X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fllite%2Ffile.c;h=14e38cf7050c35306ea957d2579feb1ce9498d90;hp=c186b5a20cd8ad712eba620c1dd2af77ddc9d78e;hb=a7811634d45319dc3812fa652c727367fa290a24;hpb=255e37f1639fa4edec5b929228afe7c0e8b56724 diff --git a/lustre/llite/file.c b/lustre/llite/file.c index c186b5a..14e38cf 100644 --- a/lustre/llite/file.c +++ b/lustre/llite/file.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -28,6 +26,8 @@ /* * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2012, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -43,7 +43,6 @@ #define DEBUG_SUBSYSTEM S_LLITE #include #include -#include #include #include #include "llite_internal.h" @@ -53,10 +52,11 @@ struct ll_file_data *ll_file_data_get(void) { - struct ll_file_data *fd; + struct ll_file_data *fd; - OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, CFS_ALLOC_IO); - return fd; + OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, CFS_ALLOC_IO); + fd->fd_write_failed = false; + return fd; } static void ll_file_data_put(struct ll_file_data *fd) @@ -81,6 +81,9 @@ void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data, if (fh) op_data->op_handle = *fh; op_data->op_capa1 = ll_mdscapa_get(inode); + + if (LLIF_DATA_MODIFIED & ll_i2info(inode)->lli_flags) + op_data->op_bias |= MDS_DATA_MODIFIED; } /** @@ -154,6 +157,17 @@ static int ll_close_inode_openhandle(struct obd_export *md_exp, CERROR("inode %lu mdc close failed: rc = %d\n", inode->i_ino, rc); } + + /* DATA_MODIFIED flag was successfully sent on close, cancel data + * modification flag. */ + if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) { + struct ll_inode_info *lli = ll_i2info(inode); + + spin_lock(&lli->lli_lock); + lli->lli_flags &= ~LLIF_DATA_MODIFIED; + spin_unlock(&lli->lli_lock); + } + ll_finish_md_op_data(op_data); if (rc == 0) { @@ -201,15 +215,15 @@ int ll_md_real_close(struct inode *inode, int flags) och_usecount = &lli->lli_open_fd_read_count; } - cfs_down(&lli->lli_och_sem); + mutex_lock(&lli->lli_och_mutex); if (*och_usecount) { /* There are still users of this handle, so skip freeing it. */ - cfs_up(&lli->lli_och_sem); + mutex_unlock(&lli->lli_och_mutex); RETURN(0); } och=*och_p; *och_p = NULL; - cfs_up(&lli->lli_och_sem); + mutex_unlock(&lli->lli_och_mutex); if (och) { /* There might be a race and somebody have freed this och already */ @@ -241,7 +255,7 @@ int ll_md_close(struct obd_export *md_exp, struct inode *inode, struct inode *inode = file->f_dentry->d_inode; ldlm_policy_data_t policy = {.l_inodebits={MDS_INODELOCK_OPEN}}; - cfs_down(&lli->lli_och_sem); + mutex_lock(&lli->lli_och_mutex); if (fd->fd_omode & FMODE_WRITE) { lockmode = LCK_CW; LASSERT(lli->lli_open_fd_write_count); @@ -255,7 +269,7 @@ int ll_md_close(struct obd_export *md_exp, struct inode *inode, LASSERT(lli->lli_open_fd_read_count); lli->lli_open_fd_read_count--; } - cfs_up(&lli->lli_och_sem); + mutex_unlock(&lli->lli_och_mutex); if (!md_lock_match(md_exp, flags, ll_inode2fid(inode), LDLM_IBITS, &policy, lockmode, @@ -275,8 +289,6 @@ int ll_md_close(struct obd_export *md_exp, struct inode *inode, RETURN(rc); } -int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm); - /* While this returns an error code, fput() the caller does not, so we need * to make every effort to clean up all of our state here. Also, applications * rarely check close errors and even if an error is returned they will not @@ -287,7 +299,6 @@ int ll_file_release(struct inode *inode, struct file *file) struct ll_file_data *fd; struct ll_sb_info *sbi = ll_i2sbi(inode); struct ll_inode_info *lli = ll_i2info(inode); - struct lov_stripe_md *lsm = lli->lli_smd; int rc; ENTRY; @@ -316,7 +327,8 @@ int ll_file_release(struct inode *inode, struct file *file) /* The last ref on @file, maybe not the the owner pid of statahead. * Different processes can open the same dir, "ll_opendir_key" means: * it is me that should stop the statahead thread. */ - if (lli->lli_opendir_key == fd && lli->lli_opendir_pid != 0) + if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd && + lli->lli_opendir_pid != 0) ll_stop_statahead(inode, lli->lli_opendir_key); if (inode->i_sb->s_root == file->f_dentry) { @@ -325,9 +337,10 @@ int ll_file_release(struct inode *inode, struct file *file) RETURN(0); } - if (lsm) - lov_test_and_clear_async_rc(lsm); - lli->lli_async_rc = 0; + if (!S_ISDIR(inode->i_mode)) { + lov_read_and_clear_async_rc(lli->lli_clob); + lli->lli_async_rc = 0; + } rc = ll_md_close(sbi->ll_md_exp, inode, file); @@ -374,6 +387,7 @@ static int ll_intent_file_open(struct file *file, void *lmm, if (IS_ERR(op_data)) RETURN(PTR_ERR(op_data)); + itp->it_flags |= MDS_OPEN_BY_FID; rc = md_intent_lock(sbi->ll_md_exp, op_data, lmm, lmmsize, itp, 0 /*unused */, &req, ll_md_blocking_ast, 0); ll_finish_md_op_data(op_data); @@ -388,17 +402,19 @@ static int ll_intent_file_open(struct file *file, void *lmm, GOTO(out, rc); } + if (it_disposition(itp, DISP_LOOKUP_NEG)) + GOTO(out, rc = -ENOENT); + if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) { rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp); CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc); GOTO(out, rc); } - rc = ll_prep_inode(&file->f_dentry->d_inode, req, NULL); + rc = ll_prep_inode(&file->f_dentry->d_inode, req, NULL, itp); if (!rc && itp->d.lustre.it_lock_mode) - md_set_lock_data(sbi->ll_md_exp, - &itp->d.lustre.it_lock_handle, - file->f_dentry->d_inode, NULL); + ll_set_lock_data(sbi->ll_md_exp, file->f_dentry->d_inode, + itp, NULL); out: ptlrpc_req_finished(itp->d.lustre.it_data); @@ -477,9 +493,7 @@ int ll_local_open(struct file *file, struct lookup_intent *it, /* Open a file, and (for the very first open) create objects on the OSTs at * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object - * creation or open until ll_lov_setstripe() ioctl is called. We grab - * lli_open_sem to ensure no other process will create objects, send the - * stripe MD to the MDS, or try to destroy the objects if that fails. + * creation or open until ll_lov_setstripe() ioctl is called. * * If we already have the stripe MD locally then we don't request it in * md_open(), by passing a lmm_size = 0. @@ -495,10 +509,8 @@ int ll_file_open(struct inode *inode, struct file *file) struct ll_inode_info *lli = ll_i2info(inode); struct lookup_intent *it, oit = { .it_op = IT_OPEN, .it_flags = file->f_flags }; - struct lov_stripe_md *lsm; - struct ptlrpc_request *req = NULL; - struct obd_client_handle **och_p; - __u64 *och_usecount; + struct obd_client_handle **och_p = NULL; + __u64 *och_usecount = NULL; struct ll_file_data *fd; int rc = 0, opendir_set = 0; ENTRY; @@ -511,18 +523,18 @@ int ll_file_open(struct inode *inode, struct file *file) fd = ll_file_data_get(); if (fd == NULL) - RETURN(-ENOMEM); + GOTO(out_och_free, rc = -ENOMEM); fd->fd_file = file; if (S_ISDIR(inode->i_mode)) { - cfs_spin_lock(&lli->lli_sa_lock); - if (lli->lli_opendir_key == NULL && lli->lli_opendir_pid == 0) { - LASSERT(lli->lli_sai == NULL); - lli->lli_opendir_key = fd; - lli->lli_opendir_pid = cfs_curproc_pid(); - opendir_set = 1; - } - cfs_spin_unlock(&lli->lli_sa_lock); + spin_lock(&lli->lli_sa_lock); + if (lli->lli_opendir_key == NULL && lli->lli_sai == NULL && + lli->lli_opendir_pid == 0) { + lli->lli_opendir_key = fd; + lli->lli_opendir_pid = cfs_curproc_pid(); + opendir_set = 1; + } + spin_unlock(&lli->lli_sa_lock); } if (inode->i_sb->s_root == file->f_dentry) { @@ -572,28 +584,25 @@ restart: och_usecount = &lli->lli_open_fd_read_count; } - cfs_down(&lli->lli_och_sem); + mutex_lock(&lli->lli_och_mutex); if (*och_p) { /* Open handle is present */ if (it_disposition(it, DISP_OPEN_OPEN)) { /* Well, there's extra open request that we do not need, let's close it somehow. This will decref request. */ rc = it_open_error(DISP_OPEN_OPEN, it); if (rc) { - cfs_up(&lli->lli_och_sem); - ll_file_data_put(fd); + mutex_unlock(&lli->lli_och_mutex); GOTO(out_openerr, rc); } + ll_release_openhandle(file->f_dentry, it); - lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, - LPROC_LL_OPEN); } (*och_usecount)++; rc = ll_local_open(file, it, fd, NULL); if (rc) { (*och_usecount)--; - cfs_up(&lli->lli_och_sem); - ll_file_data_put(fd); + mutex_unlock(&lli->lli_och_mutex); GOTO(out_openerr, rc); } } else { @@ -602,31 +611,22 @@ restart: /* We cannot just request lock handle now, new ELC code means that one of other OPEN locks for this file could be cancelled, and since blocking ast handler - would attempt to grab och_sem as well, that would + would attempt to grab och_mutex as well, that would result in a deadlock */ - cfs_up(&lli->lli_och_sem); + mutex_unlock(&lli->lli_och_mutex); it->it_create_mode |= M_CHECK_STALE; rc = ll_intent_file_open(file, NULL, 0, it); it->it_create_mode &= ~M_CHECK_STALE; - if (rc) { - ll_file_data_put(fd); + if (rc) GOTO(out_openerr, rc); - } - /* Got some error? Release the request */ - if (it->d.lustre.it_status < 0) { - req = it->d.lustre.it_data; - ptlrpc_req_finished(req); - } goto restart; } OBD_ALLOC(*och_p, sizeof (struct obd_client_handle)); - if (!*och_p) { - ll_file_data_put(fd); + if (!*och_p) GOTO(out_och_free, rc = -ENOMEM); - } + (*och_usecount)++; - req = it->d.lustre.it_data; /* md_intent_lock() didn't get a request ref if there was an * open error, so don't do cleanup on the request here @@ -634,55 +634,59 @@ restart: /* XXX (green): Should not we bail out on any error here, not * just open error? */ rc = it_open_error(DISP_OPEN_OPEN, it); - if (rc) { - ll_file_data_put(fd); + if (rc) GOTO(out_och_free, rc); - } - ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1); + LASSERT(it_disposition(it, DISP_ENQ_OPEN_REF)); + rc = ll_local_open(file, it, fd, *och_p); - if (rc) { - ll_file_data_put(fd); + if (rc) GOTO(out_och_free, rc); - } } - cfs_up(&lli->lli_och_sem); + mutex_unlock(&lli->lli_och_mutex); + fd = NULL; - /* Must do this outside lli_och_sem lock to prevent deadlock where + /* Must do this outside lli_och_mutex lock to prevent deadlock where different kind of OPEN lock for this same inode gets cancelled by ldlm_cancel_lru */ if (!S_ISREG(inode->i_mode)) - GOTO(out, rc); + GOTO(out_och_free, rc); ll_capa_open(inode); - lsm = lli->lli_smd; - if (lsm == NULL) { + if (!lli->lli_has_smd) { if (file->f_flags & O_LOV_DELAY_CREATE || !(file->f_mode & FMODE_WRITE)) { CDEBUG(D_INODE, "object creation was delayed\n"); - GOTO(out, rc); + GOTO(out_och_free, rc); } } file->f_flags &= ~O_LOV_DELAY_CREATE; - GOTO(out, rc); -out: - ptlrpc_req_finished(req); - if (req) - it_clear_disposition(it, DISP_ENQ_OPEN_REF); + GOTO(out_och_free, rc); + out_och_free: if (rc) { - if (*och_p) { + if (och_p && *och_p) { OBD_FREE(*och_p, sizeof (struct obd_client_handle)); *och_p = NULL; /* OBD_FREE writes some magic there */ (*och_usecount)--; } - cfs_up(&lli->lli_och_sem); + mutex_unlock(&lli->lli_och_mutex); + out_openerr: if (opendir_set != 0) ll_stop_statahead(inode, lli->lli_opendir_key); + if (fd != NULL) + ll_file_data_put(fd); + } else { + ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1); } + if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) { + ptlrpc_req_finished(it->d.lustre.it_data); + it_clear_disposition(it, DISP_ENQ_OPEN_REF); + } + return rc; } @@ -701,15 +705,15 @@ static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp, oinfo.oi_md = lsm; oinfo.oi_oa = obdo; - oinfo.oi_oa->o_id = lsm->lsm_object_id; - oinfo.oi_oa->o_seq = lsm->lsm_object_seq; + oinfo.oi_oa->o_oi = lsm->lsm_oi; oinfo.oi_oa->o_mode = S_IFREG; oinfo.oi_oa->o_ioepoch = ioepoch; oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ | OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME | - OBD_MD_FLGROUP | OBD_MD_FLEPOCH; + OBD_MD_FLGROUP | OBD_MD_FLEPOCH | + OBD_MD_FLDATAVERSION; oinfo.oi_capa = capa; if (sync) { oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS; @@ -729,7 +733,8 @@ static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp, if (rc == 0) oinfo.oi_oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ | OBD_MD_FLATIME | OBD_MD_FLMTIME | - OBD_MD_FLCTIME | OBD_MD_FLSIZE); + OBD_MD_FLCTIME | OBD_MD_FLSIZE | + OBD_MD_FLDATAVERSION); RETURN(rc); } @@ -740,55 +745,71 @@ static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp, int ll_inode_getattr(struct inode *inode, struct obdo *obdo, __u64 ioepoch, int sync) { - struct ll_inode_info *lli = ll_i2info(inode); - struct obd_capa *capa = ll_mdscapa_get(inode); - int rc; - ENTRY; - - rc = ll_lsm_getattr(lli->lli_smd, ll_i2dtexp(inode), - capa, obdo, ioepoch, sync); - capa_put(capa); - if (rc == 0) { - obdo_refresh_inode(inode, obdo, obdo->o_valid); - CDEBUG(D_INODE, - "objid "LPX64" size %Lu, blocks %llu, blksize %lu\n", - lli->lli_smd->lsm_object_id, i_size_read(inode), - (unsigned long long)inode->i_blocks, - (unsigned long)ll_inode_blksize(inode)); - } - RETURN(rc); -} - -int ll_merge_lvb(struct inode *inode) -{ - struct ll_inode_info *lli = ll_i2info(inode); - struct ll_sb_info *sbi = ll_i2sbi(inode); - struct ost_lvb lvb; - int rc; - - ENTRY; - - ll_inode_size_lock(inode, 1); - inode_init_lvb(inode, &lvb); - - /* merge timestamps the most resently obtained from mds with - timestamps obtained from osts */ - lvb.lvb_atime = lli->lli_lvb.lvb_atime; - lvb.lvb_mtime = lli->lli_lvb.lvb_mtime; - lvb.lvb_ctime = lli->lli_lvb.lvb_ctime; - rc = obd_merge_lvb(sbi->ll_dt_exp, lli->lli_smd, &lvb, 0); - cl_isize_write_nolock(inode, lvb.lvb_size); - - CDEBUG(D_VFSTRACE, DFID" updating i_size "LPU64"\n", - PFID(&lli->lli_fid), lvb.lvb_size); - inode->i_blocks = lvb.lvb_blocks; - - LTIME_S(inode->i_mtime) = lvb.lvb_mtime; - LTIME_S(inode->i_atime) = lvb.lvb_atime; - LTIME_S(inode->i_ctime) = lvb.lvb_ctime; - ll_inode_size_unlock(inode, 1); - - RETURN(rc); + struct obd_capa *capa = ll_mdscapa_get(inode); + struct lov_stripe_md *lsm; + int rc; + ENTRY; + + lsm = ccc_inode_lsm_get(inode); + rc = ll_lsm_getattr(lsm, ll_i2dtexp(inode), + capa, obdo, ioepoch, sync); + capa_put(capa); + if (rc == 0) { + struct ost_id *oi = lsm ? &lsm->lsm_oi : &obdo->o_oi; + + obdo_refresh_inode(inode, obdo, obdo->o_valid); + CDEBUG(D_INODE, "objid "DOSTID" size %llu, blocks %llu," + " blksize %lu\n", POSTID(oi), i_size_read(inode), + (unsigned long long)inode->i_blocks, + (unsigned long)ll_inode_blksize(inode)); + } + ccc_inode_lsm_put(inode, lsm); + RETURN(rc); +} + +int ll_merge_lvb(const struct lu_env *env, struct inode *inode) +{ + struct ll_inode_info *lli = ll_i2info(inode); + struct cl_object *obj = lli->lli_clob; + struct cl_attr *attr = ccc_env_thread_attr(env); + struct ost_lvb lvb; + int rc = 0; + + ENTRY; + + ll_inode_size_lock(inode); + /* merge timestamps the most recently obtained from mds with + timestamps obtained from osts */ + LTIME_S(inode->i_atime) = lli->lli_lvb.lvb_atime; + LTIME_S(inode->i_mtime) = lli->lli_lvb.lvb_mtime; + LTIME_S(inode->i_ctime) = lli->lli_lvb.lvb_ctime; + inode_init_lvb(inode, &lvb); + + cl_object_attr_lock(obj); + rc = cl_object_attr_get(env, obj, attr); + cl_object_attr_unlock(obj); + + if (rc == 0) { + if (lvb.lvb_atime < attr->cat_atime) + lvb.lvb_atime = attr->cat_atime; + if (lvb.lvb_ctime < attr->cat_ctime) + lvb.lvb_ctime = attr->cat_ctime; + if (lvb.lvb_mtime < attr->cat_mtime) + lvb.lvb_mtime = attr->cat_mtime; + + CDEBUG(D_VFSTRACE, DFID" updating i_size "LPU64"\n", + PFID(&lli->lli_fid), attr->cat_size); + cl_isize_write_nolock(inode, attr->cat_size); + + inode->i_blocks = attr->cat_blocks; + + LTIME_S(inode->i_mtime) = lvb.lvb_mtime; + LTIME_S(inode->i_atime) = lvb.lvb_atime; + LTIME_S(inode->i_ctime) = lvb.lvb_ctime; + } + ll_inode_size_unlock(inode); + + RETURN(rc); } int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm, @@ -813,8 +834,12 @@ void ll_io_init(struct cl_io *io, const struct file *file, int write) struct inode *inode = file->f_dentry->d_inode; io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK; - if (write) - io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND); + if (write) { + io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND); + io->u.ci_wr.wr_sync = file->f_flags & O_SYNC || + file->f_flags & O_DIRECT || + IS_SYNC(inode); + } io->ci_obj = ll_i2info(inode)->lli_clob; io->ci_lockreq = CILR_MAYBE; if (ll_file_nolock(file)) { @@ -825,22 +850,25 @@ void ll_io_init(struct cl_io *io, const struct file *file, int write) } } -static ssize_t ll_file_io_generic(const struct lu_env *env, - struct vvp_io_args *args, struct file *file, - enum cl_io_type iot, loff_t *ppos, size_t count) +static ssize_t +ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args, + struct file *file, enum cl_io_type iot, + loff_t *ppos, size_t count) { - struct ll_inode_info *lli = ll_i2info(file->f_dentry->d_inode); + struct ll_inode_info *lli = ll_i2info(file->f_dentry->d_inode); + struct ll_file_data *fd = LUSTRE_FPRIVATE(file); struct cl_io *io; ssize_t result; ENTRY; +restart: io = ccc_env_thread_io(env); ll_io_init(io, file, iot == CIT_WRITE); if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) { struct vvp_io *vio = vvp_env_io(env); struct ccc_io *cio = ccc_env_io(env); - int write_sem_locked = 0; + int write_mutex_locked = 0; cio->cui_fd = LUSTRE_FPRIVATE(file); vio->cui_io_subtype = args->via_io_subtype; @@ -855,11 +883,12 @@ static ssize_t ll_file_io_generic(const struct lu_env *env, #endif if ((iot == CIT_WRITE) && !(cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) { - if(cfs_down_interruptible(&lli->lli_write_sem)) + if (mutex_lock_interruptible(&lli-> + lli_write_mutex)) GOTO(out, result = -ERESTARTSYS); - write_sem_locked = 1; + write_mutex_locked = 1; } else if (iot == CIT_READ) { - cfs_down_read(&lli->lli_trunc_sem); + down_read(&lli->lli_trunc_sem); } break; case IO_SENDFILE: @@ -875,10 +904,10 @@ static ssize_t ll_file_io_generic(const struct lu_env *env, LBUG(); } result = cl_io_loop(env, io); - if (write_sem_locked) - cfs_up(&lli->lli_write_sem); + if (write_mutex_locked) + mutex_unlock(&lli->lli_write_mutex); else if (args->via_io_subtype == IO_NORMAL && iot == CIT_READ) - cfs_up_read(&lli->lli_trunc_sem); + up_read(&lli->lli_trunc_sem); } else { /* cl_io_rw_init() handled IO */ result = io->ci_result; @@ -891,9 +920,32 @@ static ssize_t ll_file_io_generic(const struct lu_env *env, GOTO(out, result); out: cl_io_fini(env, io); - if (iot == CIT_WRITE) - lli->lli_write_rc = result < 0 ? : 0; - return result; + /* If any bit been read/written (result != 0), we just return + * short read/write instead of restart io. */ + if (result == 0 && io->ci_need_restart) { + CDEBUG(D_VFSTRACE, "Restart %s on %s from %lld, count:%zd\n", + iot == CIT_READ ? "read" : "write", + file->f_dentry->d_name.name, *ppos, count); + LASSERTF(io->u.ci_rw.crw_count == count, "%zd != %zd\n", + io->u.ci_rw.crw_count, count); + goto restart; + } + + if (iot == CIT_READ) { + if (result >= 0) + ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode), + LPROC_LL_READ_BYTES, result); + } else if (iot == CIT_WRITE) { + if (result >= 0) { + ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode), + LPROC_LL_WRITE_BYTES, result); + fd->fd_write_failed = false; + } else if (result != -ERESTARTSYS) { + fd->fd_write_failed = true; + } + } + + return result; } @@ -1209,22 +1261,22 @@ static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos, static int ll_lov_recreate(struct inode *inode, obd_id id, obd_seq seq, obd_count ost_idx) { - struct obd_export *exp = ll_i2dtexp(inode); - struct obd_trans_info oti = { 0 }; - struct obdo *oa = NULL; - int lsm_size; - int rc = 0; - struct lov_stripe_md *lsm, *lsm2; - ENTRY; + struct obd_export *exp = ll_i2dtexp(inode); + struct obd_trans_info oti = { 0 }; + struct obdo *oa = NULL; + int lsm_size; + int rc = 0; + struct lov_stripe_md *lsm = NULL, *lsm2; + ENTRY; - OBDO_ALLOC(oa); - if (oa == NULL) - RETURN(-ENOMEM); + OBDO_ALLOC(oa); + if (oa == NULL) + RETURN(-ENOMEM); - ll_inode_size_lock(inode, 0); - lsm = ll_i2info(inode)->lli_smd; + lsm = ccc_inode_lsm_get(inode); if (lsm == NULL) GOTO(out, rc = -ENOENT); + lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) * (lsm->lsm_stripe_count)); @@ -1237,76 +1289,76 @@ static int ll_lov_recreate(struct inode *inode, obd_id id, obd_seq seq, oa->o_nlink = ost_idx; oa->o_flags |= OBD_FL_RECREATE_OBJS; oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP; - obdo_from_inode(oa, inode, &ll_i2info(inode)->lli_fid, OBD_MD_FLTYPE | - OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME); + obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME | + OBD_MD_FLMTIME | OBD_MD_FLCTIME); + obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid); memcpy(lsm2, lsm, lsm_size); - rc = obd_create(exp, oa, &lsm2, &oti); + ll_inode_size_lock(inode); + rc = obd_create(NULL, exp, oa, &lsm2, &oti); + ll_inode_size_unlock(inode); - OBD_FREE_LARGE(lsm2, lsm_size); - GOTO(out, rc); + OBD_FREE_LARGE(lsm2, lsm_size); + GOTO(out, rc); out: - ll_inode_size_unlock(inode, 0); - OBDO_FREE(oa); - return rc; + ccc_inode_lsm_put(inode, lsm); + OBDO_FREE(oa); + return rc; } static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg) { - struct ll_recreate_obj ucreat; - ENTRY; + struct ll_recreate_obj ucreat; + ENTRY; - if (!cfs_capable(CFS_CAP_SYS_ADMIN)) - RETURN(-EPERM); + if (!cfs_capable(CFS_CAP_SYS_ADMIN)) + RETURN(-EPERM); - if (cfs_copy_from_user(&ucreat, (struct ll_recreate_obj *)arg, - sizeof(struct ll_recreate_obj))) - RETURN(-EFAULT); + if (copy_from_user(&ucreat, (struct ll_recreate_obj *)arg, + sizeof(ucreat))) + RETURN(-EFAULT); - RETURN(ll_lov_recreate(inode, ucreat.lrc_id, 0, - ucreat.lrc_ost_idx)); + RETURN(ll_lov_recreate(inode, ucreat.lrc_id, 0, + ucreat.lrc_ost_idx)); } static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg) { - struct lu_fid fid; - obd_id id; - obd_count ost_idx; + struct lu_fid fid; + obd_id id; + obd_count ost_idx; ENTRY; - if (!cfs_capable(CFS_CAP_SYS_ADMIN)) - RETURN(-EPERM); + if (!cfs_capable(CFS_CAP_SYS_ADMIN)) + RETURN(-EPERM); - if (cfs_copy_from_user(&fid, (struct lu_fid *)arg, - sizeof(struct lu_fid))) - RETURN(-EFAULT); + if (copy_from_user(&fid, (struct lu_fid *)arg, sizeof(fid))) + RETURN(-EFAULT); - id = fid_oid(&fid) | ((fid_seq(&fid) & 0xffff) << 32); - ost_idx = (fid_seq(&fid) >> 16) & 0xffff; - RETURN(ll_lov_recreate(inode, id, 0, ost_idx)); + id = fid_oid(&fid) | ((fid_seq(&fid) & 0xffff) << 32); + ost_idx = (fid_seq(&fid) >> 16) & 0xffff; + RETURN(ll_lov_recreate(inode, id, 0, ost_idx)); } int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file, int flags, struct lov_user_md *lum, int lum_size) { - struct lov_stripe_md *lsm; - struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags}; - int rc = 0; - ENTRY; + struct lov_stripe_md *lsm = NULL; + struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags}; + int rc = 0; + ENTRY; - ll_inode_size_lock(inode, 0); - lsm = ll_i2info(inode)->lli_smd; - if (lsm) { - ll_inode_size_unlock(inode, 0); - CDEBUG(D_IOCTL, "stripe already exists for ino %lu\n", - inode->i_ino); - RETURN(-EEXIST); - } + lsm = ccc_inode_lsm_get(inode); + if (lsm != NULL) { + ccc_inode_lsm_put(inode, lsm); + CDEBUG(D_IOCTL, "stripe already exists for ino %lu\n", + inode->i_ino); + RETURN(-EEXIST); + } + ll_inode_size_lock(inode); rc = ll_intent_file_open(file, lum, lum_size, &oit); if (rc) GOTO(out, rc); - if (it_disposition(&oit, DISP_LOOKUP_NEG)) - GOTO(out_req_free, rc = -ENOENT); rc = oit.d.lustre.it_status; if (rc < 0) GOTO(out_req_free, rc); @@ -1314,12 +1366,13 @@ int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file, ll_release_openhandle(file->f_dentry, &oit); out: - ll_inode_size_unlock(inode, 0); - ll_intent_release(&oit); - RETURN(rc); + ll_inode_size_unlock(inode); + ll_intent_release(&oit); + ccc_inode_lsm_put(inode, lsm); + RETURN(rc); out_req_free: - ptlrpc_req_finished((struct ptlrpc_request *) oit.d.lustre.it_data); - goto out; + ptlrpc_req_finished((struct ptlrpc_request *) oit.d.lustre.it_data); + goto out; } int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename, @@ -1340,8 +1393,8 @@ int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename, op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, strlen(filename), lmmsize, LUSTRE_OPC_ANY, NULL); - if (op_data == NULL) - RETURN(-ENOMEM); + if (IS_ERR(op_data)) + RETURN(PTR_ERR(op_data)); op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA; rc = md_getattr_name(sbi->ll_md_exp, op_data, &req); @@ -1403,74 +1456,81 @@ out: static int ll_lov_setea(struct inode *inode, struct file *file, unsigned long arg) { - int flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE; - struct lov_user_md *lump; - int lum_size = sizeof(struct lov_user_md) + - sizeof(struct lov_user_ost_data); - int rc; - ENTRY; + int flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE; + struct lov_user_md *lump; + int lum_size = sizeof(struct lov_user_md) + + sizeof(struct lov_user_ost_data); + int rc; + ENTRY; - if (!cfs_capable(CFS_CAP_SYS_ADMIN)) - RETURN(-EPERM); + if (!cfs_capable(CFS_CAP_SYS_ADMIN)) + RETURN(-EPERM); - OBD_ALLOC_LARGE(lump, lum_size); - if (lump == NULL) { + OBD_ALLOC_LARGE(lump, lum_size); + if (lump == NULL) RETURN(-ENOMEM); - } - if (cfs_copy_from_user(lump, (struct lov_user_md *)arg, lum_size)) { - OBD_FREE_LARGE(lump, lum_size); - RETURN(-EFAULT); - } - rc = ll_lov_setstripe_ea_info(inode, file, flags, lump, lum_size); + if (copy_from_user(lump, (struct lov_user_md *)arg, lum_size)) { + OBD_FREE_LARGE(lump, lum_size); + RETURN(-EFAULT); + } - OBD_FREE_LARGE(lump, lum_size); - RETURN(rc); + rc = ll_lov_setstripe_ea_info(inode, file, flags, lump, lum_size); + + OBD_FREE_LARGE(lump, lum_size); + RETURN(rc); } static int ll_lov_setstripe(struct inode *inode, struct file *file, - unsigned long arg) -{ - struct lov_user_md_v3 lumv3; - struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3; - struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg; - struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg; - int lum_size; - int rc; - int flags = FMODE_WRITE; - ENTRY; - - /* first try with v1 which is smaller than v3 */ - lum_size = sizeof(struct lov_user_md_v1); - if (cfs_copy_from_user(lumv1, lumv1p, lum_size)) - RETURN(-EFAULT); - - if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) { - lum_size = sizeof(struct lov_user_md_v3); - if (cfs_copy_from_user(&lumv3, lumv3p, lum_size)) - RETURN(-EFAULT); - } - - rc = ll_lov_setstripe_ea_info(inode, file, flags, lumv1, lum_size); - if (rc == 0) { - put_user(0, &lumv1p->lmm_stripe_count); - rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), - 0, ll_i2info(inode)->lli_smd, - (void *)arg); - } - RETURN(rc); + unsigned long arg) +{ + struct lov_user_md_v3 lumv3; + struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3; + struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg; + struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg; + int lum_size, rc; + int flags = FMODE_WRITE; + ENTRY; + + /* first try with v1 which is smaller than v3 */ + lum_size = sizeof(struct lov_user_md_v1); + if (copy_from_user(lumv1, lumv1p, lum_size)) + RETURN(-EFAULT); + + if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) { + lum_size = sizeof(struct lov_user_md_v3); + if (copy_from_user(&lumv3, lumv3p, lum_size)) + RETURN(-EFAULT); + } + + rc = ll_lov_setstripe_ea_info(inode, file, flags, lumv1, lum_size); + if (rc == 0) { + struct lov_stripe_md *lsm; + __u32 gen; + + put_user(0, &lumv1p->lmm_stripe_count); + + ll_layout_refresh(inode, &gen); + lsm = ccc_inode_lsm_get(inode); + rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), + 0, lsm, (void *)arg); + ccc_inode_lsm_put(inode, lsm); + } + RETURN(rc); } static int ll_lov_getstripe(struct inode *inode, unsigned long arg) { - struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd; - int rc = -ENODATA; - ENTRY; + struct lov_stripe_md *lsm; + int rc = -ENODATA; + ENTRY; - if (lsm != NULL) - rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0, - lsm, (void *)arg); - RETURN(rc); + lsm = ccc_inode_lsm_get(inode); + if (lsm != NULL) + rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0, + lsm, (void *)arg); + ccc_inode_lsm_put(inode, lsm); + RETURN(rc); } int ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg) @@ -1484,47 +1544,47 @@ int ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg) if (ll_file_nolock(file)) RETURN(-EOPNOTSUPP); - cfs_spin_lock(&lli->lli_lock); - if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { - CWARN("group lock already existed with gid %lu\n", - fd->fd_grouplock.cg_gid); - cfs_spin_unlock(&lli->lli_lock); - RETURN(-EINVAL); - } - LASSERT(fd->fd_grouplock.cg_lock == NULL); - cfs_spin_unlock(&lli->lli_lock); - - rc = cl_get_grouplock(cl_i2info(inode)->lli_clob, - arg, (file->f_flags & O_NONBLOCK), &grouplock); - if (rc) - RETURN(rc); - - cfs_spin_lock(&lli->lli_lock); - if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { - cfs_spin_unlock(&lli->lli_lock); - CERROR("another thread just won the race\n"); - cl_put_grouplock(&grouplock); - RETURN(-EINVAL); - } - - fd->fd_flags |= LL_FILE_GROUP_LOCKED; - fd->fd_grouplock = grouplock; - cfs_spin_unlock(&lli->lli_lock); - - CDEBUG(D_INFO, "group lock %lu obtained\n", arg); - RETURN(0); + spin_lock(&lli->lli_lock); + if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { + CWARN("group lock already existed with gid %lu\n", + fd->fd_grouplock.cg_gid); + spin_unlock(&lli->lli_lock); + RETURN(-EINVAL); + } + LASSERT(fd->fd_grouplock.cg_lock == NULL); + spin_unlock(&lli->lli_lock); + + rc = cl_get_grouplock(cl_i2info(inode)->lli_clob, + arg, (file->f_flags & O_NONBLOCK), &grouplock); + if (rc) + RETURN(rc); + + spin_lock(&lli->lli_lock); + if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { + spin_unlock(&lli->lli_lock); + CERROR("another thread just won the race\n"); + cl_put_grouplock(&grouplock); + RETURN(-EINVAL); + } + + fd->fd_flags |= LL_FILE_GROUP_LOCKED; + fd->fd_grouplock = grouplock; + spin_unlock(&lli->lli_lock); + + CDEBUG(D_INFO, "group lock %lu obtained\n", arg); + RETURN(0); } int ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg) { - struct ll_inode_info *lli = ll_i2info(inode); - struct ll_file_data *fd = LUSTRE_FPRIVATE(file); - struct ccc_grouplock grouplock; - ENTRY; + struct ll_inode_info *lli = ll_i2info(inode); + struct ll_file_data *fd = LUSTRE_FPRIVATE(file); + struct ccc_grouplock grouplock; + ENTRY; - cfs_spin_lock(&lli->lli_lock); - if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) { - cfs_spin_unlock(&lli->lli_lock); + spin_lock(&lli->lli_lock); + if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) { + spin_unlock(&lli->lli_lock); CWARN("no group lock held\n"); RETURN(-EINVAL); } @@ -1533,18 +1593,18 @@ int ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg) if (fd->fd_grouplock.cg_gid != arg) { CWARN("group lock %lu doesn't match current id %lu\n", arg, fd->fd_grouplock.cg_gid); - cfs_spin_unlock(&lli->lli_lock); - RETURN(-EINVAL); - } + spin_unlock(&lli->lli_lock); + RETURN(-EINVAL); + } - grouplock = fd->fd_grouplock; - memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock)); - fd->fd_flags &= ~LL_FILE_GROUP_LOCKED; - cfs_spin_unlock(&lli->lli_lock); + grouplock = fd->fd_grouplock; + memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock)); + fd->fd_flags &= ~LL_FILE_GROUP_LOCKED; + spin_unlock(&lli->lli_lock); - cl_put_grouplock(&grouplock); - CDEBUG(D_INFO, "group lock %lu released\n", arg); - RETURN(0); + cl_put_grouplock(&grouplock); + CDEBUG(D_INFO, "group lock %lu released\n", arg); + RETURN(0); } /** @@ -1586,9 +1646,10 @@ int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it) inode, och); out: /* this one is in place of ll_file_open */ - if (it_disposition(it, DISP_ENQ_OPEN_REF)) + if (it_disposition(it, DISP_ENQ_OPEN_REF)) { ptlrpc_req_finished(it->d.lustre.it_data); - it_clear_disposition(it, DISP_ENQ_OPEN_REF); + it_clear_disposition(it, DISP_ENQ_OPEN_REF); + } RETURN(rc); } @@ -1599,8 +1660,8 @@ int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it) int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap, int num_bytes) { - struct obd_export *exp = ll_i2dtexp(inode); - struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd; + struct obd_export *exp = ll_i2dtexp(inode); + struct lov_stripe_md *lsm = NULL; struct ll_fiemap_info_key fm_key = { .name = KEY_FIEMAP, }; int vallen = num_bytes; int rc; @@ -1619,68 +1680,80 @@ int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap, return rc; } - /* If the stripe_count > 1 and the application does not understand - * DEVICE_ORDER flag, then it cannot interpret the extents correctly. - */ - if (lsm->lsm_stripe_count > 1 && - !(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) - return -EOPNOTSUPP; + lsm = ccc_inode_lsm_get(inode); + if (lsm == NULL) + return -ENOENT; + + /* If the stripe_count > 1 and the application does not understand + * DEVICE_ORDER flag, then it cannot interpret the extents correctly. + */ + if (lsm->lsm_stripe_count > 1 && + !(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) + GOTO(out, rc = -EOPNOTSUPP); - fm_key.oa.o_id = lsm->lsm_object_id; - fm_key.oa.o_seq = lsm->lsm_object_seq; + fm_key.oa.o_oi = lsm->lsm_oi; fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP; - obdo_from_inode(&fm_key.oa, inode, &ll_i2info(inode)->lli_fid, - OBD_MD_FLSIZE); + obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE); + obdo_set_parent_fid(&fm_key.oa, &ll_i2info(inode)->lli_fid); /* If filesize is 0, then there would be no objects for mapping */ if (fm_key.oa.o_size == 0) { fiemap->fm_mapped_extents = 0; - RETURN(0); + GOTO(out, rc = 0); } memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap)); - rc = obd_get_info(exp, sizeof(fm_key), &fm_key, &vallen, fiemap, lsm); + rc = obd_get_info(NULL, exp, sizeof(fm_key), &fm_key, &vallen, + fiemap, lsm); if (rc) CERROR("obd_get_info failed: rc = %d\n", rc); - RETURN(rc); -} - -int ll_fid2path(struct obd_export *exp, void *arg) -{ - struct getinfo_fid2path *gfout, *gfin; - int outsize, rc; - ENTRY; - - /* Need to get the buflen */ - OBD_ALLOC_PTR(gfin); - if (gfin == NULL) - RETURN(-ENOMEM); - if (cfs_copy_from_user(gfin, arg, sizeof(*gfin))) { - OBD_FREE_PTR(gfin); - RETURN(-EFAULT); - } - - outsize = sizeof(*gfout) + gfin->gf_pathlen; - OBD_ALLOC(gfout, outsize); - if (gfout == NULL) { - OBD_FREE_PTR(gfin); - RETURN(-ENOMEM); - } - memcpy(gfout, gfin, sizeof(*gfout)); - OBD_FREE_PTR(gfin); - - /* Call mdc_iocontrol */ - rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL); - if (rc) - GOTO(gf_free, rc); - if (cfs_copy_to_user(arg, gfout, outsize)) - rc = -EFAULT; +out: + ccc_inode_lsm_put(inode, lsm); + RETURN(rc); +} + +int ll_fid2path(struct inode *inode, void *arg) +{ + struct obd_export *exp = ll_i2mdexp(inode); + struct getinfo_fid2path *gfout, *gfin; + int outsize, rc; + ENTRY; + + if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) && + !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH)) + RETURN(-EPERM); + + /* Need to get the buflen */ + OBD_ALLOC_PTR(gfin); + if (gfin == NULL) + RETURN(-ENOMEM); + if (copy_from_user(gfin, arg, sizeof(*gfin))) { + OBD_FREE_PTR(gfin); + RETURN(-EFAULT); + } + + outsize = sizeof(*gfout) + gfin->gf_pathlen; + OBD_ALLOC(gfout, outsize); + if (gfout == NULL) { + OBD_FREE_PTR(gfin); + RETURN(-ENOMEM); + } + memcpy(gfout, gfin, sizeof(*gfout)); + OBD_FREE_PTR(gfin); + + /* Call mdc_iocontrol */ + rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL); + if (rc) + GOTO(gf_free, rc); + + if (copy_to_user(arg, gfout, outsize)) + rc = -EFAULT; gf_free: - OBD_FREE(gfout, outsize); - RETURN(rc); + OBD_FREE(gfout, outsize); + RETURN(rc); } static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg) @@ -1702,10 +1775,10 @@ static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg) if (fiemap_s == NULL) RETURN(-ENOMEM); - /* get the fiemap value */ - if (copy_from_user(fiemap_s,(struct ll_user_fiemap __user *)arg, - sizeof(*fiemap_s))) - GOTO(error, rc = -EFAULT); + /* get the fiemap value */ + if (copy_from_user(fiemap_s, (struct ll_user_fiemap __user *)arg, + sizeof(*fiemap_s))) + GOTO(error, rc = -EFAULT); /* If fm_extent_count is non-zero, read the first extent since * it is used to calculate end_offset and device from previous @@ -1727,26 +1800,217 @@ static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg) ret_bytes += (fiemap_s->fm_mapped_extents * sizeof(struct ll_fiemap_extent)); - if (copy_to_user((void *)arg, fiemap_s, ret_bytes)) - rc = -EFAULT; + if (copy_to_user((void *)arg, fiemap_s, ret_bytes)) + rc = -EFAULT; error: OBD_FREE_LARGE(fiemap_s, num_bytes); RETURN(rc); } -#ifdef HAVE_UNLOCKED_IOCTL +/* + * Read the data_version for inode. + * + * This value is computed using stripe object version on OST. + * Version is computed using server side locking. + * + * @param extent_lock Take extent lock. Not needed if a process is already + * holding the OST object group locks. + */ +int ll_data_version(struct inode *inode, __u64 *data_version, + int extent_lock) +{ + struct lov_stripe_md *lsm = NULL; + struct ll_sb_info *sbi = ll_i2sbi(inode); + struct obdo *obdo = NULL; + int rc; + ENTRY; + + /* If no stripe, we consider version is 0. */ + lsm = ccc_inode_lsm_get(inode); + if (lsm == NULL) { + *data_version = 0; + CDEBUG(D_INODE, "No object for inode\n"); + RETURN(0); + } + + OBD_ALLOC_PTR(obdo); + if (obdo == NULL) { + ccc_inode_lsm_put(inode, lsm); + RETURN(-ENOMEM); + } + + rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, obdo, 0, extent_lock); + if (!rc) { + if (!(obdo->o_valid & OBD_MD_FLDATAVERSION)) + rc = -EOPNOTSUPP; + else + *data_version = obdo->o_data_version; + } + + OBD_FREE_PTR(obdo); + ccc_inode_lsm_put(inode, lsm); + + RETURN(rc); +} + +struct ll_swap_stack { + struct iattr ia1, ia2; + __u64 dv1, dv2; + struct inode *inode1, *inode2; + bool check_dv1, check_dv2; +}; + +static int ll_swap_layouts(struct file *file1, struct file *file2, + struct lustre_swap_layouts *lsl) +{ + struct mdc_swap_layouts msl; + struct md_op_data *op_data; + __u32 gid; + __u64 dv; + struct ll_swap_stack *llss = NULL; + int rc, rc1; + + OBD_ALLOC_PTR(llss); + if (llss == NULL) + RETURN(-ENOMEM); + + llss->inode1 = file1->f_dentry->d_inode; + llss->inode2 = file2->f_dentry->d_inode; + + if (!S_ISREG(llss->inode2->i_mode)) + GOTO(free, rc = -EINVAL); + + if (ll_permission(llss->inode1, MAY_WRITE, NULL) || + ll_permission(llss->inode2, MAY_WRITE, NULL)) + GOTO(free, rc = -EPERM); + + if (llss->inode2->i_sb != llss->inode1->i_sb) + GOTO(free, rc = -EXDEV); + + /* we use 2 bool because it is easier to swap than 2 bits */ + if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1) + llss->check_dv1 = true; + + if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2) + llss->check_dv2 = true; + + /* we cannot use lsl->sl_dvX directly because we may swap them */ + llss->dv1 = lsl->sl_dv1; + llss->dv2 = lsl->sl_dv2; + + rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2)); + if (rc == 0) /* same file, done! */ + GOTO(free, rc = 0); + + if (rc < 0) { /* sequentialize it */ + swap(llss->inode1, llss->inode2); + swap(file1, file2); + swap(llss->dv1, llss->dv2); + swap(llss->check_dv1, llss->check_dv2); + } + + gid = lsl->sl_gid; + if (gid != 0) { /* application asks to flush dirty cache */ + rc = ll_get_grouplock(llss->inode1, file1, gid); + if (rc < 0) + GOTO(free, rc); + + rc = ll_get_grouplock(llss->inode2, file2, gid); + if (rc < 0) { + ll_put_grouplock(llss->inode1, file1, gid); + GOTO(free, rc); + } + } + + /* to be able to restore mtime and atime after swap + * we need to first save them */ + if (lsl->sl_flags & + (SWAP_LAYOUTS_KEEP_MTIME | SWAP_LAYOUTS_KEEP_ATIME)) { + llss->ia1.ia_mtime = llss->inode1->i_mtime; + llss->ia1.ia_atime = llss->inode1->i_atime; + llss->ia1.ia_valid = ATTR_MTIME | ATTR_ATIME; + llss->ia2.ia_mtime = llss->inode2->i_mtime; + llss->ia2.ia_atime = llss->inode2->i_atime; + llss->ia2.ia_valid = ATTR_MTIME | ATTR_ATIME; + } + + /* ultimate check, before swaping the layouts we check if + * dataversion has changed (if requested) */ + if (llss->check_dv1) { + rc = ll_data_version(llss->inode1, &dv, 0); + if (rc) + GOTO(putgl, rc); + if (dv != llss->dv1) + GOTO(putgl, rc = -EAGAIN); + } + + if (llss->check_dv2) { + rc = ll_data_version(llss->inode2, &dv, 0); + if (rc) + GOTO(putgl, rc); + if (dv != llss->dv2) + GOTO(putgl, rc = -EAGAIN); + } + + /* struct md_op_data is used to send the swap args to the mdt + * only flags is missing, so we use struct mdc_swap_layouts + * through the md_op_data->op_data */ + /* flags from user space have to be converted before they are send to + * server, no flag is sent today, they are only used on the client */ + msl.msl_flags = 0; + rc = -ENOMEM; + op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0, + 0, LUSTRE_OPC_ANY, &msl); + if (op_data != NULL) { + rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, + ll_i2mdexp(llss->inode1), + sizeof(*op_data), op_data, NULL); + ll_finish_md_op_data(op_data); + } + +putgl: + if (gid != 0) { + ll_put_grouplock(llss->inode2, file2, gid); + ll_put_grouplock(llss->inode1, file1, gid); + } + + /* rc can be set from obd_iocontrol() or from a GOTO(putgl, ...) */ + if (rc != 0) + GOTO(free, rc); + + /* clear useless flags */ + if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_MTIME)) { + llss->ia1.ia_valid &= ~ATTR_MTIME; + llss->ia2.ia_valid &= ~ATTR_MTIME; + } + + if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_ATIME)) { + llss->ia1.ia_valid &= ~ATTR_ATIME; + llss->ia2.ia_valid &= ~ATTR_ATIME; + } + + /* update time if requested */ + rc = rc1 = 0; + if (llss->ia2.ia_valid != 0) + rc = ll_setattr(file1->f_dentry, &llss->ia2); + + if (llss->ia1.ia_valid != 0) + rc1 = ll_setattr(file2->f_dentry, &llss->ia1); + +free: + if (llss != NULL) + OBD_FREE_PTR(llss); + + RETURN(rc); +} + long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg) { - struct inode *inode = file->f_dentry->d_inode; -#else -int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd, - unsigned long arg) -{ -#endif - struct ll_file_data *fd = LUSTRE_FPRIVATE(file); - int flags; - ENTRY; + struct inode *inode = file->f_dentry->d_inode; + struct ll_file_data *fd = LUSTRE_FPRIVATE(file); + int flags, rc; + ENTRY; CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),cmd=%x\n", inode->i_ino, inode->i_generation, inode, cmd); @@ -1786,6 +2050,27 @@ int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd, RETURN(ll_lov_setstripe(inode, file, arg)); case LL_IOC_LOV_SETEA: RETURN(ll_lov_setea(inode, file, arg)); + case LL_IOC_LOV_SWAP_LAYOUTS: { + struct file *file2; + struct lustre_swap_layouts lsl; + + if (cfs_copy_from_user(&lsl, (char *)arg, + sizeof(struct lustre_swap_layouts))) + RETURN(-EFAULT); + + if ((file->f_flags & O_ACCMODE) == 0) /* O_RDONLY */ + RETURN(-EPERM); + + file2 = fget(lsl.sl_fd); + if (file2 == NULL) + RETURN(-EBADF); + + rc = -EPERM; + if ((file2->f_flags & O_ACCMODE) != 0) /* O_WRONLY or O_RDWR */ + rc = ll_swap_layouts(file, file2, &lsl); + fput(file2); + RETURN(rc); + } case LL_IOC_LOV_GETSTRIPE: RETURN(ll_lov_getstripe(inode, arg)); case LL_IOC_RECREATE_OBJ: @@ -1813,17 +2098,32 @@ int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd, case FSFILT_IOC_SETVERSION_OLD: case FSFILT_IOC_SETVERSION: */ - case LL_IOC_FLUSHCTX: - RETURN(ll_flush_ctx(inode)); - case LL_IOC_PATH2FID: { - if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode), - sizeof(struct lu_fid))) - RETURN(-EFAULT); + case LL_IOC_FLUSHCTX: + RETURN(ll_flush_ctx(inode)); + case LL_IOC_PATH2FID: { + if (copy_to_user((void *)arg, ll_inode2fid(inode), + sizeof(struct lu_fid))) + RETURN(-EFAULT); - RETURN(0); - } - case OBD_IOC_FID2PATH: - RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg)); + RETURN(0); + } + case OBD_IOC_FID2PATH: + RETURN(ll_fid2path(inode, (void *)arg)); + case LL_IOC_DATA_VERSION: { + struct ioc_data_version idv; + int rc; + + if (copy_from_user(&idv, (char *)arg, sizeof(idv))) + RETURN(-EFAULT); + + rc = ll_data_version(inode, &idv.idv_version, + !(idv.idv_flags & LL_DV_NOFLUSH)); + + if (rc == 0 && copy_to_user((char *) arg, &idv, sizeof(idv))) + RETURN(-EFAULT); + + RETURN(rc); + } case LL_IOC_GET_MDTIDX: { int mdtidx; @@ -1837,113 +2137,328 @@ int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd, RETURN(0); } - - default: { - int err; - - if (LLIOC_STOP == - ll_iocontrol_call(inode, file, cmd, arg, &err)) - RETURN(err); - - RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL, - (void *)arg)); - } - } + case OBD_IOC_GETDTNAME: + case OBD_IOC_GETMDNAME: + RETURN(ll_get_obd_name(inode, cmd, arg)); + case LL_IOC_HSM_STATE_GET: { + struct md_op_data *op_data; + struct hsm_user_state *hus; + int rc; + + OBD_ALLOC_PTR(hus); + if (hus == NULL) + RETURN(-ENOMEM); + + op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, + LUSTRE_OPC_ANY, hus); + if (op_data == NULL) { + OBD_FREE_PTR(hus); + RETURN(-ENOMEM); + } + + rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data), + op_data, NULL); + + if (copy_to_user((void *)arg, hus, sizeof(*hus))) + rc = -EFAULT; + + ll_finish_md_op_data(op_data); + OBD_FREE_PTR(hus); + RETURN(rc); + } + case LL_IOC_HSM_STATE_SET: { + struct md_op_data *op_data; + struct hsm_state_set *hss; + int rc; + + OBD_ALLOC_PTR(hss); + if (hss == NULL) + RETURN(-ENOMEM); + if (copy_from_user(hss, (char *)arg, sizeof(*hss))) { + OBD_FREE_PTR(hss); + RETURN(-EFAULT); + } + + /* Non-root users are forbidden to set or clear flags which are + * NOT defined in HSM_USER_MASK. */ + if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) + && !cfs_capable(CFS_CAP_SYS_ADMIN)) { + OBD_FREE_PTR(hss); + RETURN(-EPERM); + } + + op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, + LUSTRE_OPC_ANY, hss); + if (op_data == NULL) { + OBD_FREE_PTR(hss); + RETURN(-ENOMEM); + } + + rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data), + op_data, NULL); + + ll_finish_md_op_data(op_data); + + OBD_FREE_PTR(hss); + RETURN(rc); + } + case LL_IOC_HSM_ACTION: { + struct md_op_data *op_data; + struct hsm_current_action *hca; + int rc; + + OBD_ALLOC_PTR(hca); + if (hca == NULL) + RETURN(-ENOMEM); + + op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, + LUSTRE_OPC_ANY, hca); + if (op_data == NULL) { + OBD_FREE_PTR(hca); + RETURN(-ENOMEM); + } + + rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data), + op_data, NULL); + + if (cfs_copy_to_user((char *)arg, hca, sizeof(*hca))) + rc = -EFAULT; + + ll_finish_md_op_data(op_data); + OBD_FREE_PTR(hca); + RETURN(rc); + } + default: { + int err; + + if (LLIOC_STOP == + ll_iocontrol_call(inode, file, cmd, arg, &err)) + RETURN(err); + + RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL, + (void *)arg)); + } + } +} + +#ifndef HAVE_FILE_LLSEEK_SIZE +static inline loff_t +llseek_execute(struct file *file, loff_t offset, loff_t maxsize) +{ + if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) + return -EINVAL; + if (offset > maxsize) + return -EINVAL; + + if (offset != file->f_pos) { + file->f_pos = offset; + file->f_version = 0; + } + return offset; +} + +static loff_t +generic_file_llseek_size(struct file *file, loff_t offset, int origin, + loff_t maxsize, loff_t eof) +{ + struct inode *inode = file->f_dentry->d_inode; + + switch (origin) { + case SEEK_END: + offset += eof; + break; + case SEEK_CUR: + /* + * Here we special-case the lseek(fd, 0, SEEK_CUR) + * position-querying operation. Avoid rewriting the "same" + * f_pos value back to the file because a concurrent read(), + * write() or lseek() might have altered it + */ + if (offset == 0) + return file->f_pos; + /* + * f_lock protects against read/modify/write race with other + * SEEK_CURs. Note that parallel writes and reads behave + * like SEEK_SET. + */ + mutex_lock(&inode->i_mutex); + offset = llseek_execute(file, file->f_pos + offset, maxsize); + mutex_unlock(&inode->i_mutex); + return offset; + case SEEK_DATA: + /* + * In the generic case the entire file is data, so as long as + * offset isn't at the end of the file then the offset is data. + */ + if (offset >= eof) + return -ENXIO; + break; + case SEEK_HOLE: + /* + * There is a virtual hole at the end of the file, so as long as + * offset isn't i_size or larger, return i_size. + */ + if (offset >= eof) + return -ENXIO; + offset = eof; + break; + } + + return llseek_execute(file, offset, maxsize); } +#endif loff_t ll_file_seek(struct file *file, loff_t offset, int origin) { - struct inode *inode = file->f_dentry->d_inode; - loff_t retval; - ENTRY; - retval = offset + ((origin == 2) ? i_size_read(inode) : - (origin == 1) ? file->f_pos : 0); - CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), to=%Lu=%#Lx(%s)\n", - inode->i_ino, inode->i_generation, inode, retval, retval, - origin == 2 ? "SEEK_END": origin == 1 ? "SEEK_CUR" : "SEEK_SET"); - ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1); - - if (origin == 2) { /* SEEK_END */ - int nonblock = 0, rc; - - if (file->f_flags & O_NONBLOCK) - nonblock = LDLM_FL_BLOCK_NOWAIT; - - rc = cl_glimpse_size(inode); - if (rc != 0) - RETURN(rc); + struct inode *inode = file->f_dentry->d_inode; + loff_t retval, eof = 0; - offset += i_size_read(inode); - } else if (origin == 1) { /* SEEK_CUR */ - offset += file->f_pos; - } + ENTRY; + retval = offset + ((origin == SEEK_END) ? i_size_read(inode) : + (origin == SEEK_CUR) ? file->f_pos : 0); + CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), to=%llu=%#llx(%d)\n", + inode->i_ino, inode->i_generation, inode, retval, retval, + origin); + ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1); - retval = -EINVAL; - if (offset >= 0 && offset <= ll_file_maxbytes(inode)) { - if (offset != file->f_pos) { - file->f_pos = offset; - } - retval = offset; - } + if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) { + retval = ll_glimpse_size(inode); + if (retval != 0) + RETURN(retval); + eof = i_size_read(inode); + } - RETURN(retval); + retval = ll_generic_file_llseek_size(file, offset, origin, + ll_file_maxbytes(inode), eof); + RETURN(retval); } -#ifdef HAVE_FLUSH_OWNER_ID int ll_flush(struct file *file, fl_owner_t id) -#else -int ll_flush(struct file *file) -#endif { - struct inode *inode = file->f_dentry->d_inode; - struct ll_inode_info *lli = ll_i2info(inode); - struct lov_stripe_md *lsm = lli->lli_smd; - int rc, err; + struct inode *inode = file->f_dentry->d_inode; + struct ll_inode_info *lli = ll_i2info(inode); + struct ll_file_data *fd = LUSTRE_FPRIVATE(file); + int rc, err; - /* the application should know write failure already. */ - if (lli->lli_write_rc) - return 0; + LASSERT(!S_ISDIR(inode->i_mode)); - /* catch async errors that were recorded back when async writeback - * failed for pages in this mapping. */ - rc = lli->lli_async_rc; - lli->lli_async_rc = 0; - if (lsm) { - err = lov_test_and_clear_async_rc(lsm); - if (rc == 0) - rc = err; - } + /* catch async errors that were recorded back when async writeback + * failed for pages in this mapping. */ + rc = lli->lli_async_rc; + lli->lli_async_rc = 0; + err = lov_read_and_clear_async_rc(lli->lli_clob); + if (rc == 0) + rc = err; - return rc ? -EIO : 0; + /* The application has been told write failure already. + * Do not report failure again. */ + if (fd->fd_write_failed) + return 0; + return rc ? -EIO : 0; } +/** + * Called to make sure a portion of file has been written out. + * if @local_only is not true, it will send OST_SYNC RPCs to ost. + * + * Return how many pages have been written. + */ +int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end, + enum cl_fsync_mode mode) +{ + struct cl_env_nest nest; + struct lu_env *env; + struct cl_io *io; + struct obd_capa *capa = NULL; + struct cl_fsync_io *fio; + int result; + ENTRY; + + if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL && + mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL) + RETURN(-EINVAL); + + env = cl_env_nested_get(&nest); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); + + capa = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE); + + io = ccc_env_thread_io(env); + io->ci_obj = cl_i2info(inode)->lli_clob; + io->ci_ignore_layout = 1; + + /* initialize parameters for sync */ + fio = &io->u.ci_fsync; + fio->fi_capa = capa; + fio->fi_start = start; + fio->fi_end = end; + fio->fi_fid = ll_inode2fid(inode); + fio->fi_mode = mode; + fio->fi_nr_written = 0; + + if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0) + result = cl_io_loop(env, io); + else + result = io->ci_result; + if (result == 0) + result = fio->fi_nr_written; + cl_io_fini(env, io); + cl_env_nested_put(&nest, env); + + capa_put(capa); + + RETURN(result); +} + +/* + * When dentry is provided (the 'else' case), *file->f_dentry may be + * null and dentry must be used directly rather than pulled from + * *file->f_dentry as is done otherwise. + */ + +#ifdef HAVE_FILE_FSYNC_4ARGS +int ll_fsync(struct file *file, loff_t start, loff_t end, int data) +{ + struct dentry *dentry = file->f_dentry; +#elif defined(HAVE_FILE_FSYNC_2ARGS) +int ll_fsync(struct file *file, int data) +{ + struct dentry *dentry = file->f_dentry; +#else int ll_fsync(struct file *file, struct dentry *dentry, int data) { +#endif struct inode *inode = dentry->d_inode; struct ll_inode_info *lli = ll_i2info(inode); - struct lov_stripe_md *lsm = lli->lli_smd; struct ptlrpc_request *req; struct obd_capa *oc; int rc, err; ENTRY; + CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino, inode->i_generation, inode); ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1); +#ifdef HAVE_FILE_FSYNC_4ARGS + rc = filemap_write_and_wait_range(inode->i_mapping, start, end); + mutex_lock(&inode->i_mutex); +#else /* fsync's caller has already called _fdata{sync,write}, we want * that IO to finish before calling the osc and mdc sync methods */ rc = filemap_fdatawait(inode->i_mapping); +#endif /* catch async errors that were recorded back when async writeback * failed for pages in this mapping. */ - err = lli->lli_async_rc; - lli->lli_async_rc = 0; - if (rc == 0) - rc = err; - if (lsm) { - err = lov_test_and_clear_async_rc(lsm); + if (!S_ISDIR(inode->i_mode)) { + err = lli->lli_async_rc; + lli->lli_async_rc = 0; if (rc == 0) rc = err; + err = lov_read_and_clear_async_rc(lli->lli_clob); + if (rc == 0) + rc = err; } oc = ll_mdscapa_get(inode); @@ -1955,37 +2470,23 @@ int ll_fsync(struct file *file, struct dentry *dentry, int data) if (!err) ptlrpc_req_finished(req); - if (data && lsm) { - struct obd_info *oinfo; - - OBD_ALLOC_PTR(oinfo); - if (!oinfo) - RETURN(rc ? rc : -ENOMEM); - OBDO_ALLOC(oinfo->oi_oa); - if (!oinfo->oi_oa) { - OBD_FREE_PTR(oinfo); - RETURN(rc ? rc : -ENOMEM); - } - oinfo->oi_oa->o_id = lsm->lsm_object_id; - oinfo->oi_oa->o_seq = lsm->lsm_object_seq; - oinfo->oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP; - obdo_from_inode(oinfo->oi_oa, inode, &ll_i2info(inode)->lli_fid, - OBD_MD_FLTYPE | OBD_MD_FLATIME | - OBD_MD_FLMTIME | OBD_MD_FLCTIME | - OBD_MD_FLGROUP); - oinfo->oi_md = lsm; - oinfo->oi_capa = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE); - err = obd_sync_rqset(ll_i2sbi(inode)->ll_dt_exp, oinfo, 0, - OBD_OBJECT_EOF); - capa_put(oinfo->oi_capa); - if (!rc) - rc = err; - OBDO_FREE(oinfo->oi_oa); - OBD_FREE_PTR(oinfo); - lli->lli_write_rc = err < 0 ? : 0; - } - - RETURN(rc); + if (data) { + struct ll_file_data *fd = LUSTRE_FPRIVATE(file); + + err = cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, + CL_FSYNC_ALL); + if (rc == 0 && err < 0) + rc = err; + if (rc < 0) + fd->fd_write_failed = true; + else + fd->fd_write_failed = false; + } + +#ifdef HAVE_FILE_FSYNC_4ARGS + mutex_unlock(&inode->i_mutex); +#endif + RETURN(rc); } int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) @@ -2000,6 +2501,7 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) ldlm_policy_data_t flock = {{0}}; int flags = 0; int rc; + int rc2 = 0; ENTRY; CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu file_lock=%p\n", @@ -2022,6 +2524,16 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) } flock.l_flock.pid = file_lock->fl_pid; + /* Somewhat ugly workaround for svc lockd. + * lockd installs custom fl_lmops->lm_compare_owner that checks + * for the fl_owner to be the same (which it always is on local node + * I guess between lockd processes) and then compares pid. + * As such we assign pid to the owner field to make it all work, + * conflict with normal locks is unlikely since pid space and + * pointer space for current->files are not intersecting */ + if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner) + flock.l_flock.owner = (unsigned long)file_lock->fl_pid; + switch (file_lock->fl_type) { case F_RDLCK: einfo.ei_mode = LCK_PR; @@ -2041,8 +2553,9 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) einfo.ei_mode = LCK_PW; break; default: - CERROR("unknown fcntl lock type: %d\n", file_lock->fl_type); - RETURN (-EINVAL); + CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", + file_lock->fl_type); + RETURN (-ENOTSUPP); } switch (cmd) { @@ -2084,17 +2597,22 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, op_data, &lockh, &flock, 0, NULL /* req */, flags); - ll_finish_md_op_data(op_data); - if ((file_lock->fl_flags & FL_FLOCK) && (rc == 0 || file_lock->fl_type == F_UNLCK)) - ll_flock_lock_file_wait(file, file_lock, (cmd == F_SETLKW)); -#ifdef HAVE_F_OP_FLOCK + rc2 = flock_lock_file_wait(file, file_lock); if ((file_lock->fl_flags & FL_POSIX) && (rc == 0 || file_lock->fl_type == F_UNLCK) && !(flags & LDLM_FL_TEST_LOCK)) - posix_lock_file_wait(file, file_lock); -#endif + rc2 = posix_lock_file_wait(file, file_lock); + + if (rc2 && file_lock->fl_type != F_UNLCK) { + einfo.ei_mode = LCK_NL; + md_enqueue(sbi->ll_md_exp, &einfo, NULL, + op_data, &lockh, &flock, 0, NULL /* req */, flags); + rc = rc2; + } + + ll_finish_md_op_data(op_data); RETURN(rc); } @@ -2106,14 +2624,25 @@ int ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock) RETURN(-ENOSYS); } -int ll_have_md_lock(struct inode *inode, __u64 bits, ldlm_mode_t l_req_mode) +/** + * test if some locks matching bits and l_req_mode are acquired + * - bits can be in different locks + * - if found clear the common lock bits in *bits + * - the bits not found, are kept in *bits + * \param inode [IN] + * \param bits [IN] searched lock bits [IN] + * \param l_req_mode [IN] searched lock mode + * \retval boolean, true iff all bits are found + */ +int ll_have_md_lock(struct inode *inode, __u64 *bits, ldlm_mode_t l_req_mode) { struct lustre_handle lockh; - ldlm_policy_data_t policy = { .l_inodebits = {bits}}; + ldlm_policy_data_t policy; ldlm_mode_t mode = (l_req_mode == LCK_MINMODE) ? (LCK_CR|LCK_CW|LCK_PR|LCK_PW) : l_req_mode; struct lu_fid *fid; - int flags; + __u64 flags; + int i; ENTRY; if (!inode) @@ -2124,50 +2653,62 @@ int ll_have_md_lock(struct inode *inode, __u64 bits, ldlm_mode_t l_req_mode) ldlm_lockname[mode]); flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK; - if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy, - mode, &lockh)) { - RETURN(1); + for (i = 0; i < MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) { + policy.l_inodebits.bits = *bits & (1 << i); + if (policy.l_inodebits.bits == 0) + continue; + + if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS, + &policy, mode, &lockh)) { + struct ldlm_lock *lock; + + lock = ldlm_handle2lock(&lockh); + if (lock) { + *bits &= + ~(lock->l_policy_data.l_inodebits.bits); + LDLM_LOCK_PUT(lock); + } else { + *bits &= ~policy.l_inodebits.bits; + } + } } - RETURN(0); + RETURN(*bits == 0); } ldlm_mode_t ll_take_md_lock(struct inode *inode, __u64 bits, - struct lustre_handle *lockh) + struct lustre_handle *lockh, __u64 flags) { ldlm_policy_data_t policy = { .l_inodebits = {bits}}; struct lu_fid *fid; ldlm_mode_t rc; - int flags; ENTRY; fid = &ll_i2info(inode)->lli_fid; CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid)); - flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING; - rc = md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy, + rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags, + fid, LDLM_IBITS, &policy, LCK_CR|LCK_CW|LCK_PR|LCK_PW, lockh); RETURN(rc); } -static int ll_inode_revalidate_fini(struct inode *inode, int rc) { - if (rc == -ENOENT) { /* Already unlinked. Just update nlink - * and return success */ - inode->i_nlink = 0; - /* This path cannot be hit for regular files unless in - * case of obscure races, so no need to to validate - * size. */ - if (!S_ISREG(inode->i_mode) && - !S_ISDIR(inode->i_mode)) - return 0; - } - - if (rc) { - CERROR("failure %d inode %lu\n", rc, inode->i_ino); - return -abs(rc); - - } +static int ll_inode_revalidate_fini(struct inode *inode, int rc) +{ + /* Already unlinked. Just update nlink and return success */ + if (rc == -ENOENT) { + clear_nlink(inode); + /* This path cannot be hit for regular files unless in + * case of obscure races, so no need to to validate + * size. */ + if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode)) + return 0; + } else if (rc != 0) { + CERROR("%s: revalidate FID "DFID" error: rc = %d\n", + ll_get_fsname(inode->i_sb, NULL, 0), + PFID(ll_inode2fid(inode)), rc); + } - return 0; + return rc; } int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, @@ -2175,16 +2716,11 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, { struct inode *inode = dentry->d_inode; struct ptlrpc_request *req = NULL; - struct ll_sb_info *sbi; struct obd_export *exp; int rc = 0; ENTRY; - if (!inode) { - CERROR("REPORT THIS LINE TO PETER\n"); - RETURN(0); - } - sbi = ll_i2sbi(inode); + LASSERT(inode != NULL); CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),name=%s\n", inode->i_ino, inode->i_generation, inode, dentry->d_name.name); @@ -2194,7 +2730,7 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC. * But under CMD case, it caused some lock issues, should be fixed * with new CMD ibits lock. See bug 12718 */ - if (exp->exp_connect_flags & OBD_CONNECT_ATTRFID) { + if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) { struct lookup_intent oit = { .it_op = IT_GETATTR }; struct md_op_data *op_data; @@ -2231,16 +2767,11 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, do_lookup() -> ll_revalidate_it(). We cannot use d_drop here to preserve get_cwd functionality on 2.6. Bug 10503 */ - if (!dentry->d_inode->i_nlink) { - cfs_spin_lock(&ll_lookup_lock); - spin_lock(&dcache_lock); - ll_drop_dentry(dentry); - spin_unlock(&dcache_lock); - cfs_spin_unlock(&ll_lookup_lock); - } + if (!dentry->d_inode->i_nlink) + d_lustre_invalidate(dentry); ll_lookup_finish_locks(&oit, dentry); - } else if (!ll_have_md_lock(dentry->d_inode, ibits, LCK_MINMODE)) { + } else if (!ll_have_md_lock(dentry->d_inode, &ibits, LCK_MINMODE)) { struct ll_sb_info *sbi = ll_i2sbi(dentry->d_inode); obd_valid valid = OBD_MD_FLGETATTR; struct md_op_data *op_data; @@ -2256,8 +2787,8 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, ealen, LUSTRE_OPC_ANY, NULL); - if (op_data == NULL) - RETURN(-ENOMEM); + if (IS_ERR(op_data)) + RETURN(PTR_ERR(op_data)); op_data->op_valid = valid; /* Once OBD_CONNECT_ATTRFID is not supported, we can't find one @@ -2270,7 +2801,7 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, RETURN(rc); } - rc = ll_prep_inode(&inode, req, NULL); + rc = ll_prep_inode(&inode, req, NULL, NULL); } out: ptlrpc_req_finished(req); @@ -2285,21 +2816,17 @@ int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, ENTRY; rc = __ll_inode_revalidate_it(dentry, it, ibits); - - /* if object not yet allocated, don't validate size */ - if (rc == 0 && ll_i2info(dentry->d_inode)->lli_smd == NULL) { - LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_lvb.lvb_atime; - LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_lvb.lvb_mtime; - LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_lvb.lvb_ctime; - RETURN(0); - } - - /* cl_glimpse_size will prefer locally cached writes if they extend - * the file */ - - if (rc == 0) - rc = cl_glimpse_size(inode); - + if (rc != 0) + RETURN(rc); + + /* if object isn't regular file, don't validate size */ + if (!S_ISREG(inode->i_mode)) { + LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_lvb.lvb_atime; + LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_lvb.lvb_mtime; + LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_lvb.lvb_ctime; + } else { + rc = ll_glimpse_size(inode); + } RETURN(rc); } @@ -2327,15 +2854,11 @@ int ll_getattr_it(struct vfsmount *mnt, struct dentry *de, stat->nlink = inode->i_nlink; stat->uid = inode->i_uid; stat->gid = inode->i_gid; - stat->rdev = kdev_t_to_nr(inode->i_rdev); + stat->rdev = inode->i_rdev; stat->atime = inode->i_atime; stat->mtime = inode->i_mtime; stat->ctime = inode->i_ctime; -#ifdef HAVE_INODE_BLKSIZE - stat->blksize = inode->i_blksize; -#else - stat->blksize = 1 << inode->i_blkbits; -#endif + stat->blksize = 1 << inode->i_blkbits; stat->size = i_size_read(inode); stat->blocks = inode->i_blocks; @@ -2384,42 +2907,73 @@ int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, } #endif +struct posix_acl * ll_get_acl(struct inode *inode, int type) +{ + struct ll_inode_info *lli = ll_i2info(inode); + struct posix_acl *acl = NULL; + ENTRY; + + spin_lock(&lli->lli_lock); + /* VFS' acl_permission_check->check_acl will release the refcount */ + acl = posix_acl_dup(lli->lli_posix_acl); + spin_unlock(&lli->lli_lock); + + RETURN(acl); +} -static -int lustre_check_acl(struct inode *inode, int mask) +#ifndef HAVE_GENERIC_PERMISSION_2ARGS +static int +# ifdef HAVE_GENERIC_PERMISSION_4ARGS +ll_check_acl(struct inode *inode, int mask, unsigned int flags) +# else +ll_check_acl(struct inode *inode, int mask) +# endif { -#ifdef CONFIG_FS_POSIX_ACL - struct ll_inode_info *lli = ll_i2info(inode); - struct posix_acl *acl; - int rc; - ENTRY; +# ifdef CONFIG_FS_POSIX_ACL + struct posix_acl *acl; + int rc; + ENTRY; - cfs_spin_lock(&lli->lli_lock); - acl = posix_acl_dup(lli->lli_posix_acl); - cfs_spin_unlock(&lli->lli_lock); +# ifdef HAVE_GENERIC_PERMISSION_4ARGS + if (flags & IPERM_FLAG_RCU) + return -ECHILD; +# endif + acl = ll_get_acl(inode, ACL_TYPE_ACCESS); - if (!acl) - RETURN(-EAGAIN); + if (!acl) + RETURN(-EAGAIN); - rc = posix_acl_permission(inode, acl, mask); - posix_acl_release(acl); + rc = posix_acl_permission(inode, acl, mask); + posix_acl_release(acl); - RETURN(rc); -#else - return -EAGAIN; -#endif + RETURN(rc); +# else /* !CONFIG_FS_POSIX_ACL */ + return -EAGAIN; +# endif /* CONFIG_FS_POSIX_ACL */ } +#endif /* HAVE_GENERIC_PERMISSION_2ARGS */ -#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,10)) -#ifndef HAVE_INODE_PERMISION_2ARGS -int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd) +#ifdef HAVE_GENERIC_PERMISSION_4ARGS +int ll_inode_permission(struct inode *inode, int mask, unsigned int flags) #else +# ifdef HAVE_INODE_PERMISION_2ARGS int ll_inode_permission(struct inode *inode, int mask) +# else +int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd) +# endif #endif { int rc = 0; ENTRY; +#ifdef MAY_NOT_BLOCK + if (mask & MAY_NOT_BLOCK) + return -ECHILD; +#elif defined(HAVE_GENERIC_PERMISSION_4ARGS) + if (flags & IPERM_FLAG_RCU) + return -ECHILD; +#endif + /* as root inode are NOT getting validated in lookup operation, * need to do it before permission check. */ @@ -2432,68 +2986,17 @@ int ll_inode_permission(struct inode *inode, int mask) RETURN(rc); } - CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), inode mode %x mask %o\n", - inode->i_ino, inode->i_generation, inode, inode->i_mode, mask); - - if (ll_i2sbi(inode)->ll_flags & LL_SBI_RMT_CLIENT) - return lustre_check_remote_perm(inode, mask); - - ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1); - rc = generic_permission(inode, mask, lustre_check_acl); + CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), inode mode %x mask %o\n", + inode->i_ino, inode->i_generation, inode, inode->i_mode, mask); - RETURN(rc); -} -#else -int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd) -{ - int mode = inode->i_mode; - int rc; - - CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), mask %o\n", - inode->i_ino, inode->i_generation, inode, mask); - - if (ll_i2sbi(inode)->ll_flags & LL_SBI_RMT_CLIENT) - return lustre_check_remote_perm(inode, mask); - - ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1); - - if ((mask & MAY_WRITE) && IS_RDONLY(inode) && - (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) - return -EROFS; - if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode)) - return -EACCES; - if (cfs_curproc_fsuid() == inode->i_uid) { - mode >>= 6; - } else if (1) { - if (((mode >> 3) & mask & S_IRWXO) != mask) - goto check_groups; - rc = lustre_check_acl(inode, mask); - if (rc == -EAGAIN) - goto check_groups; - if (rc == -EACCES) - goto check_capabilities; - return rc; - } else { -check_groups: - if (cfs_curproc_is_in_groups(inode->i_gid)) - mode >>= 3; - } - if ((mode & mask & S_IRWXO) == mask) - return 0; - -check_capabilities: - if (!(mask & MAY_EXEC) || - (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode)) - if (cfs_capable(CFS_CAP_DAC_OVERRIDE)) - return 0; + if (ll_i2sbi(inode)->ll_flags & LL_SBI_RMT_CLIENT) + return lustre_check_remote_perm(inode, mask); - if (cfs_capable(CFS_CAP_DAC_READ_SEARCH) && ((mask == MAY_READ) || - (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))) - return 0; + ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1); + rc = ll_generic_permission(inode, mask, flags, ll_check_acl); - return -EACCES; + RETURN(rc); } -#endif #ifdef HAVE_FILE_READV #define READ_METHOD readv @@ -2513,11 +3016,7 @@ struct file_operations ll_file_operations = { .READ_METHOD = READ_FUNCTION, .write = ll_file_write, .WRITE_METHOD = WRITE_FUNCTION, -#ifdef HAVE_UNLOCKED_IOCTL .unlocked_ioctl = ll_file_ioctl, -#else - .ioctl = ll_file_ioctl, -#endif .open = ll_file_open, .release = ll_file_release, .mmap = ll_file_mmap, @@ -2537,11 +3036,7 @@ struct file_operations ll_file_operations_flock = { .READ_METHOD = READ_FUNCTION, .write = ll_file_write, .WRITE_METHOD = WRITE_FUNCTION, -#ifdef HAVE_UNLOCKED_IOCTL .unlocked_ioctl = ll_file_ioctl, -#else - .ioctl = ll_file_ioctl, -#endif .open = ll_file_open, .release = ll_file_release, .mmap = ll_file_mmap, @@ -2554,9 +3049,7 @@ struct file_operations ll_file_operations_flock = { #endif .fsync = ll_fsync, .flush = ll_flush, -#ifdef HAVE_F_OP_FLOCK .flock = ll_file_flock, -#endif .lock = ll_file_flock }; @@ -2566,11 +3059,7 @@ struct file_operations ll_file_operations_noflock = { .READ_METHOD = READ_FUNCTION, .write = ll_file_write, .WRITE_METHOD = WRITE_FUNCTION, -#ifdef HAVE_UNLOCKED_IOCTL .unlocked_ioctl = ll_file_ioctl, -#else - .ioctl = ll_file_ioctl, -#endif .open = ll_file_open, .release = ll_file_release, .mmap = ll_file_mmap, @@ -2583,29 +3072,29 @@ struct file_operations ll_file_operations_noflock = { #endif .fsync = ll_fsync, .flush = ll_flush, -#ifdef HAVE_F_OP_FLOCK .flock = ll_file_noflock, -#endif .lock = ll_file_noflock }; struct inode_operations ll_file_inode_operations = { - .setattr = ll_setattr, - .truncate = ll_truncate, - .getattr = ll_getattr, - .permission = ll_inode_permission, - .setxattr = ll_setxattr, - .getxattr = ll_getxattr, - .listxattr = ll_listxattr, - .removexattr = ll_removexattr, + .setattr = ll_setattr, + .getattr = ll_getattr, + .permission = ll_inode_permission, + .setxattr = ll_setxattr, + .getxattr = ll_getxattr, + .listxattr = ll_listxattr, + .removexattr = ll_removexattr, #ifdef HAVE_LINUX_FIEMAP_H - .fiemap = ll_fiemap, + .fiemap = ll_fiemap, +#endif +#ifdef HAVE_IOP_GET_ACL + .get_acl = ll_get_acl, #endif }; /* dynamic ioctl number support routins */ static struct llioc_ctl_data { - cfs_rw_semaphore_t ioc_sem; + struct rw_semaphore ioc_sem; cfs_list_t ioc_head; } llioc = { __RWSEM_INITIALIZER(llioc.ioc_sem), @@ -2642,9 +3131,9 @@ void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd) in_data->iocd_count = count; memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count); - cfs_down_write(&llioc.ioc_sem); + down_write(&llioc.ioc_sem); cfs_list_add_tail(&in_data->iocd_list, &llioc.ioc_head); - cfs_up_write(&llioc.ioc_sem); + up_write(&llioc.ioc_sem); RETURN(in_data); } @@ -2656,19 +3145,19 @@ void ll_iocontrol_unregister(void *magic) if (magic == NULL) return; - cfs_down_write(&llioc.ioc_sem); + down_write(&llioc.ioc_sem); cfs_list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) { if (tmp == magic) { unsigned int size = tmp->iocd_size; cfs_list_del(&tmp->iocd_list); - cfs_up_write(&llioc.ioc_sem); + up_write(&llioc.ioc_sem); OBD_FREE(tmp, size); return; } } - cfs_up_write(&llioc.ioc_sem); + up_write(&llioc.ioc_sem); CWARN("didn't find iocontrol register block with magic: %p\n", magic); } @@ -2683,7 +3172,7 @@ enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file, struct llioc_data *data; int rc = -EINVAL, i; - cfs_down_read(&llioc.ioc_sem); + down_read(&llioc.ioc_sem); cfs_list_for_each_entry(data, &llioc.ioc_head, iocd_list) { for (i = 0; i < data->iocd_count; i++) { if (cmd != data->iocd_cmd[i]) @@ -2696,9 +3185,252 @@ enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file, if (ret == LLIOC_STOP) break; } - cfs_up_read(&llioc.ioc_sem); + up_read(&llioc.ioc_sem); if (rcp) *rcp = rc; return ret; } + +int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf) +{ + struct ll_inode_info *lli = ll_i2info(inode); + struct cl_env_nest nest; + struct lu_env *env; + int result; + ENTRY; + + if (lli->lli_clob == NULL) + RETURN(0); + + env = cl_env_nested_get(&nest); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); + + result = cl_conf_set(env, lli->lli_clob, conf); + cl_env_nested_put(&nest, env); + + if (conf->coc_opc == OBJECT_CONF_SET) { + struct ldlm_lock *lock = conf->coc_lock; + + LASSERT(lock != NULL); + LASSERT(ldlm_has_layout(lock)); + if (result == 0) { + /* it can only be allowed to match after layout is + * applied to inode otherwise false layout would be + * seen. Applying layout shoud happen before dropping + * the intent lock. */ + ldlm_lock_allow_match(lock); + } + } + RETURN(result); +} + +/** + * Apply the layout to the inode. Layout lock is held and will be released + * in this function. + */ +static int ll_layout_lock_set(struct lustre_handle *lockh, ldlm_mode_t mode, + struct inode *inode, __u32 *gen, bool reconf) +{ + struct ll_inode_info *lli = ll_i2info(inode); + struct ll_sb_info *sbi = ll_i2sbi(inode); + struct ldlm_lock *lock; + struct lustre_md md = { NULL }; + struct cl_object_conf conf; + int rc = 0; + bool lvb_ready; + ENTRY; + + LASSERT(lustre_handle_is_used(lockh)); + + lock = ldlm_handle2lock(lockh); + LASSERT(lock != NULL); + LASSERT(ldlm_has_layout(lock)); + + LDLM_DEBUG(lock, "File %p/"DFID" being reconfigured: %d.\n", + inode, PFID(&lli->lli_fid), reconf); + + lock_res_and_lock(lock); + lvb_ready = !!(lock->l_flags & LDLM_FL_LVB_READY); + unlock_res_and_lock(lock); + /* checking lvb_ready is racy but this is okay. The worst case is + * that multi processes may configure the file on the same time. */ + if (lvb_ready || !reconf) { + LDLM_LOCK_PUT(lock); + + rc = -ENODATA; + if (lvb_ready) { + /* layout_gen must be valid if layout lock is not + * cancelled and stripe has already set */ + *gen = lli->lli_layout_gen; + rc = 0; + } + ldlm_lock_decref(lockh, mode); + RETURN(rc); + } + + /* for layout lock, lmm is returned in lock's lvb. + * lvb_data is immutable if the lock is held so it's safe to access it + * without res lock. See the description in ldlm_lock_decref_internal() + * for the condition to free lvb_data of layout lock */ + if (lock->l_lvb_data != NULL) { + rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm, + lock->l_lvb_data, lock->l_lvb_len); + if (rc >= 0) { + *gen = LL_LAYOUT_GEN_EMPTY; + if (md.lsm != NULL) + *gen = md.lsm->lsm_layout_gen; + rc = 0; + } else { + CERROR("%s: file "DFID" unpackmd error: %d\n", + ll_get_fsname(inode->i_sb, NULL, 0), + PFID(&lli->lli_fid), rc); + } + } + if (rc < 0) { + LDLM_LOCK_PUT(lock); + ldlm_lock_decref(lockh, mode); + RETURN(rc); + } + + /* set layout to file. Unlikely this will fail as old layout was + * surely eliminated */ + memset(&conf, 0, sizeof conf); + conf.coc_opc = OBJECT_CONF_SET; + conf.coc_inode = inode; + conf.coc_lock = lock; + conf.u.coc_md = &md; + rc = ll_layout_conf(inode, &conf); + LDLM_LOCK_PUT(lock); + + ldlm_lock_decref(lockh, mode); + + if (md.lsm != NULL) + obd_free_memmd(sbi->ll_dt_exp, &md.lsm); + + /* wait for IO to complete if it's still being used. */ + if (rc == -EBUSY) { + CDEBUG(D_INODE, "%s: %p/"DFID" wait for layout reconf.\n", + ll_get_fsname(inode->i_sb, NULL, 0), + inode, PFID(&lli->lli_fid)); + + memset(&conf, 0, sizeof conf); + conf.coc_opc = OBJECT_CONF_WAIT; + conf.coc_inode = inode; + rc = ll_layout_conf(inode, &conf); + if (rc == 0) + rc = -EAGAIN; + + CDEBUG(D_INODE, "file: "DFID" waiting layout return: %d.\n", + PFID(&lli->lli_fid), rc); + } + + RETURN(rc); +} + +/** + * This function checks if there exists a LAYOUT lock on the client side, + * or enqueues it if it doesn't have one in cache. + * + * This function will not hold layout lock so it may be revoked any time after + * this function returns. Any operations depend on layout should be redone + * in that case. + * + * This function should be called before lov_io_init() to get an uptodate + * layout version, the caller should save the version number and after IO + * is finished, this function should be called again to verify that layout + * is not changed during IO time. + */ +int ll_layout_refresh(struct inode *inode, __u32 *gen) +{ + struct ll_inode_info *lli = ll_i2info(inode); + struct ll_sb_info *sbi = ll_i2sbi(inode); + struct md_op_data *op_data; + struct lookup_intent it; + struct lustre_handle lockh; + ldlm_mode_t mode; + struct ldlm_enqueue_info einfo = { .ei_type = LDLM_IBITS, + .ei_mode = LCK_CR, + .ei_cb_bl = ll_md_blocking_ast, + .ei_cb_cp = ldlm_completion_ast, + .ei_cbdata = NULL }; + int rc; + ENTRY; + + *gen = LL_LAYOUT_GEN_NONE; + if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK)) + RETURN(0); + + /* sanity checks */ + LASSERT(fid_is_sane(ll_inode2fid(inode))); + LASSERT(S_ISREG(inode->i_mode)); + + /* mostly layout lock is caching on the local side, so try to match + * it before grabbing layout lock mutex. */ + mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0); + if (mode != 0) { /* hit cached lock */ + rc = ll_layout_lock_set(&lockh, mode, inode, gen, false); + if (rc == 0) + RETURN(0); + + /* better hold lli_layout_mutex to try again otherwise + * it will have starvation problem. */ + } + + /* take layout lock mutex to enqueue layout lock exclusively. */ + mutex_lock(&lli->lli_layout_mutex); + +again: + /* try again. Maybe somebody else has done this. */ + mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0); + if (mode != 0) { /* hit cached lock */ + rc = ll_layout_lock_set(&lockh, mode, inode, gen, true); + if (rc == -EAGAIN) + goto again; + + mutex_unlock(&lli->lli_layout_mutex); + RETURN(rc); + } + + op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, + 0, 0, LUSTRE_OPC_ANY, NULL); + if (IS_ERR(op_data)) { + mutex_unlock(&lli->lli_layout_mutex); + RETURN(PTR_ERR(op_data)); + } + + /* have to enqueue one */ + memset(&it, 0, sizeof(it)); + it.it_op = IT_LAYOUT; + lockh.cookie = 0ULL; + + LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file %p/"DFID".\n", + ll_get_fsname(inode->i_sb, NULL, 0), inode, + PFID(&lli->lli_fid)); + + rc = md_enqueue(sbi->ll_md_exp, &einfo, &it, op_data, &lockh, + NULL, 0, NULL, 0); + if (it.d.lustre.it_data != NULL) + ptlrpc_req_finished(it.d.lustre.it_data); + it.d.lustre.it_data = NULL; + + ll_finish_md_op_data(op_data); + + md_set_lock_data(sbi->ll_md_exp, &it.d.lustre.it_lock_handle, inode, NULL); + + mode = it.d.lustre.it_lock_mode; + it.d.lustre.it_lock_mode = 0; + ll_intent_drop_lock(&it); + + if (rc == 0) { + /* set lock data in case this is a new lock */ + ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL); + rc = ll_layout_lock_set(&lockh, mode, inode, gen, true); + if (rc == -EAGAIN) + goto again; + } + mutex_unlock(&lli->lli_layout_mutex); + + RETURN(rc); +}