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 __u64 pos = lfd->lfd_pos;
277 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
278 int api32 = ll_need_32bit_api(sbi);
279 struct md_op_data *op_data;
281 #ifdef HAVE_TOUCH_ATIME_1ARG
291 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
292 "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
293 inode, (unsigned long)pos, i_size_read(inode), api32);
295 if (pos == MDS_DIR_END_OFF)
301 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
302 LUSTRE_OPC_ANY, inode);
304 GOTO(out, rc = PTR_ERR(op_data));
306 op_data->op_hash_offset = pos;
307 op_data->op_max_pages = sbi->ll_md_brw_size >> PAGE_CACHE_SHIFT;
308 rc = ll_dir_read(inode, op_data, cookie, filldir);
310 lfd->lfd_pos = op_data->op_hash_offset;
312 if (pos == MDS_DIR_END_OFF) {
314 filp->f_pos = LL_DIR_END_OFF_32BIT;
316 filp->f_pos = LL_DIR_END_OFF;
319 filp->f_pos = op_data->op_hash_offset >> 32;
321 filp->f_pos = op_data->op_hash_offset;
324 ll_finish_md_op_data(op_data);
325 filp->f_version = inode->i_version;
326 #ifdef HAVE_TOUCH_ATIME_1ARG
327 #ifdef HAVE_F_PATH_MNT
328 path.mnt = filp->f_path.mnt;
330 path.mnt = filp->f_vfsmnt;
332 path.dentry = filp->f_dentry;
335 touch_atime(filp->f_vfsmnt, filp->f_dentry);
340 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
345 int ll_send_mgc_param(struct obd_export *mgc, char *string)
347 struct mgs_send_param *msp;
354 strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
355 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
356 sizeof(struct mgs_send_param), msp, NULL);
358 CERROR("Failed to set parameter: %d\n", rc);
364 int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
367 struct ptlrpc_request *request = NULL;
368 struct md_op_data *op_data;
369 struct ll_sb_info *sbi = ll_i2sbi(dir);
375 mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
376 op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
377 strlen(filename), mode, LUSTRE_OPC_MKDIR,
380 GOTO(err_exit, err = PTR_ERR(op_data));
382 op_data->op_cli_flags |= CLI_SET_MEA;
383 err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
384 current_fsuid(), current_fsgid(),
385 cfs_curproc_cap_pack(), 0, &request);
386 ll_finish_md_op_data(op_data);
390 ptlrpc_req_finished(request);
394 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
397 struct ll_sb_info *sbi = ll_i2sbi(inode);
398 struct md_op_data *op_data;
399 struct ptlrpc_request *req = NULL;
401 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
402 struct obd_device *mgc = lsi->lsi_mgc;
408 * This is coming from userspace, so should be in
409 * local endian. But the MDS would like it in little
410 * endian, so we swab it before we send it.
412 switch (lump->lmm_magic) {
413 case LOV_USER_MAGIC_V1: {
414 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
415 lustre_swab_lov_user_md_v1(lump);
416 lum_size = sizeof(struct lov_user_md_v1);
419 case LOV_USER_MAGIC_V3: {
420 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
421 lustre_swab_lov_user_md_v3(
422 (struct lov_user_md_v3 *)lump);
423 lum_size = sizeof(struct lov_user_md_v3);
427 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
428 " %#08x != %#08x nor %#08x\n",
429 lump->lmm_magic, LOV_USER_MAGIC_V1,
435 lum_size = sizeof(struct lov_user_md_v1);
438 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
439 LUSTRE_OPC_ANY, NULL);
441 RETURN(PTR_ERR(op_data));
443 if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
444 op_data->op_cli_flags |= CLI_SET_MEA;
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);
500 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
501 int *lmm_size, struct ptlrpc_request **request)
503 struct ll_sb_info *sbi = ll_i2sbi(inode);
504 struct mdt_body *body;
505 struct lov_mds_md *lmm = NULL;
506 struct ptlrpc_request *req = NULL;
508 struct md_op_data *op_data;
510 rc = ll_get_max_mdsize(sbi, &lmmsize);
514 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
515 0, lmmsize, LUSTRE_OPC_ANY,
518 RETURN(PTR_ERR(op_data));
520 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
521 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
522 ll_finish_md_op_data(op_data);
524 CDEBUG(D_INFO, "md_getattr failed on inode "
525 DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
529 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
530 LASSERT(body != NULL);
532 lmmsize = body->eadatasize;
534 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
536 GOTO(out, rc = -ENODATA);
539 lmm = req_capsule_server_sized_get(&req->rq_pill,
540 &RMF_MDT_MD, lmmsize);
541 LASSERT(lmm != NULL);
544 * This is coming from the MDS, so is probably in
545 * little endian. We convert it to host endian before
546 * passing it to userspace.
548 /* We don't swab objects for directories */
549 switch (le32_to_cpu(lmm->lmm_magic)) {
551 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
552 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
555 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
556 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
559 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
570 * Get MDT index for the inode.
572 int ll_get_mdt_idx(struct inode *inode)
574 struct ll_sb_info *sbi = ll_i2sbi(inode);
575 struct md_op_data *op_data;
579 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
580 0, LUSTRE_OPC_ANY, NULL);
582 RETURN(PTR_ERR(op_data));
584 op_data->op_flags |= MF_GET_MDT_IDX;
585 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
586 mdtidx = op_data->op_mds;
587 ll_finish_md_op_data(op_data);
589 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
596 * Generic handler to do any pre-copy work.
598 * It send a first hsm_progress (with extent length == 0) to coordinator as a
599 * first information for it that real work has started.
601 * Moreover, for a ARCHIVE request, it will sample the file data version and
602 * store it in \a copy.
604 * \return 0 on success.
606 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
608 struct ll_sb_info *sbi = ll_s2sbi(sb);
609 struct hsm_progress_kernel hpk;
613 /* Forge a hsm_progress based on data from copy. */
614 hpk.hpk_fid = copy->hc_hai.hai_fid;
615 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
616 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
617 hpk.hpk_extent.length = 0;
620 hpk.hpk_data_version = 0;
623 /* For archive request, we need to read the current file version. */
624 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
626 __u64 data_version = 0;
628 /* Get inode for this fid */
629 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
631 hpk.hpk_flags |= HP_FLAG_RETRY;
632 /* hpk_errval is >= 0 */
633 hpk.hpk_errval = -PTR_ERR(inode);
634 GOTO(progress, rc = PTR_ERR(inode));
637 /* Read current file data version */
638 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
641 CDEBUG(D_HSM, "Could not read file data version of "
642 DFID" (rc = %d). Archive request ("
643 LPX64") could not be done.\n",
644 PFID(©->hc_hai.hai_fid), rc,
645 copy->hc_hai.hai_cookie);
646 hpk.hpk_flags |= HP_FLAG_RETRY;
647 /* hpk_errval must be >= 0 */
648 hpk.hpk_errval = -rc;
652 /* Store it the hsm_copy for later copytool use.
653 * Always modified even if no lsm. */
654 copy->hc_data_version = data_version;
658 /* On error, the request should be considered as completed */
659 if (hpk.hpk_errval > 0)
660 hpk.hpk_flags |= HP_FLAG_COMPLETED;
661 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
668 * Generic handler to do any post-copy work.
670 * It will send the last hsm_progress update to coordinator to inform it
671 * that copy is finished and whether it was successful or not.
674 * - for ARCHIVE request, it will sample the file data version and compare it
675 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
676 * will be considered as failed.
677 * - for RESTORE request, it will sample the file data version and send it to
678 * coordinator which is useful if the file was imported as 'released'.
680 * \return 0 on success.
682 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
684 struct ll_sb_info *sbi = ll_s2sbi(sb);
685 struct hsm_progress_kernel hpk;
689 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
690 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
691 * initialized if copy_end was called with copy == NULL.
694 /* Forge a hsm_progress based on data from copy. */
695 hpk.hpk_fid = copy->hc_hai.hai_fid;
696 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
697 hpk.hpk_extent = copy->hc_hai.hai_extent;
698 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
699 hpk.hpk_errval = copy->hc_errval;
700 hpk.hpk_data_version = 0;
702 /* For archive request, we need to check the file data was not changed.
704 * For restore request, we need to send the file data version, this is
705 * useful when the file was created using hsm_import.
707 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
708 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
709 (copy->hc_errval == 0)) {
711 __u64 data_version = 0;
713 /* Get lsm for this fid */
714 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
716 hpk.hpk_flags |= HP_FLAG_RETRY;
717 /* hpk_errval must be >= 0 */
718 hpk.hpk_errval = -PTR_ERR(inode);
719 GOTO(progress, rc = PTR_ERR(inode));
722 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
725 CDEBUG(D_HSM, "Could not read file data version. "
726 "Request could not be confirmed.\n");
727 if (hpk.hpk_errval == 0)
728 hpk.hpk_errval = -rc;
732 /* Store it the hsm_copy for later copytool use.
733 * Always modified even if no lsm. */
734 hpk.hpk_data_version = data_version;
736 /* File could have been stripped during archiving, so we need
737 * to check anyway. */
738 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
739 (copy->hc_data_version != data_version)) {
740 CDEBUG(D_HSM, "File data version mismatched. "
741 "File content was changed during archiving. "
742 DFID", start:"LPX64" current:"LPX64"\n",
743 PFID(©->hc_hai.hai_fid),
744 copy->hc_data_version, data_version);
745 /* File was changed, send error to cdt. Do not ask for
746 * retry because if a file is modified frequently,
747 * the cdt will loop on retried archive requests.
748 * The policy engine will ask for a new archive later
749 * when the file will not be modified for some tunable
751 /* we do not notify caller */
752 hpk.hpk_flags &= ~HP_FLAG_RETRY;
753 /* hpk_errval must be >= 0 */
754 hpk.hpk_errval = EBUSY;
760 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
767 static int copy_and_ioctl(int cmd, struct obd_export *exp,
768 const void __user *data, size_t size)
773 OBD_ALLOC(copy, size);
777 if (copy_from_user(copy, data, size)) {
782 rc = obd_iocontrol(cmd, exp, size, copy, NULL);
784 OBD_FREE(copy, size);
789 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
791 int cmd = qctl->qc_cmd;
792 int type = qctl->qc_type;
793 int id = qctl->qc_id;
794 int valid = qctl->qc_valid;
799 case LUSTRE_Q_INVALIDATE:
800 case LUSTRE_Q_FINVALIDATE:
805 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
806 sbi->ll_flags & LL_SBI_RMT_CLIENT)
810 if (((type == USRQUOTA && current_euid() != id) ||
811 (type == GRPQUOTA && !in_egroup_p(id))) &&
812 (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
813 sbi->ll_flags & LL_SBI_RMT_CLIENT))
819 CERROR("unsupported quotactl op: %#x\n", cmd);
823 if (valid != QC_GENERAL) {
824 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
827 if (cmd == Q_GETINFO)
828 qctl->qc_cmd = Q_GETOINFO;
829 else if (cmd == Q_GETQUOTA)
830 qctl->qc_cmd = Q_GETOQUOTA;
836 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
837 sizeof(*qctl), qctl, NULL);
840 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
841 sizeof(*qctl), qctl, NULL);
844 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
845 sizeof(*qctl), qctl, NULL);
847 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
849 sizeof(*qctl), qctl, NULL);
861 struct obd_quotactl *oqctl;
863 OBD_ALLOC_PTR(oqctl);
867 QCTL_COPY(oqctl, qctl);
868 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
870 if (rc != -EALREADY && cmd == Q_QUOTAON) {
871 oqctl->qc_cmd = Q_QUOTAOFF;
872 obd_quotactl(sbi->ll_md_exp, oqctl);
877 /* If QIF_SPACE is not set, client should collect the
878 * space usage from OSSs by itself */
879 if (cmd == Q_GETQUOTA &&
880 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
881 !oqctl->qc_dqblk.dqb_curspace) {
882 struct obd_quotactl *oqctl_tmp;
884 OBD_ALLOC_PTR(oqctl_tmp);
885 if (oqctl_tmp == NULL)
886 GOTO(out, rc = -ENOMEM);
888 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
889 oqctl_tmp->qc_id = oqctl->qc_id;
890 oqctl_tmp->qc_type = oqctl->qc_type;
892 /* collect space usage from OSTs */
893 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
894 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
895 if (!rc || rc == -EREMOTEIO) {
896 oqctl->qc_dqblk.dqb_curspace =
897 oqctl_tmp->qc_dqblk.dqb_curspace;
898 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
901 /* collect space & inode usage from MDTs */
902 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
903 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
904 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
905 if (!rc || rc == -EREMOTEIO) {
906 oqctl->qc_dqblk.dqb_curspace +=
907 oqctl_tmp->qc_dqblk.dqb_curspace;
908 oqctl->qc_dqblk.dqb_curinodes =
909 oqctl_tmp->qc_dqblk.dqb_curinodes;
910 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
912 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
915 OBD_FREE_PTR(oqctl_tmp);
918 QCTL_COPY(qctl, oqctl);
926 ll_getname(const char __user *filename)
929 char *tmp = __getname();
932 return ERR_PTR(-ENOMEM);
934 len = strncpy_from_user(tmp, filename, PATH_MAX);
937 else if (len > PATH_MAX)
947 #define ll_putname(filename) __putname(filename)
949 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
951 struct inode *inode = file->f_dentry->d_inode;
952 struct ll_sb_info *sbi = ll_i2sbi(inode);
953 struct obd_ioctl_data *data;
957 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
958 PFID(ll_inode2fid(inode)), inode, cmd);
960 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
961 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
964 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
966 case FSFILT_IOC_GETFLAGS:
967 case FSFILT_IOC_SETFLAGS:
968 RETURN(ll_iocontrol(inode, file, cmd, arg));
969 case FSFILT_IOC_GETVERSION_OLD:
970 case FSFILT_IOC_GETVERSION:
971 RETURN(put_user(inode->i_generation, (int *)arg));
972 /* We need to special case any other ioctls we want to handle,
973 * to send them to the MDS/OST as appropriate and to properly
974 * network encode the arg field.
975 case FSFILT_IOC_SETVERSION_OLD:
976 case FSFILT_IOC_SETVERSION:
978 case LL_IOC_GET_MDTIDX: {
981 mdtidx = ll_get_mdt_idx(inode);
985 if (put_user((int)mdtidx, (int*)arg))
990 case IOC_MDC_LOOKUP: {
991 struct ptlrpc_request *request = NULL;
992 int namelen, len = 0;
995 struct md_op_data *op_data;
997 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1002 filename = data->ioc_inlbuf1;
1003 namelen = strlen(filename);
1006 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1007 GOTO(out_free, rc = -EINVAL);
1010 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1011 0, LUSTRE_OPC_ANY, NULL);
1012 if (IS_ERR(op_data))
1013 GOTO(out_free, rc = PTR_ERR(op_data));
1015 op_data->op_valid = OBD_MD_FLID;
1016 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1017 ll_finish_md_op_data(op_data);
1019 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1022 ptlrpc_req_finished(request);
1025 obd_ioctl_freedata(buf, len);
1028 case LL_IOC_LMV_SETSTRIPE: {
1029 struct lmv_user_md *lum;
1037 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1042 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1043 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1044 GOTO(lmv_out_free, rc = -EINVAL);
1046 filename = data->ioc_inlbuf1;
1047 namelen = data->ioc_inllen1;
1050 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1051 GOTO(lmv_out_free, rc = -EINVAL);
1053 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1054 lumlen = data->ioc_inllen2;
1056 if (lum->lum_magic != LMV_USER_MAGIC ||
1057 lumlen != sizeof(*lum)) {
1058 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1059 filename, lum->lum_magic, lumlen, -EFAULT);
1060 GOTO(lmv_out_free, rc = -EINVAL);
1064 * ll_dir_setdirstripe will be used to set dir stripe
1065 * mdc_create--->mdt_reint_create (with dirstripe)
1067 rc = ll_dir_setdirstripe(inode, lum, filename);
1069 obd_ioctl_freedata(buf, len);
1073 case LL_IOC_LOV_SETSTRIPE: {
1074 struct lov_user_md_v3 lumv3;
1075 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1076 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1077 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1079 int set_default = 0;
1081 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1082 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1083 sizeof(lumv3p->lmm_objects[0]));
1084 /* first try with v1 which is smaller than v3 */
1085 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1088 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1089 if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1093 if (inode->i_sb->s_root == file->f_dentry)
1096 /* in v1 and v3 cases lumv1 points to data */
1097 rc = ll_dir_setstripe(inode, lumv1, set_default);
1101 case LL_IOC_LMV_GETSTRIPE: {
1102 struct lmv_user_md *lump = (struct lmv_user_md *)arg;
1103 struct lmv_user_md lum;
1104 struct lmv_user_md *tmp;
1109 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1112 if (lum.lum_magic != LMV_MAGIC_V1)
1115 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1116 OBD_ALLOC(tmp, lum_size);
1118 GOTO(free_lmv, rc = -ENOMEM);
1120 memcpy(tmp, &lum, sizeof(lum));
1121 tmp->lum_type = LMV_STRIPE_TYPE;
1122 tmp->lum_stripe_count = 1;
1123 mdtindex = ll_get_mdt_idx(inode);
1125 GOTO(free_lmv, rc = -ENOMEM);
1127 tmp->lum_stripe_offset = mdtindex;
1128 tmp->lum_objects[0].lum_mds = mdtindex;
1129 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1130 sizeof(struct lu_fid));
1131 if (copy_to_user((void *)arg, tmp, lum_size))
1132 GOTO(free_lmv, rc = -EFAULT);
1135 OBD_FREE(tmp, lum_size);
1138 case LL_IOC_REMOVE_ENTRY: {
1139 char *filename = NULL;
1143 /* Here is a little hack to avoid sending REINT_RMENTRY to
1144 * unsupported server, which might crash the server(LU-2730),
1145 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1146 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1147 * server will support REINT_RMENTRY XXX*/
1148 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1151 filename = ll_getname((const char *)arg);
1152 if (IS_ERR(filename))
1153 RETURN(PTR_ERR(filename));
1155 namelen = strlen(filename);
1157 GOTO(out_rmdir, rc = -EINVAL);
1159 rc = ll_rmdir_entry(inode, filename, namelen);
1162 ll_putname(filename);
1165 case LL_IOC_LOV_SWAP_LAYOUTS:
1167 case LL_IOC_OBD_STATFS:
1168 RETURN(ll_obd_statfs(inode, (void *)arg));
1169 case LL_IOC_LOV_GETSTRIPE:
1170 case LL_IOC_MDC_GETINFO:
1171 case IOC_MDC_GETFILEINFO:
1172 case IOC_MDC_GETFILESTRIPE: {
1173 struct ptlrpc_request *request = NULL;
1174 struct lov_user_md *lump;
1175 struct lov_mds_md *lmm = NULL;
1176 struct mdt_body *body;
1177 char *filename = NULL;
1180 if (cmd == IOC_MDC_GETFILEINFO ||
1181 cmd == IOC_MDC_GETFILESTRIPE) {
1182 filename = ll_getname((const char *)arg);
1183 if (IS_ERR(filename))
1184 RETURN(PTR_ERR(filename));
1186 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1187 &lmmsize, &request);
1189 rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1193 body = req_capsule_server_get(&request->rq_pill,
1195 LASSERT(body != NULL);
1201 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1202 cmd == LL_IOC_MDC_GETINFO))
1203 GOTO(skip_lmm, rc = 0);
1208 if (cmd == IOC_MDC_GETFILESTRIPE ||
1209 cmd == LL_IOC_LOV_GETSTRIPE) {
1210 lump = (struct lov_user_md *)arg;
1212 struct lov_user_mds_data *lmdp;
1213 lmdp = (struct lov_user_mds_data *)arg;
1214 lump = &lmdp->lmd_lmm;
1216 if (copy_to_user(lump, lmm, lmmsize)) {
1217 if (copy_to_user(lump, lmm, sizeof(*lump)))
1218 GOTO(out_req, rc = -EFAULT);
1222 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1223 struct lov_user_mds_data *lmdp;
1226 st.st_dev = inode->i_sb->s_dev;
1227 st.st_mode = body->mode;
1228 st.st_nlink = body->nlink;
1229 st.st_uid = body->uid;
1230 st.st_gid = body->gid;
1231 st.st_rdev = body->rdev;
1232 st.st_size = body->size;
1233 st.st_blksize = PAGE_CACHE_SIZE;
1234 st.st_blocks = body->blocks;
1235 st.st_atime = body->atime;
1236 st.st_mtime = body->mtime;
1237 st.st_ctime = body->ctime;
1238 st.st_ino = inode->i_ino;
1240 lmdp = (struct lov_user_mds_data *)arg;
1241 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1242 GOTO(out_req, rc = -EFAULT);
1247 ptlrpc_req_finished(request);
1249 ll_putname(filename);
1252 case IOC_LOV_GETINFO: {
1253 struct lov_user_mds_data *lumd;
1254 struct lov_stripe_md *lsm;
1255 struct lov_user_md *lum;
1256 struct lov_mds_md *lmm;
1260 lumd = (struct lov_user_mds_data *)arg;
1261 lum = &lumd->lmd_lmm;
1263 rc = ll_get_max_mdsize(sbi, &lmmsize);
1267 OBD_ALLOC_LARGE(lmm, lmmsize);
1271 if (copy_from_user(lmm, lum, lmmsize))
1272 GOTO(free_lmm, rc = -EFAULT);
1274 switch (lmm->lmm_magic) {
1275 case LOV_USER_MAGIC_V1:
1276 if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1278 /* swab objects first so that stripes num will be sane */
1279 lustre_swab_lov_user_md_objects(
1280 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1281 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1282 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1284 case LOV_USER_MAGIC_V3:
1285 if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1287 /* swab objects first so that stripes num will be sane */
1288 lustre_swab_lov_user_md_objects(
1289 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1290 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1291 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1294 GOTO(free_lmm, rc = -EINVAL);
1297 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1299 GOTO(free_lmm, rc = -ENOMEM);
1301 /* Perform glimpse_size operation. */
1302 memset(&st, 0, sizeof(st));
1304 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1308 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1309 GOTO(free_lsm, rc = -EFAULT);
1313 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1315 OBD_FREE_LARGE(lmm, lmmsize);
1318 case OBD_IOC_LLOG_CATINFO: {
1319 RETURN(-EOPNOTSUPP);
1321 case OBD_IOC_QUOTACHECK: {
1322 struct obd_quotactl *oqctl;
1325 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1326 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1329 OBD_ALLOC_PTR(oqctl);
1332 oqctl->qc_type = arg;
1333 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1335 CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1339 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1341 CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1343 OBD_FREE_PTR(oqctl);
1346 case OBD_IOC_POLL_QUOTACHECK: {
1347 struct if_quotacheck *check;
1349 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1350 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1353 OBD_ALLOC_PTR(check);
1357 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1360 CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1361 if (copy_to_user((void *)arg, check,
1363 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1367 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1370 CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1371 if (copy_to_user((void *)arg, check,
1373 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1377 OBD_FREE_PTR(check);
1380 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1381 case LL_IOC_QUOTACTL_18: {
1382 /* copy the old 1.x quota struct for internal use, then copy
1383 * back into old format struct. For 1.8 compatibility. */
1384 struct if_quotactl_18 *qctl_18;
1385 struct if_quotactl *qctl_20;
1387 OBD_ALLOC_PTR(qctl_18);
1391 OBD_ALLOC_PTR(qctl_20);
1393 GOTO(out_quotactl_18, rc = -ENOMEM);
1395 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1396 GOTO(out_quotactl_20, rc = -ENOMEM);
1398 QCTL_COPY(qctl_20, qctl_18);
1399 qctl_20->qc_idx = 0;
1401 /* XXX: dqb_valid was borrowed as a flag to mark that
1402 * only mds quota is wanted */
1403 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1404 qctl_18->qc_dqblk.dqb_valid) {
1405 qctl_20->qc_valid = QC_MDTIDX;
1406 qctl_20->qc_dqblk.dqb_valid = 0;
1407 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1408 qctl_20->qc_valid = QC_UUID;
1409 qctl_20->obd_uuid = qctl_18->obd_uuid;
1411 qctl_20->qc_valid = QC_GENERAL;
1414 rc = quotactl_ioctl(sbi, qctl_20);
1417 QCTL_COPY(qctl_18, qctl_20);
1418 qctl_18->obd_uuid = qctl_20->obd_uuid;
1420 if (copy_to_user((void *)arg, qctl_18,
1426 OBD_FREE_PTR(qctl_20);
1428 OBD_FREE_PTR(qctl_18);
1432 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1433 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1434 case LL_IOC_QUOTACTL: {
1435 struct if_quotactl *qctl;
1437 OBD_ALLOC_PTR(qctl);
1441 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1442 GOTO(out_quotactl, rc = -EFAULT);
1444 rc = quotactl_ioctl(sbi, qctl);
1446 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1453 case OBD_IOC_GETDTNAME:
1454 case OBD_IOC_GETMDNAME:
1455 RETURN(ll_get_obd_name(inode, cmd, arg));
1456 case LL_IOC_FLUSHCTX:
1457 RETURN(ll_flush_ctx(inode));
1458 #ifdef CONFIG_FS_POSIX_ACL
1459 case LL_IOC_RMTACL: {
1460 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1461 inode == inode->i_sb->s_root->d_inode) {
1462 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1464 LASSERT(fd != NULL);
1465 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1467 fd->fd_flags |= LL_FILE_RMTACL;
1473 case LL_IOC_GETOBDCOUNT: {
1475 struct obd_export *exp;
1477 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1480 /* get ost count when count is zero, get mdt count otherwise */
1481 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1482 vallen = sizeof(count);
1483 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1484 KEY_TGT_COUNT, &vallen, &count, NULL);
1486 CERROR("get target count failed: %d\n", rc);
1490 if (copy_to_user((int *)arg, &count, sizeof(int)))
1495 case LL_IOC_PATH2FID:
1496 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1497 sizeof(struct lu_fid)))
1500 case LL_IOC_GET_CONNECT_FLAGS: {
1501 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1503 case OBD_IOC_CHANGELOG_SEND:
1504 case OBD_IOC_CHANGELOG_CLEAR:
1505 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1506 sizeof(struct ioc_changelog));
1508 case OBD_IOC_FID2PATH:
1509 RETURN(ll_fid2path(inode, (void *)arg));
1510 case LL_IOC_HSM_REQUEST: {
1511 struct hsm_user_request *hur;
1518 /* We don't know the true size yet; copy the fixed-size part */
1519 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1524 /* Compute the whole struct size */
1525 totalsize = hur_len(hur);
1528 /* Make sure the size is reasonable */
1529 if (totalsize >= MDS_MAXREQSIZE)
1532 OBD_ALLOC_LARGE(hur, totalsize);
1536 /* Copy the whole struct */
1537 if (copy_from_user(hur, (void *)arg, totalsize)) {
1538 OBD_FREE_LARGE(hur, totalsize);
1542 if (hur->hur_request.hr_action == HUA_RELEASE) {
1543 const struct lu_fid *fid;
1547 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1548 fid = &hur->hur_user_item[i].hui_fid;
1549 f = search_inode_for_lustre(inode->i_sb, fid);
1555 rc = ll_hsm_release(f);
1561 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1565 OBD_FREE_LARGE(hur, totalsize);
1569 case LL_IOC_HSM_PROGRESS: {
1570 struct hsm_progress_kernel hpk;
1571 struct hsm_progress hp;
1573 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1576 hpk.hpk_fid = hp.hp_fid;
1577 hpk.hpk_cookie = hp.hp_cookie;
1578 hpk.hpk_extent = hp.hp_extent;
1579 hpk.hpk_flags = hp.hp_flags;
1580 hpk.hpk_errval = hp.hp_errval;
1581 hpk.hpk_data_version = 0;
1583 /* File may not exist in Lustre; all progress
1584 * reported to Lustre root */
1585 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1589 case LL_IOC_HSM_CT_START:
1590 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1593 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1594 sizeof(struct lustre_kernelcomm));
1597 case LL_IOC_HSM_COPY_START: {
1598 struct hsm_copy *copy;
1601 OBD_ALLOC_PTR(copy);
1604 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1609 rc = ll_ioc_copy_start(inode->i_sb, copy);
1610 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1616 case LL_IOC_HSM_COPY_END: {
1617 struct hsm_copy *copy;
1620 OBD_ALLOC_PTR(copy);
1623 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1628 rc = ll_ioc_copy_end(inode->i_sb, copy);
1629 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1636 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1641 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1643 struct inode *inode = file->f_mapping->host;
1644 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1645 struct ll_sb_info *sbi = ll_i2sbi(inode);
1646 int api32 = ll_need_32bit_api(sbi);
1647 loff_t ret = -EINVAL;
1650 mutex_lock(&inode->i_mutex);
1655 offset += file->f_pos;
1661 offset += LL_DIR_END_OFF_32BIT;
1663 offset += LL_DIR_END_OFF;
1670 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1671 (!api32 && offset <= LL_DIR_END_OFF))) {
1672 if (offset != file->f_pos) {
1673 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1674 (!api32 && offset == LL_DIR_END_OFF))
1675 fd->lfd_pos = MDS_DIR_END_OFF;
1676 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1677 fd->lfd_pos = offset << 32;
1679 fd->lfd_pos = offset;
1680 file->f_pos = offset;
1681 file->f_version = 0;
1688 mutex_unlock(&inode->i_mutex);
1692 int ll_dir_open(struct inode *inode, struct file *file)
1695 RETURN(ll_file_open(inode, file));
1698 int ll_dir_release(struct inode *inode, struct file *file)
1701 RETURN(ll_file_release(inode, file));
1704 struct file_operations ll_dir_operations = {
1705 .llseek = ll_dir_seek,
1706 .open = ll_dir_open,
1707 .release = ll_dir_release,
1708 .read = generic_read_dir,
1709 .readdir = ll_readdir,
1710 .unlocked_ioctl = ll_dir_ioctl,