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, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Directory code for lustre client.
42 #include <linux/pagemap.h>
44 #include <linux/version.h>
45 #include <asm/uaccess.h>
46 #include <linux/buffer_head.h> // for wait_on_buffer
47 #include <linux/pagevec.h>
49 #define DEBUG_SUBSYSTEM S_LLITE
51 #include <lustre/lustre_idl.h>
52 #include <obd_support.h>
53 #include <obd_class.h>
54 #include <lustre_lib.h>
55 #include <lustre/lustre_idl.h>
56 #include <lustre_lite.h>
57 #include <lustre_dlm.h>
58 #include <lustre_fid.h>
59 #include "llite_internal.h"
62 * (new) readdir implementation overview.
64 * Original lustre readdir implementation cached exact copy of raw directory
65 * pages on the client. These pages were indexed in client page cache by
66 * logical offset in the directory file. This design, while very simple and
67 * intuitive had some inherent problems:
69 * . it implies that byte offset to the directory entry serves as a
70 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
71 * ext3/htree directory entries may move due to splits, and more
74 * . it is incompatible with the design of split directories for cmd3,
75 * that assumes that names are distributed across nodes based on their
76 * hash, and so readdir should be done in hash order.
78 * New readdir implementation does readdir in hash order, and uses hash of a
79 * file name as a telldir/seekdir cookie. This led to number of complications:
81 * . hash is not unique, so it cannot be used to index cached directory
82 * pages on the client (note, that it requires a whole pageful of hash
83 * collided entries to cause two pages to have identical hashes);
85 * . hash is not unique, so it cannot, strictly speaking, be used as an
86 * entry cookie. ext3/htree has the same problem and lustre implementation
87 * mimics their solution: seekdir(hash) positions directory at the first
88 * entry with the given hash.
94 * Client caches directory pages using hash of the first entry as an index. As
95 * noted above hash is not unique, so this solution doesn't work as is:
96 * special processing is needed for "page hash chains" (i.e., sequences of
97 * pages filled with entries all having the same hash value).
99 * First, such chains have to be detected. To this end, server returns to the
100 * client the hash of the first entry on the page next to one returned. When
101 * client detects that this hash is the same as hash of the first entry on the
102 * returned page, page hash collision has to be handled. Pages in the
103 * hash chain, except first one, are termed "overflow pages".
105 * Solution to index uniqueness problem is to not cache overflow
106 * pages. Instead, when page hash collision is detected, all overflow pages
107 * from emerging chain are immediately requested from the server and placed in
108 * a special data structure (struct ll_dir_chain). This data structure is used
109 * by ll_readdir() to process entries from overflow pages. When readdir
110 * invocation finishes, overflow pages are discarded. If page hash collision
111 * chain weren't completely processed, next call to readdir will again detect
112 * page hash collision, again read overflow pages in, process next portion of
113 * entries and again discard the pages. This is not as wasteful as it looks,
114 * because, given reasonable hash, page hash collisions are extremely rare.
116 * 1. directory positioning
118 * When seekdir(hash) is called, original
129 * identification of and access to overflow pages
133 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
134 * a header lu_dirpage which describes the start/end hash, and whether this
135 * page is empty (contains no dir entry) or hash collide with next page.
136 * After client receives reply, several pages will be integrated into dir page
137 * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
138 * lu_dirpage for this integrated page will be adjusted. See
139 * lmv_adjust_dirpages().
143 * The following three APIs will be used by llite to iterate directory
144 * entries from MDC dir page caches.
146 * ll_dir_entry_start(next) will lookup(return) entry by op_hash_offset.
147 * To avoid extra memory allocation, the @entry will be pointed to
148 * the dir entries in MDC page directly, so these pages can not be released
149 * until the entry has been accessed in ll_readdir(or statahead).
151 * The iterate process will be
153 * ll_dir_entry_start: locate the page in MDC, and return the first entry.
156 * ll_dir_entry_next: return the next entry in the current page, if it reaches
157 * to the end, release current page.
159 * ll_dir_entry_end: release the last page.
161 struct lu_dirent *ll_dir_entry_start(struct inode *dir,
162 struct md_op_data *op_data,
165 struct lu_dirent *entry;
166 struct md_callback cb_op;
169 LASSERT(*ppage == NULL);
170 cb_op.md_blocking_ast = ll_md_blocking_ast;
171 rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
177 struct lu_dirent *ll_dir_entry_next(struct inode *dir,
178 struct md_op_data *op_data,
179 struct lu_dirent *ent,
182 struct lu_dirent *entry;
183 struct md_callback cb_op;
186 LASSERT(*ppage != NULL);
187 cb_op.md_blocking_ast = ll_md_blocking_ast;
188 op_data->op_hash_offset = le64_to_cpu(ent->lde_hash);
190 page_cache_release(*ppage);
192 rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
198 int ll_dir_read(struct inode *inode, struct md_op_data *op_data,
199 void *cookie, filldir_t filldir)
201 struct ll_sb_info *sbi = ll_i2sbi(inode);
202 struct ll_dir_chain chain;
203 struct lu_dirent *ent;
204 int api32 = ll_need_32bit_api(sbi);
205 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
208 __u64 hash = MDS_DIR_END_OFF;
209 __u64 last_hash = MDS_DIR_END_OFF;
210 struct page *page = NULL;
213 ll_dir_chain_init(&chain);
214 for (ent = ll_dir_entry_start(inode, op_data, &page);
215 ent != NULL && !IS_ERR(ent) && !done;
216 ent = ll_dir_entry_next(inode, op_data, ent, &page)) {
223 hash = le64_to_cpu(ent->lde_hash);
224 if (hash < op_data->op_hash_offset)
226 * Skip until we find target hash
230 namelen = le16_to_cpu(ent->lde_namelen);
241 fid_le_to_cpu(&fid, &ent->lde_fid);
242 ino = cl_fid_build_ino(&fid, api32);
243 type = ll_dirent_type_get(ent);
244 /* For 'll_nfs_get_name_filldir()', it will try
245 * to access the 'ent' through its 'lde_name',
246 * so the parameter 'name' for 'filldir()' must
247 * be part of the 'ent'. */
248 done = filldir(cookie, ent->lde_name, namelen, lhash,
251 if (op_data->op_hash_offset != MDS_DIR_END_OFF)
252 op_data->op_hash_offset = last_hash;
264 page_cache_release(page);
267 ll_dir_chain_fini(&chain);
271 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
273 struct inode *inode = filp->f_dentry->d_inode;
274 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
275 struct ll_sb_info *sbi = ll_i2sbi(inode);
276 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
277 int api32 = ll_need_32bit_api(sbi);
278 struct md_op_data *op_data;
288 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
289 "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
290 inode, (unsigned long)pos, i_size_read(inode), api32);
292 if (pos == MDS_DIR_END_OFF)
298 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
299 LUSTRE_OPC_ANY, inode);
301 GOTO(out, rc = PTR_ERR(op_data));
303 op_data->op_hash_offset = pos;
304 op_data->op_max_pages = sbi->ll_md_brw_size >> PAGE_CACHE_SHIFT;
305 rc = ll_dir_read(inode, op_data, cookie, filldir);
307 lfd->lfd_pos = op_data->op_hash_offset;
309 if (pos == MDS_DIR_END_OFF) {
311 filp->f_pos = LL_DIR_END_OFF_32BIT;
313 filp->f_pos = LL_DIR_END_OFF;
316 filp->f_pos = op_data->op_hash_offset >> 32;
318 filp->f_pos = op_data->op_hash_offset;
321 ll_finish_md_op_data(op_data);
322 filp->f_version = inode->i_version;
326 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
331 int ll_send_mgc_param(struct obd_export *mgc, char *string)
333 struct mgs_send_param *msp;
340 strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
341 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
342 sizeof(struct mgs_send_param), msp, NULL);
344 CERROR("Failed to set parameter: %d\n", rc);
350 static int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
351 const char *filename)
353 struct ptlrpc_request *request = NULL;
354 struct md_op_data *op_data;
355 struct ll_sb_info *sbi = ll_i2sbi(dir);
360 if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
363 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
364 "stripe_offset %d, stripe_count: %u\n",
365 PFID(ll_inode2fid(dir)), dir, filename,
366 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
368 if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
369 lustre_swab_lmv_user_md(lump);
371 mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
372 op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
373 strlen(filename), mode, LUSTRE_OPC_MKDIR,
376 GOTO(err_exit, err = PTR_ERR(op_data));
378 op_data->op_cli_flags |= CLI_SET_MEA;
379 err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
380 current_fsuid(), current_fsgid(),
381 cfs_curproc_cap_pack(), 0, &request);
382 ll_finish_md_op_data(op_data);
386 ptlrpc_req_finished(request);
390 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
393 struct ll_sb_info *sbi = ll_i2sbi(inode);
394 struct md_op_data *op_data;
395 struct ptlrpc_request *req = NULL;
397 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
398 struct obd_device *mgc = lsi->lsi_mgc;
404 * This is coming from userspace, so should be in
405 * local endian. But the MDS would like it in little
406 * endian, so we swab it before we send it.
408 switch (lump->lmm_magic) {
409 case LOV_USER_MAGIC_V1: {
410 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
411 lustre_swab_lov_user_md_v1(lump);
412 lum_size = sizeof(struct lov_user_md_v1);
415 case LOV_USER_MAGIC_V3: {
416 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
417 lustre_swab_lov_user_md_v3(
418 (struct lov_user_md_v3 *)lump);
419 lum_size = sizeof(struct lov_user_md_v3);
422 case LMV_USER_MAGIC: {
423 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
424 lustre_swab_lmv_user_md(
425 (struct lmv_user_md *)lump);
426 lum_size = sizeof(struct lmv_user_md);
430 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
431 " %#08x != %#08x nor %#08x\n",
432 lump->lmm_magic, LOV_USER_MAGIC_V1,
438 lum_size = sizeof(struct lov_user_md_v1);
441 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
442 LUSTRE_OPC_ANY, NULL);
444 RETURN(PTR_ERR(op_data));
446 /* swabbing is done in lov_setstripe() on server side */
447 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
448 NULL, 0, &req, NULL);
449 ll_finish_md_op_data(op_data);
450 ptlrpc_req_finished(req);
452 if (rc != -EPERM && rc != -EACCES)
453 CERROR("mdc_setattr fails: rc = %d\n", rc);
456 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
457 LOV_USER_MAGIC_V3 have the same initial fields so we do not
458 need the make the distiction between the 2 versions */
459 if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
463 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
465 GOTO(end, rc = -ENOMEM);
468 /* Get fsname and assume devname to be -MDT0000. */
469 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
470 strcat(buf, "-MDT0000.lov");
473 /* Set root stripesize */
474 sprintf(buf, ".stripesize=%u",
475 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
476 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
480 /* Set root stripecount */
481 sprintf(buf, ".stripecount=%hd",
482 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
483 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
487 /* Set root stripeoffset */
488 sprintf(buf, ".stripeoffset=%hd",
489 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
490 (typeof(lump->lmm_stripe_offset))(-1));
491 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
495 OBD_FREE(param, MGS_PARAM_MAXLEN);
501 * This function will be used to get default LOV/LMV/Default LMV
502 * @valid will be used to indicate which stripe it will retrieve
503 * OBD_MD_MEA LMV stripe EA
504 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
505 * otherwise Default LOV EA.
506 * Each time, it can only retrieve 1 stripe EA
508 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
509 struct ptlrpc_request **request, obd_valid valid)
511 struct ll_sb_info *sbi = ll_i2sbi(inode);
512 struct mdt_body *body;
513 struct lov_mds_md *lmm = NULL;
514 struct ptlrpc_request *req = NULL;
516 struct md_op_data *op_data;
519 rc = ll_get_max_mdsize(sbi, &lmm_size);
523 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
524 0, lmm_size, LUSTRE_OPC_ANY,
527 RETURN(PTR_ERR(op_data));
529 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
530 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
531 ll_finish_md_op_data(op_data);
533 CDEBUG(D_INFO, "md_getattr failed on inode "
534 DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
538 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
539 LASSERT(body != NULL);
541 lmm_size = body->eadatasize;
543 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
545 GOTO(out, rc = -ENODATA);
548 lmm = req_capsule_server_sized_get(&req->rq_pill,
549 &RMF_MDT_MD, lmm_size);
550 LASSERT(lmm != NULL);
553 * This is coming from the MDS, so is probably in
554 * little endian. We convert it to host endian before
555 * passing it to userspace.
557 /* We don't swab objects for directories */
558 switch (le32_to_cpu(lmm->lmm_magic)) {
560 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
561 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
564 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
565 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
568 case LMV_MAGIC_MIGRATE:
569 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
570 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
573 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
574 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
577 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
582 *plmm_size = lmm_size;
587 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
589 struct md_op_data *op_data;
594 OBD_ALLOC_PTR(op_data);
598 op_data->op_flags |= MF_GET_MDT_IDX;
599 op_data->op_fid1 = *fid;
600 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
601 mdt_index = op_data->op_mds;
602 OBD_FREE_PTR(op_data);
610 * Get MDT index for the inode.
612 int ll_get_mdt_idx(struct inode *inode)
614 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
618 * Generic handler to do any pre-copy work.
620 * It send a first hsm_progress (with extent length == 0) to coordinator as a
621 * first information for it that real work has started.
623 * Moreover, for a ARCHIVE request, it will sample the file data version and
624 * store it in \a copy.
626 * \return 0 on success.
628 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
630 struct ll_sb_info *sbi = ll_s2sbi(sb);
631 struct hsm_progress_kernel hpk;
635 /* Forge a hsm_progress based on data from copy. */
636 hpk.hpk_fid = copy->hc_hai.hai_fid;
637 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
638 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
639 hpk.hpk_extent.length = 0;
642 hpk.hpk_data_version = 0;
645 /* For archive request, we need to read the current file version. */
646 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
648 __u64 data_version = 0;
650 /* Get inode for this fid */
651 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
653 hpk.hpk_flags |= HP_FLAG_RETRY;
654 /* hpk_errval is >= 0 */
655 hpk.hpk_errval = -PTR_ERR(inode);
656 GOTO(progress, rc = PTR_ERR(inode));
659 /* Read current file data version */
660 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
663 CDEBUG(D_HSM, "Could not read file data version of "
664 DFID" (rc = %d). Archive request ("
665 LPX64") could not be done.\n",
666 PFID(©->hc_hai.hai_fid), rc,
667 copy->hc_hai.hai_cookie);
668 hpk.hpk_flags |= HP_FLAG_RETRY;
669 /* hpk_errval must be >= 0 */
670 hpk.hpk_errval = -rc;
674 /* Store it the hsm_copy for later copytool use.
675 * Always modified even if no lsm. */
676 copy->hc_data_version = data_version;
680 /* On error, the request should be considered as completed */
681 if (hpk.hpk_errval > 0)
682 hpk.hpk_flags |= HP_FLAG_COMPLETED;
683 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
690 * Generic handler to do any post-copy work.
692 * It will send the last hsm_progress update to coordinator to inform it
693 * that copy is finished and whether it was successful or not.
696 * - for ARCHIVE request, it will sample the file data version and compare it
697 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
698 * will be considered as failed.
699 * - for RESTORE request, it will sample the file data version and send it to
700 * coordinator which is useful if the file was imported as 'released'.
702 * \return 0 on success.
704 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
706 struct ll_sb_info *sbi = ll_s2sbi(sb);
707 struct hsm_progress_kernel hpk;
711 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
712 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
713 * initialized if copy_end was called with copy == NULL.
716 /* Forge a hsm_progress based on data from copy. */
717 hpk.hpk_fid = copy->hc_hai.hai_fid;
718 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
719 hpk.hpk_extent = copy->hc_hai.hai_extent;
720 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
721 hpk.hpk_errval = copy->hc_errval;
722 hpk.hpk_data_version = 0;
724 /* For archive request, we need to check the file data was not changed.
726 * For restore request, we need to send the file data version, this is
727 * useful when the file was created using hsm_import.
729 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
730 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
731 (copy->hc_errval == 0)) {
733 __u64 data_version = 0;
735 /* Get lsm for this fid */
736 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
738 hpk.hpk_flags |= HP_FLAG_RETRY;
739 /* hpk_errval must be >= 0 */
740 hpk.hpk_errval = -PTR_ERR(inode);
741 GOTO(progress, rc = PTR_ERR(inode));
744 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
747 CDEBUG(D_HSM, "Could not read file data version. "
748 "Request could not be confirmed.\n");
749 if (hpk.hpk_errval == 0)
750 hpk.hpk_errval = -rc;
754 /* Store it the hsm_copy for later copytool use.
755 * Always modified even if no lsm. */
756 hpk.hpk_data_version = data_version;
758 /* File could have been stripped during archiving, so we need
759 * to check anyway. */
760 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
761 (copy->hc_data_version != data_version)) {
762 CDEBUG(D_HSM, "File data version mismatched. "
763 "File content was changed during archiving. "
764 DFID", start:"LPX64" current:"LPX64"\n",
765 PFID(©->hc_hai.hai_fid),
766 copy->hc_data_version, data_version);
767 /* File was changed, send error to cdt. Do not ask for
768 * retry because if a file is modified frequently,
769 * the cdt will loop on retried archive requests.
770 * The policy engine will ask for a new archive later
771 * when the file will not be modified for some tunable
773 /* we do not notify caller */
774 hpk.hpk_flags &= ~HP_FLAG_RETRY;
775 /* hpk_errval must be >= 0 */
776 hpk.hpk_errval = EBUSY;
782 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
789 static int copy_and_ioctl(int cmd, struct obd_export *exp,
790 const void __user *data, size_t size)
795 OBD_ALLOC(copy, size);
799 if (copy_from_user(copy, data, size)) {
804 rc = obd_iocontrol(cmd, exp, size, copy, NULL);
806 OBD_FREE(copy, size);
811 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
813 int cmd = qctl->qc_cmd;
814 int type = qctl->qc_type;
815 int id = qctl->qc_id;
816 int valid = qctl->qc_valid;
821 case LUSTRE_Q_INVALIDATE:
822 case LUSTRE_Q_FINVALIDATE:
827 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
828 sbi->ll_flags & LL_SBI_RMT_CLIENT)
832 if (((type == USRQUOTA && current_euid() != id) ||
833 (type == GRPQUOTA && !in_egroup_p(id))) &&
834 (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
835 sbi->ll_flags & LL_SBI_RMT_CLIENT))
841 CERROR("unsupported quotactl op: %#x\n", cmd);
845 if (valid != QC_GENERAL) {
846 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
849 if (cmd == Q_GETINFO)
850 qctl->qc_cmd = Q_GETOINFO;
851 else if (cmd == Q_GETQUOTA)
852 qctl->qc_cmd = Q_GETOQUOTA;
858 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
859 sizeof(*qctl), qctl, NULL);
862 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
863 sizeof(*qctl), qctl, NULL);
866 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
867 sizeof(*qctl), qctl, NULL);
869 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
871 sizeof(*qctl), qctl, NULL);
883 struct obd_quotactl *oqctl;
885 OBD_ALLOC_PTR(oqctl);
889 QCTL_COPY(oqctl, qctl);
890 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
892 if (rc != -EALREADY && cmd == Q_QUOTAON) {
893 oqctl->qc_cmd = Q_QUOTAOFF;
894 obd_quotactl(sbi->ll_md_exp, oqctl);
899 /* If QIF_SPACE is not set, client should collect the
900 * space usage from OSSs by itself */
901 if (cmd == Q_GETQUOTA &&
902 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
903 !oqctl->qc_dqblk.dqb_curspace) {
904 struct obd_quotactl *oqctl_tmp;
906 OBD_ALLOC_PTR(oqctl_tmp);
907 if (oqctl_tmp == NULL)
908 GOTO(out, rc = -ENOMEM);
910 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
911 oqctl_tmp->qc_id = oqctl->qc_id;
912 oqctl_tmp->qc_type = oqctl->qc_type;
914 /* collect space usage from OSTs */
915 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
916 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
917 if (!rc || rc == -EREMOTEIO) {
918 oqctl->qc_dqblk.dqb_curspace =
919 oqctl_tmp->qc_dqblk.dqb_curspace;
920 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
923 /* collect space & inode usage from MDTs */
924 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
925 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
926 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
927 if (!rc || rc == -EREMOTEIO) {
928 oqctl->qc_dqblk.dqb_curspace +=
929 oqctl_tmp->qc_dqblk.dqb_curspace;
930 oqctl->qc_dqblk.dqb_curinodes =
931 oqctl_tmp->qc_dqblk.dqb_curinodes;
932 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
934 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
937 OBD_FREE_PTR(oqctl_tmp);
940 QCTL_COPY(qctl, oqctl);
948 ll_getname(const char __user *filename)
951 char *tmp = __getname();
954 return ERR_PTR(-ENOMEM);
956 len = strncpy_from_user(tmp, filename, PATH_MAX);
959 else if (len > PATH_MAX)
969 #define ll_putname(filename) __putname(filename)
971 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
973 struct inode *inode = file->f_dentry->d_inode;
974 struct ll_sb_info *sbi = ll_i2sbi(inode);
975 struct obd_ioctl_data *data;
979 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
980 PFID(ll_inode2fid(inode)), inode, cmd);
982 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
983 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
986 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
988 case FSFILT_IOC_GETFLAGS:
989 case FSFILT_IOC_SETFLAGS:
990 RETURN(ll_iocontrol(inode, file, cmd, arg));
991 case FSFILT_IOC_GETVERSION_OLD:
992 case FSFILT_IOC_GETVERSION:
993 RETURN(put_user(inode->i_generation, (int *)arg));
994 /* We need to special case any other ioctls we want to handle,
995 * to send them to the MDS/OST as appropriate and to properly
996 * network encode the arg field.
997 case FSFILT_IOC_SETVERSION_OLD:
998 case FSFILT_IOC_SETVERSION:
1000 case LL_IOC_GET_MDTIDX: {
1003 mdtidx = ll_get_mdt_idx(inode);
1007 if (put_user((int)mdtidx, (int*)arg))
1012 case IOC_MDC_LOOKUP: {
1013 struct ptlrpc_request *request = NULL;
1014 int namelen, len = 0;
1017 struct md_op_data *op_data;
1019 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1024 filename = data->ioc_inlbuf1;
1025 namelen = strlen(filename);
1028 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1029 GOTO(out_free, rc = -EINVAL);
1032 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1033 0, LUSTRE_OPC_ANY, NULL);
1034 if (IS_ERR(op_data))
1035 GOTO(out_free, rc = PTR_ERR(op_data));
1037 op_data->op_valid = OBD_MD_FLID;
1038 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1039 ll_finish_md_op_data(op_data);
1041 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1044 ptlrpc_req_finished(request);
1047 obd_ioctl_freedata(buf, len);
1050 case LL_IOC_LMV_SETSTRIPE: {
1051 struct lmv_user_md *lum;
1059 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1064 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1065 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1066 GOTO(lmv_out_free, rc = -EINVAL);
1068 filename = data->ioc_inlbuf1;
1069 namelen = data->ioc_inllen1;
1072 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1073 GOTO(lmv_out_free, rc = -EINVAL);
1075 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1076 lumlen = data->ioc_inllen2;
1078 if (lum->lum_magic != LMV_USER_MAGIC ||
1079 lumlen != sizeof(*lum)) {
1080 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1081 filename, lum->lum_magic, lumlen, -EFAULT);
1082 GOTO(lmv_out_free, rc = -EINVAL);
1086 * ll_dir_setdirstripe will be used to set dir stripe
1087 * mdc_create--->mdt_reint_create (with dirstripe)
1089 rc = ll_dir_setdirstripe(inode, lum, filename);
1091 obd_ioctl_freedata(buf, len);
1095 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1096 struct lmv_user_md lum;
1097 struct lmv_user_md __user *ulump =
1098 (struct lmv_user_md __user *)arg;
1101 if (copy_from_user(&lum, ulump, sizeof(lum)))
1104 if (lum.lum_magic != LMV_USER_MAGIC)
1107 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1111 case LL_IOC_LOV_SETSTRIPE: {
1112 struct lov_user_md_v3 lumv3;
1113 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1114 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1115 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1117 int set_default = 0;
1119 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1120 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1121 sizeof(lumv3p->lmm_objects[0]));
1122 /* first try with v1 which is smaller than v3 */
1123 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1126 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1127 if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1131 if (inode->i_sb->s_root == file->f_dentry)
1134 /* in v1 and v3 cases lumv1 points to data */
1135 rc = ll_dir_setstripe(inode, lumv1, set_default);
1139 case LL_IOC_LMV_GETSTRIPE: {
1140 struct lmv_user_md __user *ulmv =
1141 (struct lmv_user_md __user *)arg;
1142 struct lmv_user_md lum;
1143 struct ptlrpc_request *request = NULL;
1144 union lmv_mds_md *lmm = NULL;
1146 obd_valid valid = 0;
1147 struct lmv_user_md *tmp = NULL;
1154 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1157 /* lum_magic will indicate which stripe the ioctl will like
1158 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1159 * is for default LMV stripe */
1160 if (lum.lum_magic == LMV_MAGIC_V1)
1161 valid |= OBD_MD_MEA;
1162 else if (lum.lum_magic == LMV_USER_MAGIC)
1163 valid |= OBD_MD_DEFAULT_MEA;
1167 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1169 if (rc != 0 && rc != -ENODATA)
1170 GOTO(finish_req, rc);
1172 /* Get default LMV EA */
1173 if (lum.lum_magic == LMV_USER_MAGIC) {
1175 GOTO(finish_req, rc);
1177 if (lmmsize > sizeof(*ulmv))
1178 GOTO(finish_req, rc = -EINVAL);
1180 if (copy_to_user(ulmv, lmm, lmmsize))
1181 GOTO(finish_req, rc = -EFAULT);
1183 GOTO(finish_req, rc);
1186 /* Get normal LMV EA */
1187 if (rc == -ENODATA) {
1190 LASSERT(lmm != NULL);
1191 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1194 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1195 OBD_ALLOC(tmp, lum_size);
1197 GOTO(finish_req, rc = -ENOMEM);
1199 tmp->lum_magic = LMV_MAGIC_V1;
1200 tmp->lum_stripe_count = 1;
1201 mdt_index = ll_get_mdt_idx(inode);
1203 GOTO(out_tmp, rc = -ENOMEM);
1204 tmp->lum_stripe_offset = mdt_index;
1205 tmp->lum_objects[0].lum_mds = mdt_index;
1206 tmp->lum_objects[0].lum_fid = *ll_inode2fid(inode);
1207 for (i = 1; i < stripe_count; i++) {
1208 struct lmv_mds_md_v1 *lmm1;
1210 lmm1 = &lmm->lmv_md_v1;
1211 mdt_index = ll_get_mdt_idx_by_fid(sbi,
1212 &lmm1->lmv_stripe_fids[i]);
1214 GOTO(out_tmp, rc = mdt_index);
1216 tmp->lum_objects[i].lum_mds = mdt_index;
1217 tmp->lum_objects[i].lum_fid = lmm1->lmv_stripe_fids[i];
1218 tmp->lum_stripe_count++;
1221 if (copy_to_user(ulmv, tmp, lum_size))
1222 GOTO(out_tmp, rc = -EFAULT);
1224 OBD_FREE(tmp, lum_size);
1226 ptlrpc_req_finished(request);
1230 case LL_IOC_REMOVE_ENTRY: {
1231 char *filename = NULL;
1235 /* Here is a little hack to avoid sending REINT_RMENTRY to
1236 * unsupported server, which might crash the server(LU-2730),
1237 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1238 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1239 * server will support REINT_RMENTRY XXX*/
1240 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1243 filename = ll_getname((const char *)arg);
1244 if (IS_ERR(filename))
1245 RETURN(PTR_ERR(filename));
1247 namelen = strlen(filename);
1249 GOTO(out_rmdir, rc = -EINVAL);
1251 rc = ll_rmdir_entry(inode, filename, namelen);
1254 ll_putname(filename);
1257 case LL_IOC_LOV_SWAP_LAYOUTS:
1259 case LL_IOC_OBD_STATFS:
1260 RETURN(ll_obd_statfs(inode, (void *)arg));
1261 case LL_IOC_LOV_GETSTRIPE:
1262 case LL_IOC_MDC_GETINFO:
1263 case IOC_MDC_GETFILEINFO:
1264 case IOC_MDC_GETFILESTRIPE: {
1265 struct ptlrpc_request *request = NULL;
1266 struct lov_user_md *lump;
1267 struct lov_mds_md *lmm = NULL;
1268 struct mdt_body *body;
1269 char *filename = NULL;
1272 if (cmd == IOC_MDC_GETFILEINFO ||
1273 cmd == IOC_MDC_GETFILESTRIPE) {
1274 filename = ll_getname((const char *)arg);
1275 if (IS_ERR(filename))
1276 RETURN(PTR_ERR(filename));
1278 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1279 &lmmsize, &request);
1281 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1286 body = req_capsule_server_get(&request->rq_pill,
1288 LASSERT(body != NULL);
1294 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1295 cmd == LL_IOC_MDC_GETINFO))
1296 GOTO(skip_lmm, rc = 0);
1301 if (cmd == IOC_MDC_GETFILESTRIPE ||
1302 cmd == LL_IOC_LOV_GETSTRIPE) {
1303 lump = (struct lov_user_md *)arg;
1305 struct lov_user_mds_data *lmdp;
1306 lmdp = (struct lov_user_mds_data *)arg;
1307 lump = &lmdp->lmd_lmm;
1309 if (copy_to_user(lump, lmm, lmmsize)) {
1310 if (copy_to_user(lump, lmm, sizeof(*lump)))
1311 GOTO(out_req, rc = -EFAULT);
1315 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1316 struct lov_user_mds_data *lmdp;
1319 st.st_dev = inode->i_sb->s_dev;
1320 st.st_mode = body->mode;
1321 st.st_nlink = body->nlink;
1322 st.st_uid = body->uid;
1323 st.st_gid = body->gid;
1324 st.st_rdev = body->rdev;
1325 st.st_size = body->size;
1326 st.st_blksize = PAGE_CACHE_SIZE;
1327 st.st_blocks = body->blocks;
1328 st.st_atime = body->atime;
1329 st.st_mtime = body->mtime;
1330 st.st_ctime = body->ctime;
1331 st.st_ino = inode->i_ino;
1333 lmdp = (struct lov_user_mds_data *)arg;
1334 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1335 GOTO(out_req, rc = -EFAULT);
1340 ptlrpc_req_finished(request);
1342 ll_putname(filename);
1345 case IOC_LOV_GETINFO: {
1346 struct lov_user_mds_data *lumd;
1347 struct lov_stripe_md *lsm;
1348 struct lov_user_md *lum;
1349 struct lov_mds_md *lmm;
1353 lumd = (struct lov_user_mds_data *)arg;
1354 lum = &lumd->lmd_lmm;
1356 rc = ll_get_max_mdsize(sbi, &lmmsize);
1360 OBD_ALLOC_LARGE(lmm, lmmsize);
1364 if (copy_from_user(lmm, lum, lmmsize))
1365 GOTO(free_lmm, rc = -EFAULT);
1367 switch (lmm->lmm_magic) {
1368 case LOV_USER_MAGIC_V1:
1369 if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1371 /* swab objects first so that stripes num will be sane */
1372 lustre_swab_lov_user_md_objects(
1373 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1374 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1375 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1377 case LOV_USER_MAGIC_V3:
1378 if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1380 /* swab objects first so that stripes num will be sane */
1381 lustre_swab_lov_user_md_objects(
1382 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1383 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1384 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1387 GOTO(free_lmm, rc = -EINVAL);
1390 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1392 GOTO(free_lmm, rc = -ENOMEM);
1394 /* Perform glimpse_size operation. */
1395 memset(&st, 0, sizeof(st));
1397 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1401 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1402 GOTO(free_lsm, rc = -EFAULT);
1406 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1408 OBD_FREE_LARGE(lmm, lmmsize);
1411 case OBD_IOC_LLOG_CATINFO: {
1412 RETURN(-EOPNOTSUPP);
1414 case OBD_IOC_QUOTACHECK: {
1415 struct obd_quotactl *oqctl;
1418 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1419 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1422 OBD_ALLOC_PTR(oqctl);
1425 oqctl->qc_type = arg;
1426 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1428 CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1432 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1434 CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1436 OBD_FREE_PTR(oqctl);
1439 case OBD_IOC_POLL_QUOTACHECK: {
1440 struct if_quotacheck *check;
1442 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1443 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1446 OBD_ALLOC_PTR(check);
1450 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1453 CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1454 if (copy_to_user((void *)arg, check,
1456 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1460 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1463 CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1464 if (copy_to_user((void *)arg, check,
1466 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1470 OBD_FREE_PTR(check);
1473 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1474 case LL_IOC_QUOTACTL_18: {
1475 /* copy the old 1.x quota struct for internal use, then copy
1476 * back into old format struct. For 1.8 compatibility. */
1477 struct if_quotactl_18 *qctl_18;
1478 struct if_quotactl *qctl_20;
1480 OBD_ALLOC_PTR(qctl_18);
1484 OBD_ALLOC_PTR(qctl_20);
1486 GOTO(out_quotactl_18, rc = -ENOMEM);
1488 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1489 GOTO(out_quotactl_20, rc = -ENOMEM);
1491 QCTL_COPY(qctl_20, qctl_18);
1492 qctl_20->qc_idx = 0;
1494 /* XXX: dqb_valid was borrowed as a flag to mark that
1495 * only mds quota is wanted */
1496 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1497 qctl_18->qc_dqblk.dqb_valid) {
1498 qctl_20->qc_valid = QC_MDTIDX;
1499 qctl_20->qc_dqblk.dqb_valid = 0;
1500 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1501 qctl_20->qc_valid = QC_UUID;
1502 qctl_20->obd_uuid = qctl_18->obd_uuid;
1504 qctl_20->qc_valid = QC_GENERAL;
1507 rc = quotactl_ioctl(sbi, qctl_20);
1510 QCTL_COPY(qctl_18, qctl_20);
1511 qctl_18->obd_uuid = qctl_20->obd_uuid;
1513 if (copy_to_user((void *)arg, qctl_18,
1519 OBD_FREE_PTR(qctl_20);
1521 OBD_FREE_PTR(qctl_18);
1525 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1526 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1527 case LL_IOC_QUOTACTL: {
1528 struct if_quotactl *qctl;
1530 OBD_ALLOC_PTR(qctl);
1534 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1535 GOTO(out_quotactl, rc = -EFAULT);
1537 rc = quotactl_ioctl(sbi, qctl);
1539 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1546 case OBD_IOC_GETDTNAME:
1547 case OBD_IOC_GETMDNAME:
1548 RETURN(ll_get_obd_name(inode, cmd, arg));
1549 case LL_IOC_FLUSHCTX:
1550 RETURN(ll_flush_ctx(inode));
1551 #ifdef CONFIG_FS_POSIX_ACL
1552 case LL_IOC_RMTACL: {
1553 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1554 inode == inode->i_sb->s_root->d_inode) {
1555 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1557 LASSERT(fd != NULL);
1558 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1560 fd->fd_flags |= LL_FILE_RMTACL;
1566 case LL_IOC_GETOBDCOUNT: {
1568 struct obd_export *exp;
1570 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1573 /* get ost count when count is zero, get mdt count otherwise */
1574 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1575 vallen = sizeof(count);
1576 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1577 KEY_TGT_COUNT, &vallen, &count, NULL);
1579 CERROR("get target count failed: %d\n", rc);
1583 if (copy_to_user((int *)arg, &count, sizeof(int)))
1588 case LL_IOC_PATH2FID:
1589 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1590 sizeof(struct lu_fid)))
1593 case LL_IOC_GET_CONNECT_FLAGS: {
1594 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1596 case OBD_IOC_CHANGELOG_SEND:
1597 case OBD_IOC_CHANGELOG_CLEAR:
1598 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1599 sizeof(struct ioc_changelog));
1601 case OBD_IOC_FID2PATH:
1602 RETURN(ll_fid2path(inode, (void *)arg));
1603 case LL_IOC_HSM_REQUEST: {
1604 struct hsm_user_request *hur;
1611 /* We don't know the true size yet; copy the fixed-size part */
1612 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1617 /* Compute the whole struct size */
1618 totalsize = hur_len(hur);
1621 /* Make sure the size is reasonable */
1622 if (totalsize >= MDS_MAXREQSIZE)
1625 OBD_ALLOC_LARGE(hur, totalsize);
1629 /* Copy the whole struct */
1630 if (copy_from_user(hur, (void *)arg, totalsize)) {
1631 OBD_FREE_LARGE(hur, totalsize);
1635 if (hur->hur_request.hr_action == HUA_RELEASE) {
1636 const struct lu_fid *fid;
1640 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1641 fid = &hur->hur_user_item[i].hui_fid;
1642 f = search_inode_for_lustre(inode->i_sb, fid);
1648 rc = ll_hsm_release(f);
1654 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 *)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 *)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 *)arg, sizeof(*copy))) {
1702 rc = ll_ioc_copy_start(inode->i_sb, copy);
1703 if (copy_to_user((char *)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 *)arg, sizeof(*copy))) {
1721 rc = ll_ioc_copy_end(inode->i_sb, copy);
1722 if (copy_to_user((char *)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;
1748 GOTO(migrate_free, rc = -EINVAL);
1750 if (data->ioc_inllen2 != sizeof(mdtidx))
1751 GOTO(migrate_free, rc = -EINVAL);
1752 mdtidx = *(int *)data->ioc_inlbuf2;
1754 rc = ll_migrate(inode, file, mdtidx, filename, namelen);
1756 obd_ioctl_freedata(buf, len);
1761 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1766 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1768 struct inode *inode = file->f_mapping->host;
1769 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1770 struct ll_sb_info *sbi = ll_i2sbi(inode);
1771 int api32 = ll_need_32bit_api(sbi);
1772 loff_t ret = -EINVAL;
1775 mutex_lock(&inode->i_mutex);
1780 offset += file->f_pos;
1786 offset += LL_DIR_END_OFF_32BIT;
1788 offset += LL_DIR_END_OFF;
1795 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1796 (!api32 && offset <= LL_DIR_END_OFF))) {
1797 if (offset != file->f_pos) {
1798 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1799 (!api32 && offset == LL_DIR_END_OFF))
1800 fd->lfd_pos = MDS_DIR_END_OFF;
1801 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1802 fd->lfd_pos = offset << 32;
1804 fd->lfd_pos = offset;
1805 file->f_pos = offset;
1806 file->f_version = 0;
1813 mutex_unlock(&inode->i_mutex);
1817 int ll_dir_open(struct inode *inode, struct file *file)
1820 RETURN(ll_file_open(inode, file));
1823 int ll_dir_release(struct inode *inode, struct file *file)
1826 RETURN(ll_file_release(inode, file));
1829 struct file_operations ll_dir_operations = {
1830 .llseek = ll_dir_seek,
1831 .open = ll_dir_open,
1832 .release = ll_dir_release,
1833 .read = generic_read_dir,
1834 .readdir = ll_readdir,
1835 .unlocked_ioctl = ll_dir_ioctl,