!rc || rc == -EREMOTEIO) {
1096 oqctl->qc_dqblk.dqb_curspace =
1097 oqctl_tmp->qc_dqblk.dqb_curspace;
1098 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1101 /* collect space & inode usage from MDTs */
1102 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1103 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1104 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1105 if (!rc || rc == -EREMOTEIO) {
1106 oqctl->qc_dqblk.dqb_curspace +=
1107 oqctl_tmp->qc_dqblk.dqb_curspace;
1108 oqctl->qc_dqblk.dqb_curinodes =
1109 oqctl_tmp->qc_dqblk.dqb_curinodes;
1110 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1112 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1115 OBD_FREE_PTR(oqctl_tmp);
1118 QCTL_COPY(qctl, oqctl);
1119 OBD_FREE_PTR(oqctl);
1125 /* This function tries to get a single name component,
1126 * to send to the server. No actual path traversal involved,
1127 * so we limit to NAME_MAX */
1128 static char *ll_getname(const char __user *filename)
1133 OBD_ALLOC(tmp, NAME_MAX + 1);
1136 return ERR_PTR(-ENOMEM);
1138 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1141 else if (len > NAME_MAX)
1142 ret = -ENAMETOOLONG;
1145 OBD_FREE(tmp, NAME_MAX + 1);
1151 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1153 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1155 struct dentry *dentry = file_dentry(file);
1156 struct inode *inode = file_inode(file);
1157 struct ll_sb_info *sbi = ll_i2sbi(inode);
1158 struct obd_ioctl_data *data;
1162 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1163 PFID(ll_inode2fid(inode)), inode, cmd);
1165 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1166 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1169 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1171 case FSFILT_IOC_GETFLAGS:
1172 case FSFILT_IOC_SETFLAGS:
1173 RETURN(ll_iocontrol(inode, file, cmd, arg));
1174 case FSFILT_IOC_GETVERSION_OLD:
1175 case FSFILT_IOC_GETVERSION:
1176 RETURN(put_user(inode->i_generation, (int __user *)arg));
1177 /* We need to special case any other ioctls we want to handle,
1178 * to send them to the MDS/OST as appropriate and to properly
1179 * network encode the arg field.
1180 case FSFILT_IOC_SETVERSION_OLD:
1181 case FSFILT_IOC_SETVERSION:
1183 case LL_IOC_GET_MDTIDX: {
1186 mdtidx = ll_get_mdt_idx(inode);
1190 if (put_user((int)mdtidx, (int __user *)arg))
1195 case IOC_MDC_LOOKUP: {
1196 int namelen, len = 0;
1200 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1205 filename = data->ioc_inlbuf1;
1206 namelen = strlen(filename);
1208 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1209 GOTO(out_free, rc = -EINVAL);
1212 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1214 CERROR("%s: lookup %.*s failed: rc = %d\n",
1215 ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1220 OBD_FREE_LARGE(buf, len);
1223 case LL_IOC_LMV_SETSTRIPE: {
1224 struct lmv_user_md *lum;
1233 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1238 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1239 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1240 GOTO(lmv_out_free, rc = -EINVAL);
1242 filename = data->ioc_inlbuf1;
1243 namelen = data->ioc_inllen1;
1246 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1247 GOTO(lmv_out_free, rc = -EINVAL);
1249 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1250 lumlen = data->ioc_inllen2;
1252 if ((lum->lum_magic != LMV_USER_MAGIC &&
1253 lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) ||
1254 lumlen < sizeof(*lum)) {
1255 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1256 filename, lum->lum_magic, lumlen, -EFAULT);
1257 GOTO(lmv_out_free, rc = -EINVAL);
1260 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1261 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1263 mode = data->ioc_type;
1265 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1267 OBD_FREE_LARGE(buf, len);
1271 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1272 struct lmv_user_md lum;
1273 struct lmv_user_md __user *ulump =
1274 (struct lmv_user_md __user *)arg;
1277 if (copy_from_user(&lum, ulump, sizeof(lum)))
1280 if (lum.lum_magic != LMV_USER_MAGIC)
1283 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1287 case LL_IOC_LOV_SETSTRIPE_NEW:
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 CLASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1301 CLASSERT(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_LOV_GETSTRIPE_NEW:
1441 case LL_IOC_MDC_GETINFO:
1442 case IOC_MDC_GETFILEINFO:
1443 case IOC_MDC_GETFILESTRIPE: {
1444 struct ptlrpc_request *request = NULL;
1445 struct lov_user_md __user *lump;
1446 struct lov_mds_md *lmm = NULL;
1447 struct mdt_body *body;
1448 char *filename = NULL;
1451 if (cmd == IOC_MDC_GETFILEINFO ||
1452 cmd == IOC_MDC_GETFILESTRIPE) {
1453 filename = ll_getname((const char __user *)arg);
1454 if (IS_ERR(filename))
1455 RETURN(PTR_ERR(filename));
1457 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1458 &lmmsize, &request);
1460 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1465 body = req_capsule_server_get(&request->rq_pill,
1467 LASSERT(body != NULL);
1473 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1474 cmd == LL_IOC_MDC_GETINFO))
1475 GOTO(skip_lmm, rc = 0);
1480 if (cmd == IOC_MDC_GETFILESTRIPE ||
1481 cmd == LL_IOC_LOV_GETSTRIPE ||
1482 cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1483 lump = (struct lov_user_md __user *)arg;
1485 struct lov_user_mds_data __user *lmdp;
1486 lmdp = (struct lov_user_mds_data __user *)arg;
1487 lump = &lmdp->lmd_lmm;
1489 if (copy_to_user(lump, lmm, lmmsize)) {
1490 if (copy_to_user(lump, lmm, sizeof(*lump)))
1491 GOTO(out_req, rc = -EFAULT);
1495 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1496 struct lov_user_mds_data __user *lmdp;
1499 st.st_dev = inode->i_sb->s_dev;
1500 st.st_mode = body->mbo_mode;
1501 st.st_nlink = body->mbo_nlink;
1502 st.st_uid = body->mbo_uid;
1503 st.st_gid = body->mbo_gid;
1504 st.st_rdev = body->mbo_rdev;
1505 st.st_size = body->mbo_size;
1506 st.st_blksize = PAGE_SIZE;
1507 st.st_blocks = body->mbo_blocks;
1508 st.st_atime = body->mbo_atime;
1509 st.st_mtime = body->mbo_mtime;
1510 st.st_ctime = body->mbo_ctime;
1511 st.st_ino = cl_fid_build_ino(&body->mbo_fid1,
1515 lmdp = (struct lov_user_mds_data __user *)arg;
1516 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1517 GOTO(out_req, rc = -EFAULT);
1522 ptlrpc_req_finished(request);
1524 ll_putname(filename);
1527 case OBD_IOC_QUOTACTL: {
1528 struct if_quotactl *qctl;
1530 OBD_ALLOC_PTR(qctl);
1534 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1535 GOTO(out_quotactl, rc = -EFAULT);
1537 rc = quotactl_ioctl(sbi, qctl);
1540 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1547 case OBD_IOC_GETDTNAME:
1548 case OBD_IOC_GETMDNAME:
1549 RETURN(ll_get_obd_name(inode, cmd, arg));
1550 case LL_IOC_FLUSHCTX:
1551 RETURN(ll_flush_ctx(inode));
1552 case LL_IOC_GETOBDCOUNT: {
1554 struct obd_export *exp;
1556 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1559 /* get ost count when count is zero, get mdt count otherwise */
1560 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1561 vallen = sizeof(count);
1562 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1563 KEY_TGT_COUNT, &vallen, &count);
1565 CERROR("get target count failed: %d\n", rc);
1569 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1574 case LL_IOC_PATH2FID:
1575 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1576 sizeof(struct lu_fid)))
1579 case LL_IOC_GET_CONNECT_FLAGS: {
1580 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1581 (void __user *)arg));
1583 case OBD_IOC_FID2PATH:
1584 RETURN(ll_fid2path(inode, (void __user *)arg));
1585 case LL_IOC_GETPARENT:
1586 RETURN(ll_getparent(file, (void __user *)arg));
1587 case LL_IOC_FID2MDTIDX: {
1588 struct obd_export *exp = ll_i2mdexp(inode);
1592 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1596 /* Call mdc_iocontrol */
1597 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1598 (__u32 __user *)&index);
1604 case LL_IOC_HSM_REQUEST: {
1605 struct hsm_user_request *hur;
1612 /* We don't know the true size yet; copy the fixed-size part */
1613 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1618 /* Compute the whole struct size */
1619 totalsize = hur_len(hur);
1624 /* Final size will be more than double totalsize */
1625 if (totalsize >= MDS_MAXREQSIZE / 3)
1628 OBD_ALLOC_LARGE(hur, totalsize);
1632 /* Copy the whole struct */
1633 if (copy_from_user(hur, (void __user *)arg, totalsize))
1634 GOTO(out_hur, rc = -EFAULT);
1636 if (hur->hur_request.hr_action == HUA_RELEASE) {
1637 const struct lu_fid *fid;
1641 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1642 fid = &hur->hur_user_item[i].hui_fid;
1643 f = search_inode_for_lustre(inode->i_sb, fid);
1649 rc = ll_hsm_release(f);
1655 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1660 OBD_FREE_LARGE(hur, totalsize);
1664 case LL_IOC_HSM_PROGRESS: {
1665 struct hsm_progress_kernel hpk;
1666 struct hsm_progress hp;
1668 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1671 hpk.hpk_fid = hp.hp_fid;
1672 hpk.hpk_cookie = hp.hp_cookie;
1673 hpk.hpk_extent = hp.hp_extent;
1674 hpk.hpk_flags = hp.hp_flags;
1675 hpk.hpk_errval = hp.hp_errval;
1676 hpk.hpk_data_version = 0;
1678 /* File may not exist in Lustre; all progress
1679 * reported to Lustre root */
1680 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1684 case LL_IOC_HSM_CT_START:
1685 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1688 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1689 sizeof(struct lustre_kernelcomm));
1692 case LL_IOC_HSM_COPY_START: {
1693 struct hsm_copy *copy;
1696 OBD_ALLOC_PTR(copy);
1699 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1704 rc = ll_ioc_copy_start(inode->i_sb, copy);
1705 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1711 case LL_IOC_HSM_COPY_END: {
1712 struct hsm_copy *copy;
1715 OBD_ALLOC_PTR(copy);
1718 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1723 rc = ll_ioc_copy_end(inode->i_sb, copy);
1724 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1730 case LL_IOC_MIGRATE: {
1732 const char *filename;
1738 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1742 data = (struct obd_ioctl_data *)buf;
1743 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1744 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1745 GOTO(migrate_free, rc = -EINVAL);
1747 filename = data->ioc_inlbuf1;
1748 namelen = data->ioc_inllen1;
1749 /* \0 is packed at the end of filename */
1750 if (namelen < 1 || namelen != strlen(filename) + 1)
1751 GOTO(migrate_free, rc = -EINVAL);
1753 if (data->ioc_inllen2 != sizeof(mdtidx))
1754 GOTO(migrate_free, rc = -EINVAL);
1755 mdtidx = *(int *)data->ioc_inlbuf2;
1757 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1759 OBD_FREE_LARGE(buf, len);
1763 case LL_IOC_FSGETXATTR:
1764 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
1765 case LL_IOC_FSSETXATTR:
1766 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
1768 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1769 (void __user *)arg));
1773 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1775 struct inode *inode = file->f_mapping->host;
1776 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1777 struct ll_sb_info *sbi = ll_i2sbi(inode);
1778 int api32 = ll_need_32bit_api(sbi);
1779 loff_t ret = -EINVAL;
1787 offset += file->f_pos;
1793 offset += LL_DIR_END_OFF_32BIT;
1795 offset += LL_DIR_END_OFF;
1802 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1803 (!api32 && offset <= LL_DIR_END_OFF))) {
1804 if (offset != file->f_pos) {
1805 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1806 (!api32 && offset == LL_DIR_END_OFF))
1807 fd->lfd_pos = MDS_DIR_END_OFF;
1808 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1809 fd->lfd_pos = offset << 32;
1811 fd->lfd_pos = offset;
1812 file->f_pos = offset;
1813 file->f_version = 0;
1820 inode_unlock(inode);
1824 static int ll_dir_open(struct inode *inode, struct file *file)
1827 RETURN(ll_file_open(inode, file));
1830 static int ll_dir_release(struct inode *inode, struct file *file)
1833 RETURN(ll_file_release(inode, file));
1836 const struct file_operations ll_dir_operations = {
1837 .llseek = ll_dir_seek,
1838 .open = ll_dir_open,
1839 .release = ll_dir_release,
1840 .read = generic_read_dir,
1841 #ifdef HAVE_DIR_CONTEXT
1842 .iterate = ll_iterate,
1844 .readdir = ll_readdir,
1846 .unlocked_ioctl = ll_dir_ioctl,