4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Directory code for lustre client.
42 #include <linux/pagemap.h>
44 #include <linux/version.h>
45 #include <linux/user_namespace.h>
46 #ifdef HAVE_UIDGID_HEADER
47 # include <linux/uidgid.h>
49 #include <asm/uaccess.h>
50 #include <linux/buffer_head.h> // for wait_on_buffer
51 #include <linux/pagevec.h>
53 #define DEBUG_SUBSYSTEM S_LLITE
55 #include <lustre/lustre_idl.h>
56 #include <obd_support.h>
57 #include <obd_class.h>
58 #include <lustre_ioctl.h>
59 #include <lustre_lib.h>
60 #include <lustre_dlm.h>
61 #include <lustre_fid.h>
62 #include <lustre_kernelcomm.h>
63 #include "llite_internal.h"
66 * (new) readdir implementation overview.
68 * Original lustre readdir implementation cached exact copy of raw directory
69 * pages on the client. These pages were indexed in client page cache by
70 * logical offset in the directory file. This design, while very simple and
71 * intuitive had some inherent problems:
73 * . it implies that byte offset to the directory entry serves as a
74 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
75 * ext3/htree directory entries may move due to splits, and more
78 * . it is incompatible with the design of split directories for cmd3,
79 * that assumes that names are distributed across nodes based on their
80 * hash, and so readdir should be done in hash order.
82 * New readdir implementation does readdir in hash order, and uses hash of a
83 * file name as a telldir/seekdir cookie. This led to number of complications:
85 * . hash is not unique, so it cannot be used to index cached directory
86 * pages on the client (note, that it requires a whole pageful of hash
87 * collided entries to cause two pages to have identical hashes);
89 * . hash is not unique, so it cannot, strictly speaking, be used as an
90 * entry cookie. ext3/htree has the same problem and lustre implementation
91 * mimics their solution: seekdir(hash) positions directory at the first
92 * entry with the given hash.
98 * Client caches directory pages using hash of the first entry as an index. As
99 * noted above hash is not unique, so this solution doesn't work as is:
100 * special processing is needed for "page hash chains" (i.e., sequences of
101 * pages filled with entries all having the same hash value).
103 * First, such chains have to be detected. To this end, server returns to the
104 * client the hash of the first entry on the page next to one returned. When
105 * client detects that this hash is the same as hash of the first entry on the
106 * returned page, page hash collision has to be handled. Pages in the
107 * hash chain, except first one, are termed "overflow pages".
109 * Solution to index uniqueness problem is to not cache overflow
110 * pages. Instead, when page hash collision is detected, all overflow pages
111 * from emerging chain are immediately requested from the server and placed in
112 * a special data structure (struct ll_dir_chain). This data structure is used
113 * by ll_readdir() to process entries from overflow pages. When readdir
114 * invocation finishes, overflow pages are discarded. If page hash collision
115 * chain weren't completely processed, next call to readdir will again detect
116 * page hash collision, again read overflow pages in, process next portion of
117 * entries and again discard the pages. This is not as wasteful as it looks,
118 * because, given reasonable hash, page hash collisions are extremely rare.
120 * 1. directory positioning
122 * When seekdir(hash) is called, original
133 * identification of and access to overflow pages
137 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
138 * a header lu_dirpage which describes the start/end hash, and whether this
139 * page is empty (contains no dir entry) or hash collide with next page.
140 * After client receives reply, several pages will be integrated into dir page
141 * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
142 * lu_dirpage for this integrated page will be adjusted. See
143 * mdc_adjust_dirpages().
146 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
147 __u64 offset, struct ll_dir_chain *chain)
149 struct md_callback cb_op;
153 cb_op.md_blocking_ast = ll_md_blocking_ast;
154 rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
161 void ll_release_page(struct inode *inode, struct page *page,
166 /* Always remove the page for striped dir, because the page is
167 * built from temporarily in LMV layer */
168 if (inode != NULL && S_ISDIR(inode->i_mode) &&
169 ll_i2info(inode)->lli_lsm_md != NULL) {
176 if (likely(page->mapping != NULL))
177 truncate_complete_page(page->mapping, page);
180 page_cache_release(page);
184 * return IF_* type for given lu_dirent entry.
185 * IF_* flag shld be converted to particular OS file type in
186 * platform llite module.
188 static u16 ll_dirent_type_get(struct lu_dirent *ent)
191 struct luda_type *lt;
194 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
195 const unsigned align = sizeof(struct luda_type) - 1;
197 len = le16_to_cpu(ent->lde_namelen);
198 len = (len + align) & ~align;
199 lt = (void *)ent->lde_name + len;
200 type = IFTODT(le16_to_cpu(lt->lt_type));
206 #ifdef HAVE_DIR_CONTEXT
207 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
208 struct dir_context *ctx)
211 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
212 void *cookie, filldir_t filldir)
215 struct ll_sb_info *sbi = ll_i2sbi(inode);
217 bool is_api32 = ll_need_32bit_api(sbi);
218 bool is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
220 struct ll_dir_chain chain;
225 ll_dir_chain_init(&chain);
227 page = ll_get_dir_page(inode, op_data, pos, &chain);
229 while (rc == 0 && !done) {
230 struct lu_dirpage *dp;
231 struct lu_dirent *ent;
240 hash = MDS_DIR_END_OFF;
241 dp = page_address(page);
242 for (ent = lu_dirent_start(dp); ent != NULL && !done;
243 ent = lu_dirent_next(ent)) {
250 hash = le64_to_cpu(ent->lde_hash);
253 * Skip until we find target hash
258 namelen = le16_to_cpu(ent->lde_namelen);
265 if (is_api32 && is_hash64)
269 fid_le_to_cpu(&fid, &ent->lde_fid);
270 ino = cl_fid_build_ino(&fid, is_api32);
271 type = ll_dirent_type_get(ent);
272 /* For 'll_nfs_get_name_filldir()', it will try
273 * to access the 'ent' through its 'lde_name',
274 * so the parameter 'name' for 'filldir()' must
275 * be part of the 'ent'. */
276 #ifdef HAVE_DIR_CONTEXT
278 done = !dir_emit(ctx, ent->lde_name, namelen, ino,
281 done = filldir(cookie, ent->lde_name, namelen, lhash,
288 ll_release_page(inode, page, false);
292 next = le64_to_cpu(dp->ldp_hash_end);
294 if (pos == MDS_DIR_END_OFF) {
296 * End of directory reached.
299 ll_release_page(inode, page, false);
302 * Normal case: continue to the next
305 ll_release_page(inode, page,
306 le32_to_cpu(dp->ldp_flags) &
309 page = ll_get_dir_page(inode, op_data, pos,
313 #ifdef HAVE_DIR_CONTEXT
318 ll_dir_chain_fini(&chain);
322 #ifdef HAVE_DIR_CONTEXT
323 static int ll_iterate(struct file *filp, struct dir_context *ctx)
325 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
328 struct inode *inode = filp->f_dentry->d_inode;
329 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
330 struct ll_sb_info *sbi = ll_i2sbi(inode);
331 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
332 int api32 = ll_need_32bit_api(sbi);
333 struct md_op_data *op_data;
343 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
344 "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
345 inode, (unsigned long)pos, i_size_read(inode), api32);
347 if (pos == MDS_DIR_END_OFF)
353 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
354 LUSTRE_OPC_ANY, inode);
356 GOTO(out, rc = PTR_ERR(op_data));
358 if (unlikely(op_data->op_mea1 != NULL)) {
359 /* This is only needed for striped dir to fill ..,
360 * see lmv_read_entry */
361 if (filp->f_dentry->d_parent != NULL &&
362 filp->f_dentry->d_parent->d_inode != NULL) {
363 __u64 ibits = MDS_INODELOCK_UPDATE;
364 struct inode *parent =
365 filp->f_dentry->d_parent->d_inode;
367 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
368 op_data->op_fid3 = *ll_inode2fid(parent);
371 /* If it can not find in cache, do lookup .. on the master
373 if (fid_is_zero(&op_data->op_fid3)) {
374 rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
376 ll_finish_md_op_data(op_data);
381 op_data->op_max_pages = sbi->ll_md_brw_pages;
382 #ifdef HAVE_DIR_CONTEXT
384 rc = ll_dir_read(inode, &pos, op_data, ctx);
387 rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
392 if (pos == MDS_DIR_END_OFF) {
394 pos = LL_DIR_END_OFF_32BIT;
396 pos = LL_DIR_END_OFF;
401 #ifdef HAVE_DIR_CONTEXT
406 ll_finish_md_op_data(op_data);
407 filp->f_version = inode->i_version;
411 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
416 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
418 struct mgs_send_param *msp;
425 strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
426 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
427 sizeof(struct mgs_send_param), msp, NULL);
429 CERROR("Failed to set parameter: %d\n", rc);
436 * Create striped directory with specified stripe(@lump)
438 * param[in]parent the parent of the directory.
439 * param[in]lump the specified stripes.
440 * param[in]dirname the name of the directory.
441 * param[in]mode the specified mode of the directory.
443 * retval =0 if striped directory is being created successfully.
444 * <0 if the creation is failed.
446 static int ll_dir_setdirstripe(struct inode *parent, struct lmv_user_md *lump,
447 const char *dirname, umode_t mode)
449 struct ptlrpc_request *request = NULL;
450 struct md_op_data *op_data;
451 struct ll_sb_info *sbi = ll_i2sbi(parent);
455 if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
458 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
459 "stripe_offset %d, stripe_count: %u\n",
460 PFID(ll_inode2fid(parent)), parent, dirname,
461 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
463 if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
464 lustre_swab_lmv_user_md(lump);
466 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
467 mode &= ~current_umask();
468 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
469 op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
470 strlen(dirname), mode, LUSTRE_OPC_MKDIR,
473 GOTO(err_exit, err = PTR_ERR(op_data));
475 op_data->op_cli_flags |= CLI_SET_MEA;
476 err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
477 from_kuid(&init_user_ns, current_fsuid()),
478 from_kgid(&init_user_ns, current_fsgid()),
479 cfs_curproc_cap_pack(), 0, &request);
480 ll_finish_md_op_data(op_data);
484 ptlrpc_req_finished(request);
488 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
491 struct ll_sb_info *sbi = ll_i2sbi(inode);
492 struct md_op_data *op_data;
493 struct ptlrpc_request *req = NULL;
495 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
496 struct obd_device *mgc = lsi->lsi_mgc;
502 * This is coming from userspace, so should be in
503 * local endian. But the MDS would like it in little
504 * endian, so we swab it before we send it.
506 switch (lump->lmm_magic) {
507 case LOV_USER_MAGIC_V1: {
508 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
509 lustre_swab_lov_user_md_v1(lump);
510 lum_size = sizeof(struct lov_user_md_v1);
513 case LOV_USER_MAGIC_V3: {
514 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
515 lustre_swab_lov_user_md_v3(
516 (struct lov_user_md_v3 *)lump);
517 lum_size = sizeof(struct lov_user_md_v3);
520 case LMV_USER_MAGIC: {
521 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
522 lustre_swab_lmv_user_md(
523 (struct lmv_user_md *)lump);
524 lum_size = sizeof(struct lmv_user_md);
528 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
529 " %#08x != %#08x nor %#08x\n",
530 lump->lmm_magic, LOV_USER_MAGIC_V1,
536 lum_size = sizeof(struct lov_user_md_v1);
539 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
540 LUSTRE_OPC_ANY, NULL);
542 RETURN(PTR_ERR(op_data));
544 /* swabbing is done in lov_setstripe() on server side */
545 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
546 ll_finish_md_op_data(op_data);
547 ptlrpc_req_finished(req);
549 if (rc != -EPERM && rc != -EACCES)
550 CERROR("mdc_setattr fails: rc = %d\n", rc);
553 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
554 LOV_USER_MAGIC_V3 have the same initial fields so we do not
555 need the make the distiction between the 2 versions */
556 if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
560 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
562 GOTO(end, rc = -ENOMEM);
565 /* Get fsname and assume devname to be -MDT0000. */
566 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
567 strcat(buf, "-MDT0000.lov");
570 /* Set root stripesize */
571 sprintf(buf, ".stripesize=%u",
572 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
573 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
577 /* Set root stripecount */
578 sprintf(buf, ".stripecount=%hd",
579 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
580 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
584 /* Set root stripeoffset */
585 sprintf(buf, ".stripeoffset=%hd",
586 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
587 (typeof(lump->lmm_stripe_offset))(-1));
588 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
592 OBD_FREE(param, MGS_PARAM_MAXLEN);
598 * This function will be used to get default LOV/LMV/Default LMV
599 * @valid will be used to indicate which stripe it will retrieve
600 * OBD_MD_MEA LMV stripe EA
601 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
602 * otherwise Default LOV EA.
603 * Each time, it can only retrieve 1 stripe EA
605 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
606 struct ptlrpc_request **request, u64 valid)
608 struct ll_sb_info *sbi = ll_i2sbi(inode);
609 struct mdt_body *body;
610 struct lov_mds_md *lmm = NULL;
611 struct ptlrpc_request *req = NULL;
613 struct md_op_data *op_data;
616 rc = ll_get_default_mdsize(sbi, &lmm_size);
620 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
621 0, lmm_size, LUSTRE_OPC_ANY,
624 RETURN(PTR_ERR(op_data));
626 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
627 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
628 ll_finish_md_op_data(op_data);
630 CDEBUG(D_INFO, "md_getattr failed on inode "
631 DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
635 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
636 LASSERT(body != NULL);
638 lmm_size = body->mbo_eadatasize;
640 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
642 GOTO(out, rc = -ENODATA);
645 lmm = req_capsule_server_sized_get(&req->rq_pill,
646 &RMF_MDT_MD, lmm_size);
647 LASSERT(lmm != NULL);
650 * This is coming from the MDS, so is probably in
651 * little endian. We convert it to host endian before
652 * passing it to userspace.
654 /* We don't swab objects for directories */
655 switch (le32_to_cpu(lmm->lmm_magic)) {
657 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
658 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
661 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
662 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
665 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
666 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
669 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
670 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
673 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
678 *plmm_size = lmm_size;
683 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
685 struct md_op_data *op_data;
690 OBD_ALLOC_PTR(op_data);
694 op_data->op_flags |= MF_GET_MDT_IDX;
695 op_data->op_fid1 = *fid;
696 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
697 mdt_index = op_data->op_mds;
698 OBD_FREE_PTR(op_data);
706 * Get MDT index for the inode.
708 int ll_get_mdt_idx(struct inode *inode)
710 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
714 * Generic handler to do any pre-copy work.
716 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
717 * first information for it that real work has started.
719 * Moreover, for a ARCHIVE request, it will sample the file data version and
720 * store it in \a copy.
722 * \return 0 on success.
724 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
726 struct ll_sb_info *sbi = ll_s2sbi(sb);
727 struct hsm_progress_kernel hpk;
732 /* Forge a hsm_progress based on data from copy. */
733 hpk.hpk_fid = copy->hc_hai.hai_fid;
734 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
735 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
736 hpk.hpk_extent.length = 0;
739 hpk.hpk_data_version = 0;
742 /* For archive request, we need to read the current file version. */
743 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
745 __u64 data_version = 0;
747 /* Get inode for this fid */
748 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
750 hpk.hpk_flags |= HP_FLAG_RETRY;
751 /* hpk_errval is >= 0 */
752 hpk.hpk_errval = -PTR_ERR(inode);
753 GOTO(progress, rc = PTR_ERR(inode));
756 /* Read current file data version */
757 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
760 CDEBUG(D_HSM, "Could not read file data version of "
761 DFID" (rc = %d). Archive request ("
762 LPX64") could not be done.\n",
763 PFID(©->hc_hai.hai_fid), rc,
764 copy->hc_hai.hai_cookie);
765 hpk.hpk_flags |= HP_FLAG_RETRY;
766 /* hpk_errval must be >= 0 */
767 hpk.hpk_errval = -rc;
771 /* Store in the hsm_copy for later copytool use.
772 * Always modified even if no lsm. */
773 copy->hc_data_version = data_version;
777 /* On error, the request should be considered as completed */
778 if (hpk.hpk_errval > 0)
779 hpk.hpk_flags |= HP_FLAG_COMPLETED;
781 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
784 /* Return first error */
785 RETURN(rc != 0 ? rc : rc2);
789 * Generic handler to do any post-copy work.
791 * It will send the last hsm_progress update to coordinator to inform it
792 * that copy is finished and whether it was successful or not.
795 * - for ARCHIVE request, it will sample the file data version and compare it
796 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
797 * will be considered as failed.
798 * - for RESTORE request, it will sample the file data version and send it to
799 * coordinator which is useful if the file was imported as 'released'.
801 * \return 0 on success.
803 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
805 struct ll_sb_info *sbi = ll_s2sbi(sb);
806 struct hsm_progress_kernel hpk;
811 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
812 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
813 * initialized if copy_end was called with copy == NULL.
816 /* Forge a hsm_progress based on data from copy. */
817 hpk.hpk_fid = copy->hc_hai.hai_fid;
818 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
819 hpk.hpk_extent = copy->hc_hai.hai_extent;
820 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
821 hpk.hpk_errval = copy->hc_errval;
822 hpk.hpk_data_version = 0;
824 /* For archive request, we need to check the file data was not changed.
826 * For restore request, we need to send the file data version, this is
827 * useful when the file was created using hsm_import.
829 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
830 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
831 (copy->hc_errval == 0)) {
833 __u64 data_version = 0;
835 /* Get lsm for this fid */
836 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
838 hpk.hpk_flags |= HP_FLAG_RETRY;
839 /* hpk_errval must be >= 0 */
840 hpk.hpk_errval = -PTR_ERR(inode);
841 GOTO(progress, rc = PTR_ERR(inode));
844 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
847 CDEBUG(D_HSM, "Could not read file data version. "
848 "Request could not be confirmed.\n");
849 if (hpk.hpk_errval == 0)
850 hpk.hpk_errval = -rc;
854 /* Store in the hsm_copy for later copytool use.
855 * Always modified even if no lsm. */
856 hpk.hpk_data_version = data_version;
858 /* File could have been stripped during archiving, so we need
859 * to check anyway. */
860 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
861 (copy->hc_data_version != data_version)) {
862 CDEBUG(D_HSM, "File data version mismatched. "
863 "File content was changed during archiving. "
864 DFID", start:"LPX64" current:"LPX64"\n",
865 PFID(©->hc_hai.hai_fid),
866 copy->hc_data_version, data_version);
867 /* File was changed, send error to cdt. Do not ask for
868 * retry because if a file is modified frequently,
869 * the cdt will loop on retried archive requests.
870 * The policy engine will ask for a new archive later
871 * when the file will not be modified for some tunable
873 /* we do not notify caller */
874 hpk.hpk_flags &= ~HP_FLAG_RETRY;
875 /* hpk_errval must be >= 0 */
876 hpk.hpk_errval = EBUSY;
882 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
885 /* Return first error */
886 RETURN(rc != 0 ? rc : rc2);
890 static int copy_and_ioctl(int cmd, struct obd_export *exp,
891 const void __user *data, size_t size)
896 OBD_ALLOC(copy, size);
900 if (copy_from_user(copy, data, size)) {
905 rc = obd_iocontrol(cmd, exp, size, copy, NULL);
907 OBD_FREE(copy, size);
912 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
914 int cmd = qctl->qc_cmd;
915 int type = qctl->qc_type;
916 int id = qctl->qc_id;
917 int valid = qctl->qc_valid;
922 case LUSTRE_Q_INVALIDATE:
923 case LUSTRE_Q_FINVALIDATE:
928 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
929 sbi->ll_flags & LL_SBI_RMT_CLIENT)
933 if (((type == USRQUOTA &&
934 !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
936 !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
937 (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
938 sbi->ll_flags & LL_SBI_RMT_CLIENT))
944 CERROR("unsupported quotactl op: %#x\n", cmd);
948 if (valid != QC_GENERAL) {
949 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
952 if (cmd == Q_GETINFO)
953 qctl->qc_cmd = Q_GETOINFO;
954 else if (cmd == Q_GETQUOTA)
955 qctl->qc_cmd = Q_GETOQUOTA;
961 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
962 sizeof(*qctl), qctl, NULL);
965 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
966 sizeof(*qctl), qctl, NULL);
969 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
970 sizeof(*qctl), qctl, NULL);
972 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
974 sizeof(*qctl), qctl, NULL);
986 struct obd_quotactl *oqctl;
988 OBD_ALLOC_PTR(oqctl);
992 QCTL_COPY(oqctl, qctl);
993 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
995 if (rc != -EALREADY && cmd == Q_QUOTAON) {
996 oqctl->qc_cmd = Q_QUOTAOFF;
997 obd_quotactl(sbi->ll_md_exp, oqctl);
1002 /* If QIF_SPACE is not set, client should collect the
1003 * space usage from OSSs by itself */
1004 if (cmd == Q_GETQUOTA &&
1005 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1006 !oqctl->qc_dqblk.dqb_curspace) {
1007 struct obd_quotactl *oqctl_tmp;
1009 OBD_ALLOC_PTR(oqctl_tmp);
1010 if (oqctl_tmp == NULL)
1011 GOTO(out, rc = -ENOMEM);
1013 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1014 oqctl_tmp->qc_id = oqctl->qc_id;
1015 oqctl_tmp->qc_type = oqctl->qc_type;
1017 /* collect space usage from OSTs */
1018 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1019 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1020 if (!rc || rc == -EREMOTEIO) {
1021 oqctl->qc_dqblk.dqb_curspace =
1022 oqctl_tmp->qc_dqblk.dqb_curspace;
1023 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1026 /* collect space & inode usage from MDTs */
1027 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1028 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1029 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1030 if (!rc || rc == -EREMOTEIO) {
1031 oqctl->qc_dqblk.dqb_curspace +=
1032 oqctl_tmp->qc_dqblk.dqb_curspace;
1033 oqctl->qc_dqblk.dqb_curinodes =
1034 oqctl_tmp->qc_dqblk.dqb_curinodes;
1035 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1037 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1040 OBD_FREE_PTR(oqctl_tmp);
1043 QCTL_COPY(qctl, oqctl);
1044 OBD_FREE_PTR(oqctl);
1051 ll_getname(const char __user *filename)
1054 char *tmp = __getname();
1057 return ERR_PTR(-ENOMEM);
1059 len = strncpy_from_user(tmp, filename, PATH_MAX);
1062 else if (len > PATH_MAX)
1063 ret = -ENAMETOOLONG;
1072 #define ll_putname(filename) __putname(filename)
1074 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1076 struct inode *inode = file->f_dentry->d_inode;
1077 struct ll_sb_info *sbi = ll_i2sbi(inode);
1078 struct obd_ioctl_data *data;
1082 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1083 PFID(ll_inode2fid(inode)), inode, cmd);
1085 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1086 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1089 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1091 case FSFILT_IOC_GETFLAGS:
1092 case FSFILT_IOC_SETFLAGS:
1093 RETURN(ll_iocontrol(inode, file, cmd, arg));
1094 case FSFILT_IOC_GETVERSION_OLD:
1095 case FSFILT_IOC_GETVERSION:
1096 RETURN(put_user(inode->i_generation, (int __user *)arg));
1097 /* We need to special case any other ioctls we want to handle,
1098 * to send them to the MDS/OST as appropriate and to properly
1099 * network encode the arg field.
1100 case FSFILT_IOC_SETVERSION_OLD:
1101 case FSFILT_IOC_SETVERSION:
1103 case LL_IOC_GET_MDTIDX: {
1106 mdtidx = ll_get_mdt_idx(inode);
1110 if (put_user((int)mdtidx, (int __user *)arg))
1115 case IOC_MDC_LOOKUP: {
1116 int namelen, len = 0;
1120 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1125 filename = data->ioc_inlbuf1;
1126 namelen = strlen(filename);
1128 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1129 GOTO(out_free, rc = -EINVAL);
1132 rc = ll_get_fid_by_name(inode, filename, namelen, NULL);
1134 CERROR("%s: lookup %.*s failed: rc = %d\n",
1135 ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1140 obd_ioctl_freedata(buf, len);
1143 case LL_IOC_LMV_SETSTRIPE: {
1144 struct lmv_user_md *lum;
1153 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1158 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1159 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1160 GOTO(lmv_out_free, rc = -EINVAL);
1162 filename = data->ioc_inlbuf1;
1163 namelen = data->ioc_inllen1;
1166 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1167 GOTO(lmv_out_free, rc = -EINVAL);
1169 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1170 lumlen = data->ioc_inllen2;
1172 if (lum->lum_magic != LMV_USER_MAGIC ||
1173 lumlen != sizeof(*lum)) {
1174 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1175 filename, lum->lum_magic, lumlen, -EFAULT);
1176 GOTO(lmv_out_free, rc = -EINVAL);
1179 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1180 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1182 mode = data->ioc_type;
1184 rc = ll_dir_setdirstripe(inode, lum, filename, mode);
1186 obd_ioctl_freedata(buf, len);
1190 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1191 struct lmv_user_md lum;
1192 struct lmv_user_md __user *ulump =
1193 (struct lmv_user_md __user *)arg;
1196 if (copy_from_user(&lum, ulump, sizeof(lum)))
1199 if (lum.lum_magic != LMV_USER_MAGIC)
1202 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1206 case LL_IOC_LOV_SETSTRIPE: {
1207 struct lov_user_md_v3 lumv3;
1208 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1209 struct lov_user_md_v1 __user *lumv1p =
1210 (struct lov_user_md_v1 __user *)arg;
1211 struct lov_user_md_v3 __user *lumv3p =
1212 (struct lov_user_md_v3 __user *)arg;
1214 int set_default = 0;
1216 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1217 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1218 sizeof(lumv3p->lmm_objects[0]));
1219 /* first try with v1 which is smaller than v3 */
1220 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1223 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1224 if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1228 if (inode->i_sb->s_root == file->f_dentry)
1231 /* in v1 and v3 cases lumv1 points to data */
1232 rc = ll_dir_setstripe(inode, lumv1, set_default);
1236 case LL_IOC_LMV_GETSTRIPE: {
1237 struct lmv_user_md __user *ulmv =
1238 (struct lmv_user_md __user *)arg;
1239 struct lmv_user_md lum;
1240 struct ptlrpc_request *request = NULL;
1241 union lmv_mds_md *lmm = NULL;
1244 struct lmv_user_md *tmp = NULL;
1251 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1254 /* lum_magic will indicate which stripe the ioctl will like
1255 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1256 * is for default LMV stripe */
1257 if (lum.lum_magic == LMV_MAGIC_V1)
1258 valid |= OBD_MD_MEA;
1259 else if (lum.lum_magic == LMV_USER_MAGIC)
1260 valid |= OBD_MD_DEFAULT_MEA;
1264 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1267 GOTO(finish_req, rc);
1269 /* Get default LMV EA */
1270 if (lum.lum_magic == LMV_USER_MAGIC) {
1272 GOTO(finish_req, rc);
1274 if (lmmsize > sizeof(*ulmv))
1275 GOTO(finish_req, rc = -EINVAL);
1277 if (copy_to_user(ulmv, lmm, lmmsize))
1278 GOTO(finish_req, rc = -EFAULT);
1280 GOTO(finish_req, rc);
1283 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1284 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1285 OBD_ALLOC(tmp, lum_size);
1287 GOTO(finish_req, rc = -ENOMEM);
1289 mdt_index = ll_get_mdt_idx(inode);
1291 GOTO(out_tmp, rc = -ENOMEM);
1293 tmp->lum_magic = LMV_MAGIC_V1;
1294 tmp->lum_stripe_count = 0;
1295 tmp->lum_stripe_offset = mdt_index;
1296 for (i = 0; i < stripe_count; i++) {
1299 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1300 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1302 GOTO(out_tmp, rc = mdt_index);
1304 tmp->lum_objects[i].lum_mds = mdt_index;
1305 tmp->lum_objects[i].lum_fid = fid;
1306 tmp->lum_stripe_count++;
1309 if (copy_to_user(ulmv, tmp, lum_size))
1310 GOTO(out_tmp, rc = -EFAULT);
1312 OBD_FREE(tmp, lum_size);
1314 ptlrpc_req_finished(request);
1318 case LL_IOC_REMOVE_ENTRY: {
1319 char *filename = NULL;
1323 /* Here is a little hack to avoid sending REINT_RMENTRY to
1324 * unsupported server, which might crash the server(LU-2730),
1325 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1326 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1327 * server will support REINT_RMENTRY XXX*/
1328 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1331 filename = ll_getname((const char __user *)arg);
1332 if (IS_ERR(filename))
1333 RETURN(PTR_ERR(filename));
1335 namelen = strlen(filename);
1337 GOTO(out_rmdir, rc = -EINVAL);
1339 rc = ll_rmdir_entry(inode, filename, namelen);
1342 ll_putname(filename);
1345 case LL_IOC_LOV_SWAP_LAYOUTS:
1347 case IOC_OBD_STATFS:
1348 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1349 case LL_IOC_LOV_GETSTRIPE:
1350 case LL_IOC_MDC_GETINFO:
1351 case IOC_MDC_GETFILEINFO:
1352 case IOC_MDC_GETFILESTRIPE: {
1353 struct ptlrpc_request *request = NULL;
1354 struct lov_user_md __user *lump;
1355 struct lov_mds_md *lmm = NULL;
1356 struct mdt_body *body;
1357 char *filename = NULL;
1360 if (cmd == IOC_MDC_GETFILEINFO ||
1361 cmd == IOC_MDC_GETFILESTRIPE) {
1362 filename = ll_getname((const char __user *)arg);
1363 if (IS_ERR(filename))
1364 RETURN(PTR_ERR(filename));
1366 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1367 &lmmsize, &request);
1369 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1374 body = req_capsule_server_get(&request->rq_pill,
1376 LASSERT(body != NULL);
1382 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1383 cmd == LL_IOC_MDC_GETINFO))
1384 GOTO(skip_lmm, rc = 0);
1389 if (cmd == IOC_MDC_GETFILESTRIPE ||
1390 cmd == LL_IOC_LOV_GETSTRIPE) {
1391 lump = (struct lov_user_md __user *)arg;
1393 struct lov_user_mds_data __user *lmdp;
1394 lmdp = (struct lov_user_mds_data __user *)arg;
1395 lump = &lmdp->lmd_lmm;
1397 if (copy_to_user(lump, lmm, lmmsize)) {
1398 if (copy_to_user(lump, lmm, sizeof(*lump)))
1399 GOTO(out_req, rc = -EFAULT);
1403 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1404 struct lov_user_mds_data __user *lmdp;
1407 st.st_dev = inode->i_sb->s_dev;
1408 st.st_mode = body->mbo_mode;
1409 st.st_nlink = body->mbo_nlink;
1410 st.st_uid = body->mbo_uid;
1411 st.st_gid = body->mbo_gid;
1412 st.st_rdev = body->mbo_rdev;
1413 st.st_size = body->mbo_size;
1414 st.st_blksize = PAGE_CACHE_SIZE;
1415 st.st_blocks = body->mbo_blocks;
1416 st.st_atime = body->mbo_atime;
1417 st.st_mtime = body->mbo_mtime;
1418 st.st_ctime = body->mbo_ctime;
1419 st.st_ino = inode->i_ino;
1421 lmdp = (struct lov_user_mds_data __user *)arg;
1422 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1423 GOTO(out_req, rc = -EFAULT);
1428 ptlrpc_req_finished(request);
1430 ll_putname(filename);
1433 case OBD_IOC_QUOTACHECK: {
1434 struct obd_quotactl *oqctl;
1437 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1438 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1441 OBD_ALLOC_PTR(oqctl);
1444 oqctl->qc_type = arg;
1445 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1447 CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1451 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1453 CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1455 OBD_FREE_PTR(oqctl);
1458 case OBD_IOC_POLL_QUOTACHECK: {
1459 struct if_quotacheck *check;
1461 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1462 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1465 OBD_ALLOC_PTR(check);
1469 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1472 CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1473 if (copy_to_user((void __user *)arg, check,
1475 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1479 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1482 CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1483 if (copy_to_user((void __user *)arg, check,
1485 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1489 OBD_FREE_PTR(check);
1492 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 53, 0)
1493 case LL_IOC_QUOTACTL_18: {
1494 /* copy the old 1.x quota struct for internal use, then copy
1495 * back into old format struct. For 1.8 compatibility. */
1496 struct if_quotactl_18 *qctl_18;
1497 struct if_quotactl *qctl_20;
1499 OBD_ALLOC_PTR(qctl_18);
1503 OBD_ALLOC_PTR(qctl_20);
1505 GOTO(out_quotactl_18, rc = -ENOMEM);
1507 if (copy_from_user(qctl_18, (void __user *)arg,
1509 GOTO(out_quotactl_20, rc = -ENOMEM);
1511 QCTL_COPY(qctl_20, qctl_18);
1512 qctl_20->qc_idx = 0;
1514 /* XXX: dqb_valid was borrowed as a flag to mark that
1515 * only mds quota is wanted */
1516 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1517 qctl_18->qc_dqblk.dqb_valid) {
1518 qctl_20->qc_valid = QC_MDTIDX;
1519 qctl_20->qc_dqblk.dqb_valid = 0;
1520 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1521 qctl_20->qc_valid = QC_UUID;
1522 qctl_20->obd_uuid = qctl_18->obd_uuid;
1524 qctl_20->qc_valid = QC_GENERAL;
1527 rc = quotactl_ioctl(sbi, qctl_20);
1530 QCTL_COPY(qctl_18, qctl_20);
1531 qctl_18->obd_uuid = qctl_20->obd_uuid;
1533 if (copy_to_user((void __user *)arg, qctl_18,
1539 OBD_FREE_PTR(qctl_20);
1541 OBD_FREE_PTR(qctl_18);
1544 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 53, 0) */
1545 case OBD_IOC_QUOTACTL: {
1546 struct if_quotactl *qctl;
1548 OBD_ALLOC_PTR(qctl);
1552 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1553 GOTO(out_quotactl, rc = -EFAULT);
1555 rc = quotactl_ioctl(sbi, qctl);
1558 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1565 case OBD_IOC_GETDTNAME:
1566 case OBD_IOC_GETMDNAME:
1567 RETURN(ll_get_obd_name(inode, cmd, arg));
1568 case LL_IOC_FLUSHCTX:
1569 RETURN(ll_flush_ctx(inode));
1570 #ifdef CONFIG_FS_POSIX_ACL
1571 case LL_IOC_RMTACL: {
1572 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1573 inode == inode->i_sb->s_root->d_inode) {
1574 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1576 LASSERT(fd != NULL);
1577 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1579 fd->fd_flags |= LL_FILE_RMTACL;
1585 case LL_IOC_GETOBDCOUNT: {
1587 struct obd_export *exp;
1589 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1592 /* get ost count when count is zero, get mdt count otherwise */
1593 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1594 vallen = sizeof(count);
1595 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1596 KEY_TGT_COUNT, &vallen, &count);
1598 CERROR("get target count failed: %d\n", rc);
1602 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1607 case LL_IOC_PATH2FID:
1608 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1609 sizeof(struct lu_fid)))
1612 case LL_IOC_GET_CONNECT_FLAGS: {
1613 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1614 (void __user *)arg));
1616 case OBD_IOC_CHANGELOG_SEND:
1617 case OBD_IOC_CHANGELOG_CLEAR:
1618 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1619 sizeof(struct ioc_changelog));
1621 case OBD_IOC_FID2PATH:
1622 RETURN(ll_fid2path(inode, (void __user *)arg));
1623 case LL_IOC_GETPARENT:
1624 RETURN(ll_getparent(file, (void __user *)arg));
1625 case LL_IOC_FID2MDTIDX: {
1626 struct obd_export *exp = ll_i2mdexp(inode);
1630 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1634 /* Call mdc_iocontrol */
1635 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1636 (__u32 __user *)&index);
1642 case LL_IOC_HSM_REQUEST: {
1643 struct hsm_user_request *hur;
1650 /* We don't know the true size yet; copy the fixed-size part */
1651 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1656 /* Compute the whole struct size */
1657 totalsize = hur_len(hur);
1662 /* Final size will be more than double totalsize */
1663 if (totalsize >= MDS_MAXREQSIZE / 3)
1666 OBD_ALLOC_LARGE(hur, totalsize);
1670 /* Copy the whole struct */
1671 if (copy_from_user(hur, (void __user *)arg, totalsize)) {
1672 OBD_FREE_LARGE(hur, totalsize);
1676 if (hur->hur_request.hr_action == HUA_RELEASE) {
1677 const struct lu_fid *fid;
1681 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1682 fid = &hur->hur_user_item[i].hui_fid;
1683 f = search_inode_for_lustre(inode->i_sb, fid);
1689 rc = ll_hsm_release(f);
1695 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1699 OBD_FREE_LARGE(hur, totalsize);
1703 case LL_IOC_HSM_PROGRESS: {
1704 struct hsm_progress_kernel hpk;
1705 struct hsm_progress hp;
1707 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1710 hpk.hpk_fid = hp.hp_fid;
1711 hpk.hpk_cookie = hp.hp_cookie;
1712 hpk.hpk_extent = hp.hp_extent;
1713 hpk.hpk_flags = hp.hp_flags;
1714 hpk.hpk_errval = hp.hp_errval;
1715 hpk.hpk_data_version = 0;
1717 /* File may not exist in Lustre; all progress
1718 * reported to Lustre root */
1719 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1723 case LL_IOC_HSM_CT_START:
1724 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1727 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1728 sizeof(struct lustre_kernelcomm));
1731 case LL_IOC_HSM_COPY_START: {
1732 struct hsm_copy *copy;
1735 OBD_ALLOC_PTR(copy);
1738 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1743 rc = ll_ioc_copy_start(inode->i_sb, copy);
1744 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1750 case LL_IOC_HSM_COPY_END: {
1751 struct hsm_copy *copy;
1754 OBD_ALLOC_PTR(copy);
1757 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1762 rc = ll_ioc_copy_end(inode->i_sb, copy);
1763 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1769 case LL_IOC_MIGRATE: {
1771 const char *filename;
1777 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1781 data = (struct obd_ioctl_data *)buf;
1782 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1783 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1784 GOTO(migrate_free, rc = -EINVAL);
1786 filename = data->ioc_inlbuf1;
1787 namelen = data->ioc_inllen1;
1788 /* \0 is packed at the end of filename */
1789 if (namelen < 1 || namelen != strlen(filename) + 1)
1790 GOTO(migrate_free, rc = -EINVAL);
1792 if (data->ioc_inllen2 != sizeof(mdtidx))
1793 GOTO(migrate_free, rc = -EINVAL);
1794 mdtidx = *(int *)data->ioc_inlbuf2;
1796 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1798 obd_ioctl_freedata(buf, len);
1803 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1804 (void __user *)arg));
1808 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1810 struct inode *inode = file->f_mapping->host;
1811 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1812 struct ll_sb_info *sbi = ll_i2sbi(inode);
1813 int api32 = ll_need_32bit_api(sbi);
1814 loff_t ret = -EINVAL;
1817 mutex_lock(&inode->i_mutex);
1822 offset += file->f_pos;
1828 offset += LL_DIR_END_OFF_32BIT;
1830 offset += LL_DIR_END_OFF;
1837 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1838 (!api32 && offset <= LL_DIR_END_OFF))) {
1839 if (offset != file->f_pos) {
1840 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1841 (!api32 && offset == LL_DIR_END_OFF))
1842 fd->lfd_pos = MDS_DIR_END_OFF;
1843 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1844 fd->lfd_pos = offset << 32;
1846 fd->lfd_pos = offset;
1847 file->f_pos = offset;
1848 file->f_version = 0;
1855 mutex_unlock(&inode->i_mutex);
1859 static int ll_dir_open(struct inode *inode, struct file *file)
1862 RETURN(ll_file_open(inode, file));
1865 static int ll_dir_release(struct inode *inode, struct file *file)
1868 RETURN(ll_file_release(inode, file));
1871 const struct file_operations ll_dir_operations = {
1872 .llseek = ll_dir_seek,
1873 .open = ll_dir_open,
1874 .release = ll_dir_release,
1875 .read = generic_read_dir,
1876 #ifdef HAVE_DIR_CONTEXT
1877 .iterate = ll_iterate,
1879 .readdir = ll_readdir,
1881 .unlocked_ioctl = ll_dir_ioctl,