diff options
Diffstat (limited to 'xlators/storage/bd')
| -rw-r--r-- | xlators/storage/bd/Makefile.am | 3 | ||||
| -rw-r--r-- | xlators/storage/bd/src/Makefile.am | 20 | ||||
| -rw-r--r-- | xlators/storage/bd/src/bd-aio.c | 527 | ||||
| -rw-r--r-- | xlators/storage/bd/src/bd-aio.h | 41 | ||||
| -rw-r--r-- | xlators/storage/bd/src/bd-helper.c | 783 | ||||
| -rw-r--r-- | xlators/storage/bd/src/bd.c | 2404 | ||||
| -rw-r--r-- | xlators/storage/bd/src/bd.h | 178 |
7 files changed, 3956 insertions, 0 deletions
diff --git a/xlators/storage/bd/Makefile.am b/xlators/storage/bd/Makefile.am new file mode 100644 index 000000000..a985f42a8 --- /dev/null +++ b/xlators/storage/bd/Makefile.am @@ -0,0 +1,3 @@ +SUBDIRS = src + +CLEANFILES = diff --git a/xlators/storage/bd/src/Makefile.am b/xlators/storage/bd/src/Makefile.am new file mode 100644 index 000000000..3d93f7442 --- /dev/null +++ b/xlators/storage/bd/src/Makefile.am @@ -0,0 +1,20 @@ +if ENABLE_BD_XLATOR +xlator_LTLIBRARIES = bd.la +xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/storage + +bd_la_LDFLAGS = -module -avoid-version +LIBBD = -llvm2app -lrt +bd_la_SOURCES = bd.c bd-helper.c bd-aio.c +bd_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la $(LIBBD) $(LIBAIO) + +noinst_HEADERS = bd.h bd-aio.h + +AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \ + -I$(top_srcdir)/rpc/xdr/src \ + -I$(top_srcdir)/rpc/rpc-lib/src + +AM_CFLAGS = -fno-strict-aliasing -Wall $(GF_CFLAGS) + +CLEANFILES = + +endif diff --git a/xlators/storage/bd/src/bd-aio.c b/xlators/storage/bd/src/bd-aio.c new file mode 100644 index 000000000..62d4590f7 --- /dev/null +++ b/xlators/storage/bd/src/bd-aio.c @@ -0,0 +1,527 @@ +/* + Copyright IBM, Corp. 2013 + + This file is part of GlusterFS. + + Author: M. Mohan Kumar <mohan@in.ibm.com> + + Based on posix-aio.c + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#ifndef _CONFIG_H +#define _CONFIG_H +#include "config.h" +#endif + +#include <lvm2app.h> +#include <sys/uio.h> + +#include "xlator.h" +#include "glusterfs.h" +#include "defaults.h" +#include "bd.h" +#include "bd-aio.h" + +#ifdef HAVE_LIBAIO +#include <libaio.h> + +struct bd_aio_cb { + struct iocb iocb; + call_frame_t *frame; + struct iobuf *iobuf; + struct iobref *iobref; + struct iatt prebuf; + int op; + off_t offset; + fd_t *fd; +}; + +void +__bd_fd_set_odirect (fd_t *fd, bd_fd_t *bd_fd, int opflags, + off_t offset, size_t size) +{ + int odirect = 0; + int flags = 0; + int ret = 0; + + odirect = bd_fd->odirect; + + if ((fd->flags|opflags) & O_DIRECT) { + /* if instructed, use O_DIRECT always */ + odirect = 1; + } else { + /* else use O_DIRECT when feasible */ + if ((offset|size) & 0xfff) + odirect = 0; + else + odirect = 1; + } + + if (!odirect && bd_fd->odirect) { + flags = fcntl (bd_fd->fd, F_GETFL); + ret = fcntl (bd_fd->fd, F_SETFL, (flags & (~O_DIRECT))); + bd_fd->odirect = 0; + } + + if (odirect && !bd_fd->odirect) { + flags = fcntl (bd_fd->fd, F_GETFL); + ret = fcntl (bd_fd->fd, F_SETFL, (flags | O_DIRECT)); + bd_fd->odirect = 1; + } + + if (ret) { + gf_log (THIS->name, GF_LOG_WARNING, + "fcntl() failed (%s). fd=%d flags=%d pfd->odirect=%d", + strerror (errno), bd_fd->fd, flags, bd_fd->odirect); + } +} + +int +bd_aio_readv_complete (struct bd_aio_cb *paiocb, int res, int res2) +{ + call_frame_t *frame = NULL; + xlator_t *this = NULL; + struct iobuf *iobuf = NULL; + struct iatt postbuf = {0,}; + int op_ret = -1; + int op_errno = 0; + struct iovec iov; + struct iobref *iobref = NULL; + off_t offset = 0; + bd_attr_t *bdatt = NULL; + + frame = paiocb->frame; + this = frame->this; + iobuf = paiocb->iobuf; + offset = paiocb->offset; + + if (res < 0) { + op_ret = -1; + op_errno = -res; + gf_log (this->name, GF_LOG_ERROR, + "readv(async) failed fd=%p,size=%lu,offset=%llu (%d/%s)", + paiocb->fd, paiocb->iocb.u.c.nbytes, + (unsigned long long) paiocb->offset, + res, strerror (op_errno)); + goto out; + } + + bd_inode_ctx_get (paiocb->fd->inode, this, &bdatt); + memcpy (&postbuf, &bdatt->iatt, sizeof (struct iatt)); + + op_ret = res; + op_errno = 0; + + iobref = iobref_new (); + if (!iobref) { + op_ret = -1; + op_errno = ENOMEM; + goto out; + } + + iobref_add (iobref, iobuf); + + iov.iov_base = iobuf_ptr (iobuf); + iov.iov_len = op_ret; + + /* Hack to notify higher layers of EOF. */ + if (!postbuf.ia_size || (offset + iov.iov_len) >= postbuf.ia_size) + op_errno = ENOENT; + +out: + STACK_UNWIND_STRICT (readv, frame, op_ret, op_errno, &iov, 1, + &postbuf, iobref, NULL); + if (iobuf) + iobuf_unref (iobuf); + if (iobref) + iobref_unref (iobref); + + GF_FREE (paiocb); + + return 0; +} + +int +bd_aio_readv (call_frame_t *frame, xlator_t *this, fd_t *fd, + size_t size, off_t offset, uint32_t flags, dict_t *xdata) +{ + int32_t op_errno = EINVAL; + int _fd = -1; + struct iobuf *iobuf = NULL; + bd_fd_t *bd_fd = NULL; + int ret = -1; + struct bd_aio_cb *paiocb = NULL; + bd_priv_t *priv = NULL; + struct iocb *iocb = NULL; + bd_attr_t *bdatt = NULL; + + VALIDATE_OR_GOTO (frame, err); + VALIDATE_OR_GOTO (this, err); + VALIDATE_OR_GOTO (fd, err); + + priv = this->private; + + ret = bd_fd_ctx_get (this, fd, &bd_fd); + if (ret < 0 || !bd_fd) { + STACK_WIND (frame, default_readv_cbk, FIRST_CHILD (this), + FIRST_CHILD (this)->fops->readv, fd, size, offset, + flags, xdata); + return 0; + } + _fd = bd_fd->fd; + bd_inode_ctx_get (fd->inode, this, &bdatt); + if (!size) { + op_errno = EINVAL; + gf_log (this->name, GF_LOG_WARNING, "size=%"GF_PRI_SIZET, size); + goto err; + } + + iobuf = iobuf_get2 (this->ctx->iobuf_pool, size); + if (!iobuf) { + op_errno = ENOMEM; + goto err; + } + + paiocb = CALLOC (1, sizeof (*paiocb)); + if (!paiocb) { + op_errno = ENOMEM; + goto err; + } + + paiocb->frame = frame; + paiocb->iobuf = iobuf; + paiocb->offset = offset; + paiocb->op = GF_FOP_READ; + paiocb->fd = fd; + + paiocb->iocb.data = paiocb; + paiocb->iocb.aio_fildes = _fd; + paiocb->iocb.aio_lio_opcode = IO_CMD_PREAD; + paiocb->iocb.aio_reqprio = 0; + paiocb->iocb.u.c.buf = iobuf_ptr (iobuf); + paiocb->iocb.u.c.nbytes = size; + paiocb->iocb.u.c.offset = offset; + + iocb = &paiocb->iocb; + + LOCK (&fd->lock); + { + __bd_fd_set_odirect (fd, bd_fd, flags, offset, size); + + ret = io_submit (priv->ctxp, 1, &iocb); + } + UNLOCK (&fd->lock); + + if (ret != 1) { + gf_log (this->name, GF_LOG_ERROR, + "io_submit() returned %d", ret); + op_errno = -ret; + goto err; + } + + return 0; +err: + STACK_UNWIND_STRICT (readv, frame, -1, op_errno, 0, 0, 0, 0, 0); + if (iobuf) + iobuf_unref (iobuf); + + if (paiocb) + GF_FREE (paiocb); + + return 0; +} + +int +bd_aio_writev_complete (struct bd_aio_cb *paiocb, int res, int res2) +{ + call_frame_t *frame = NULL; + xlator_t *this = NULL; + struct iatt prebuf = {0,}; + struct iatt postbuf = {0,}; + int op_ret = -1; + int op_errno = 0; + bd_attr_t *bdatt = NULL; + + frame = paiocb->frame; + prebuf = paiocb->prebuf; + this = frame->this; + + if (res < 0) { + op_ret = -1; + op_errno = -res; + gf_log (this->name, GF_LOG_ERROR, + "writev(async) failed fd=%p,offset=%llu (%d/%s)", + paiocb->fd, (unsigned long long) paiocb->offset, res, + strerror (op_errno)); + + goto out; + } + + bd_inode_ctx_get (paiocb->fd->inode, this, &bdatt); + bd_update_amtime (&bdatt->iatt, GF_SET_ATTR_MTIME); + memcpy (&postbuf, &bdatt->iatt, sizeof (struct iatt)); + + op_ret = res; + op_errno = 0; + +out: + STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, &prebuf, &postbuf, + NULL); + + if (paiocb) { + if (paiocb->iobref) + iobref_unref (paiocb->iobref); + GF_FREE (paiocb); + } + + return 0; +} + +int +bd_aio_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, + struct iovec *iov, int count, off_t offset, uint32_t flags, + struct iobref *iobref, dict_t *xdata) +{ + int32_t op_errno = EINVAL; + int _fd = -1; + bd_fd_t *bd_fd = NULL; + int ret = -1; + struct bd_aio_cb *paiocb = NULL; + bd_priv_t *priv = NULL; + struct iocb *iocb = NULL; + bd_attr_t *bdatt = NULL; + + VALIDATE_OR_GOTO (frame, err); + VALIDATE_OR_GOTO (this, err); + VALIDATE_OR_GOTO (fd, err); + + priv = this->private; + + ret = bd_fd_ctx_get (this, fd, &bd_fd); + if (ret < 0 || !bd_fd) { + STACK_WIND (frame, default_writev_cbk, + FIRST_CHILD(this), FIRST_CHILD(this)->fops->writev, + fd, iov, count, offset, flags, iobref, xdata); + return 0; + } + + bd_inode_ctx_get (fd->inode, this, &bdatt); + + _fd = bd_fd->fd; + + paiocb = CALLOC (1, sizeof (*paiocb)); + if (!paiocb) { + op_errno = ENOMEM; + goto err; + } + + + paiocb->frame = frame; + paiocb->offset = offset; + paiocb->op = GF_FOP_WRITE; + paiocb->fd = fd; + + paiocb->iocb.data = paiocb; + paiocb->iocb.aio_fildes = _fd; + paiocb->iobref = iobref_ref (iobref); + paiocb->iocb.aio_lio_opcode = IO_CMD_PWRITEV; + paiocb->iocb.aio_reqprio = 0; + paiocb->iocb.u.v.vec = iov; + paiocb->iocb.u.v.nr = count; + paiocb->iocb.u.v.offset = offset; + + iocb = &paiocb->iocb; + + memcpy (&paiocb->prebuf, &bdatt->iatt, sizeof (struct iatt)); + LOCK (&fd->lock); + { + __bd_fd_set_odirect (fd, bd_fd, flags, offset, + iov_length (iov, count)); + + ret = io_submit (priv->ctxp, 1, &iocb); + } + UNLOCK (&fd->lock); + + if (ret != 1) { + gf_log (this->name, GF_LOG_ERROR, + "io_submit() returned %d", ret); + op_errno = -ret; + goto err; + } + + return 0; +err: + STACK_UNWIND_STRICT (writev, frame, -1, op_errno, 0, 0, 0); + + if (paiocb) { + if (paiocb->iobref) + iobref_unref (paiocb->iobref); + GF_FREE (paiocb); + } + + return 0; +} + +void * +bd_aio_thread (void *data) +{ + xlator_t *this = NULL; + bd_priv_t *priv = NULL; + int ret = 0; + int i = 0; + struct io_event *event = NULL; + struct bd_aio_cb *paiocb = NULL; + struct io_event events[BD_AIO_MAX_NR_GETEVENTS]; + struct timespec ts = {0, }; + + this = data; + THIS = this; + priv = this->private; + + ts.tv_sec = 5; + for (;;) { + memset (&events[0], 0, sizeof (events)); + ret = io_getevents (priv->ctxp, 1, BD_AIO_MAX_NR_GETEVENTS, + &events[0], &ts); + if (ret < 0) { + if (ret == -EINTR) + continue; + gf_log (this->name, GF_LOG_ERROR, + "io_getevents() returned %d, exiting", ret); + break; + } + + for (i = 0; i < ret; i++) { + event = &events[i]; + + paiocb = event->data; + + switch (paiocb->op) { + case GF_FOP_READ: + bd_aio_readv_complete (paiocb, event->res, + event->res2); + break; + case GF_FOP_WRITE: + bd_aio_writev_complete (paiocb, event->res, + event->res2); + break; + default: + gf_log (this->name, GF_LOG_ERROR, + "unknown op %d found in piocb", + paiocb->op); + break; + } + } + } + + return NULL; +} + +int +bd_aio_init (xlator_t *this) +{ + bd_priv_t *priv = NULL; + int ret = 0; + + priv = this->private; + + ret = io_setup (BD_AIO_MAX_NR_EVENTS, &priv->ctxp); + if ((ret == -1 && errno == ENOSYS) || ret == -ENOSYS) { + gf_log (this->name, GF_LOG_WARNING, + "Linux AIO not available at run-time." + " Continuing with synchronous IO"); + ret = 0; + goto out; + } + + if (ret < 0) { + gf_log (this->name, GF_LOG_WARNING, + "io_setup() failed. ret=%d, errno=%d", + ret, errno); + goto out; + } + + ret = pthread_create (&priv->aiothread, NULL, + bd_aio_thread, this); + if (ret != 0) { + io_destroy (priv->ctxp); + goto out; + } + + this->fops->readv = bd_aio_readv; + this->fops->writev = bd_aio_writev; +out: + return ret; +} + + +int +bd_aio_on (xlator_t *this) +{ + bd_priv_t *priv = NULL; + int ret = 0; + + priv = this->private; + + if (!priv->aio_init_done) { + ret = bd_aio_init (this); + if (ret == 0) + priv->aio_capable = _gf_true; + else + priv->aio_capable = _gf_false; + priv->aio_init_done = _gf_true; + } + + if (priv->aio_capable) { + this->fops->readv = bd_aio_readv; + this->fops->writev = bd_aio_writev; + } + + return ret; +} + +int +bd_aio_off (xlator_t *this) +{ + this->fops->readv = bd_readv; + this->fops->writev = bd_writev; + + return 0; +} + +#else + +int +bd_aio_on (xlator_t *this) +{ + gf_log (this->name, GF_LOG_INFO, + "Linux AIO not available at build-time." + " Continuing with synchronous IO"); + return 0; +} + +int +bd_aio_off (xlator_t *this) +{ + gf_log (this->name, GF_LOG_INFO, + "Linux AIO not available at build-time." + " Continuing with synchronous IO"); + return 0; +} + +void +__bd_fd_set_odirect (fd_t *fd, struct bd_fd *pfd, int opflags, + off_t offset, size_t size) +{ + xlator_t *this = THIS; + gf_log (this->name, GF_LOG_INFO, + "Linux AIO not available at build-time." + " Continuing with synchronous IO"); + return; +} +#endif diff --git a/xlators/storage/bd/src/bd-aio.h b/xlators/storage/bd/src/bd-aio.h new file mode 100644 index 000000000..16f686a4c --- /dev/null +++ b/xlators/storage/bd/src/bd-aio.h @@ -0,0 +1,41 @@ +/* + Copyright IBM, Corp. 2013 + + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ +#ifndef _BD_AIO_H +#define _BD_AIO_H + +#ifndef _CONFIG_H +#define _CONFIG_H +#include "config.h" +#endif + +#include "xlator.h" +#include "glusterfs.h" + +/* + * Maximum number of concurrently submitted IO events. The heaviest load + * GlusterFS has been able to handle had 60-80 concurrent calls + */ +#define BD_AIO_MAX_NR_EVENTS 256 + +/* Maximum number of completed IO operations to reap per getevents syscall */ +#define BD_AIO_MAX_NR_GETEVENTS 16 + +int bd_aio_on (xlator_t *this); +int bd_aio_off (xlator_t *this); + +int bd_readv (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size, + off_t offset, uint32_t flags, dict_t *xdata); + +int bd_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, + struct iovec *vector, int32_t count, off_t offset, + uint32_t flags, struct iobref *iobref, dict_t *xdata); + +#endif /* !_BD_AIO_H */ diff --git a/xlators/storage/bd/src/bd-helper.c b/xlators/storage/bd/src/bd-helper.c new file mode 100644 index 000000000..5525e346b --- /dev/null +++ b/xlators/storage/bd/src/bd-helper.c @@ -0,0 +1,783 @@ +#ifndef _CONFIG_H +#define _CONFIG_H +#include "config.h" +#endif +#include <lvm2app.h> +#ifdef HAVE_LIBAIO +#include <libaio.h> +#endif + +#include "bd.h" +#include "run.h" + +int +bd_inode_ctx_set (inode_t *inode, xlator_t *this, bd_attr_t *ctx) +{ + int ret = -1; + uint64_t ctx_int = 0; + + GF_VALIDATE_OR_GOTO (this->name, inode, out); + GF_VALIDATE_OR_GOTO (this->name, ctx, out); + + ctx_int = (long)ctx; + ret = inode_ctx_set (inode, this, &ctx_int); +out: + return ret; +} + +int +bd_inode_ctx_get (inode_t *inode, xlator_t *this, bd_attr_t **ctx) +{ + int ret = -1; + uint64_t ctx_int = 0; + + GF_VALIDATE_OR_GOTO (this->name, inode, out); + ret = inode_ctx_get (inode, this, &ctx_int); + if (ret) + return ret; + if (ctx) + *ctx = (bd_attr_t *) ctx_int; +out: + return ret; +} + +void +bd_local_free (xlator_t *this, bd_local_t *local) +{ + if (!local) + return; + if (local->fd) + fd_unref (local->fd); + else if (local->loc.path) + loc_wipe (&local->loc); + if (local->dict) + dict_unref (local->dict); + if (local->inode) + inode_unref (local->inode); + if (local->bdatt) { + GF_FREE (local->bdatt->type); + GF_FREE (local->bdatt); + } + mem_put (local); + local = NULL; +} + +bd_local_t * +bd_local_init (call_frame_t *frame, xlator_t *this) +{ + frame->local = mem_get0 (this->local_pool); + if (!frame->local) + return NULL; + + return frame->local; +} + +/* + * VG are set with the tag in GF_XATTR_VOL_ID_KEY:<uuid> format. + * This function validates this tag agains volume-uuid. Also goes + * through LV list to find out if a thin-pool is configured or not. + */ +int bd_scan_vg (xlator_t *this, bd_priv_t *priv) +{ + vg_t brick = NULL; + data_t *tmp_data = NULL; + struct dm_list *tags = NULL; + int op_ret = -1; + uuid_t dict_uuid = {0, }; + uuid_t vg_uuid = {0, }; + gf_boolean_t uuid = _gf_false; + lvm_str_list_t *strl = NULL; + struct dm_list *lv_dm_list = NULL; + lv_list_t *lv_list = NULL; + struct dm_list *dm_seglist = NULL; + lvseg_list_t *seglist = NULL; + lvm_property_value_t prop = {0, }; + gf_boolean_t thin = _gf_false; + const char *lv_name = NULL; + + brick = lvm_vg_open (priv->handle, priv->vg, "w", 0); + if (!brick) { + gf_log (this->name, GF_LOG_CRITICAL, "VG %s is not found", + priv->vg); + return ENOENT; + } + + lv_dm_list = lvm_vg_list_lvs (brick); + if (!lv_dm_list) + goto check; + + dm_list_iterate_items (lv_list, lv_dm_list) { + dm_seglist = lvm_lv_list_lvsegs (lv_list->lv); + if (!dm_seglist) + continue; + dm_list_iterate_items (seglist, dm_seglist) { + prop = lvm_lvseg_get_property (seglist->lvseg, + "segtype"); + if (!prop.is_valid || !prop.value.string) + continue; + if (!strcmp (prop.value.string, "thin-pool")) { + thin = _gf_true; + lv_name = lvm_lv_get_name (lv_list->lv); + priv->pool = gf_strdup (lv_name); + gf_log (THIS->name, GF_LOG_INFO, "Thin Pool " + "\"%s\" will be used for thin LVs", + lv_name); + break; + } + } + } + +check: + /* If there is no volume-id set in dict, we cant validate */ + tmp_data = dict_get (this->options, "volume-id"); + if (!tmp_data) { + op_ret = 0; + goto out; + } + + op_ret = uuid_parse (tmp_data->data, dict_uuid); + if (op_ret < 0) { + gf_log (this->name, GF_LOG_ERROR, + "wrong volume-id (%s) set in volume file", + tmp_data->data); + op_ret = -1; + goto out; + } + + tags = lvm_vg_get_tags (brick); + if (!tags) { /* no tags in the VG */ + gf_log (this->name, GF_LOG_ERROR, + "Extended attribute trusted.glusterfs." + "volume-id is absent"); + op_ret = -1; + goto out; + } + dm_list_iterate_items (strl, tags) { + if (!strncmp (strl->str, GF_XATTR_VOL_ID_KEY, + strlen (GF_XATTR_VOL_ID_KEY))) { + uuid = _gf_true; + break; + } + } + /* UUID tag is not set in VG */ + if (!uuid) { + gf_log (this->name, GF_LOG_ERROR, + "Extended attribute trusted.glusterfs." + "volume-id is absent"); + op_ret = -1; + goto out; + } + + op_ret = uuid_parse (strl->str + strlen (GF_XATTR_VOL_ID_KEY) + 1, + vg_uuid); + if (op_ret < 0) { + gf_log (this->name, GF_LOG_ERROR, + "wrong volume-id (%s) set in VG", strl->str); + op_ret = -1; + goto out; + } + if (uuid_compare (dict_uuid, vg_uuid)) { + gf_log (this->name, GF_LOG_ERROR, + "mismatching volume-id (%s) received. " + "already is a part of volume %s ", + tmp_data->data, vg_uuid); + op_ret = -1; + goto out; + } + + op_ret = 0; + +out: + lvm_vg_close (brick); + + if (!thin) + gf_log (THIS->name, GF_LOG_WARNING, "No thin pool found in " + "VG %s\n", priv->vg); + else + priv->caps |= BD_CAPS_THIN; + + return op_ret; +} + +/* FIXME: Move this code to common place, so posix and bd xlator can use */ +char * +page_aligned_alloc (size_t size, char **aligned_buf) +{ + char *alloc_buf = NULL; + char *buf = NULL; + + alloc_buf = GF_CALLOC (1, (size + ALIGN_SIZE), gf_common_mt_char); + if (!alloc_buf) + return NULL; + /* page aligned buffer */ + buf = GF_ALIGN_BUF (alloc_buf, ALIGN_SIZE); + *aligned_buf = buf; + + return alloc_buf; +} + +static int +__bd_fd_ctx_get (xlator_t *this, fd_t *fd, bd_fd_t **bdfd_p) +{ + int ret = -1; + int _fd = -1; + char *devpath = NULL; + bd_fd_t *bdfd = NULL; + uint64_t tmp_bdfd = 0; + bd_priv_t *priv = this->private; + bd_gfid_t gfid = {0, }; + bd_attr_t *bdatt = NULL; + + /* not bd file */ + if (fd->inode->ia_type != IA_IFREG || + bd_inode_ctx_get (fd->inode, this, &bdatt)) + return 0; + + ret = __fd_ctx_get (fd, this, &tmp_bdfd); + if (ret == 0) { + bdfd = (void *)(long) tmp_bdfd; + *bdfd_p = bdfd; + return 0; + } + + uuid_utoa_r (fd->inode->gfid, gfid); + asprintf (&devpath, "/dev/%s/%s", priv->vg, gfid); + if (!devpath) + goto out; + + _fd = open (devpath, O_RDWR | O_LARGEFILE, 0); + if (_fd < 0) { + ret = errno; + gf_log (this->name, GF_LOG_ERROR, "open on %s: %s", devpath, + strerror (ret)); + goto out; + } + bdfd = GF_CALLOC (1, sizeof(bd_fd_t), gf_bd_fd); + BD_VALIDATE_MEM_ALLOC (bdfd, ret, out); + + bdfd->fd = _fd; + bdfd->flag = O_RDWR | O_LARGEFILE; + if (__fd_ctx_set (fd, this, (uint64_t)(long)bdfd) < 0) { + gf_log (this->name, GF_LOG_WARNING, + "failed to set the fd context fd=%p", fd); + goto out; + } + + *bdfd_p = bdfd; + + ret = 0; +out: + FREE (devpath); + if (ret) { + close (_fd); + GF_FREE (bdfd); + } + return ret; +} + +int +bd_fd_ctx_get (xlator_t *this, fd_t *fd, bd_fd_t **bdfd) +{ + int ret; + + /* FIXME: Is it ok to fd->lock here ? */ + LOCK (&fd->lock); + { + ret = __bd_fd_ctx_get (this, fd, bdfd); + } + UNLOCK (&fd->lock); + + return ret; +} + +/* + * Validates if LV exists for given inode or not. + * Returns 0 if LV exists and size also matches. + * If LV does not exist -1 returned + * If LV size mismatches, returnes 1 also lv_size is updated with actual + * size + */ +int +bd_validate_bd_xattr (xlator_t *this, char *bd, char **type, + uint64_t *lv_size, uuid_t uuid) +{ + char *path = NULL; + int ret = -1; + bd_gfid_t gfid = {0, }; + bd_priv_t *priv = this->private; + struct stat stbuf = {0, }; + uint64_t size = 0; + vg_t vg = NULL; + lv_t lv = NULL; + char *bytes = NULL; + + bytes = strrchr (bd, ':'); + if (bytes) { + *bytes = '\0'; + bytes++; + gf_string2bytesize (bytes, &size); + } + + if (strcmp (bd, BD_LV) && strcmp (bd, BD_THIN)) { + gf_log (this->name, GF_LOG_WARNING, + "invalid xattr %s", bd); + return -1; + } + *type = gf_strdup (bd); + + /* + * Check if LV really exist, there could be a failure + * after setxattr and successful LV creation + */ + uuid_utoa_r (uuid, gfid); + gf_asprintf (&path, "/dev/%s/%s", priv->vg, gfid); + if (!path) { + gf_log (this->name, GF_LOG_WARNING, + "insufficient memory"); + return 0; + } + + /* Destination file does not exist */ + if (stat (path, &stbuf)) { + gf_log (this->name, GF_LOG_WARNING, + "lstat failed for path %s", path); + return -1; + } + + vg = lvm_vg_open (priv->handle, priv->vg, "r", 0); + if (!vg) { + gf_log (this->name, GF_LOG_WARNING, + "VG %s does not exist?", priv->vg); + ret = -1; + goto out; + } + + lv = lvm_lv_from_name (vg, gfid); + if (!lv) { + gf_log (this->name, GF_LOG_WARNING, + "LV %s does not exist", gfid); + ret = -1; + goto out; + } + + *lv_size = lvm_lv_get_size (lv); + if (size == *lv_size) { + ret = 0; + goto out; + } + + ret = 1; + +out: + if (vg) + lvm_vg_close (vg); + + GF_FREE (path); + return ret; +} + +static int +create_thin_lv (char *vg, char *pool, char *lv, uint64_t extent) +{ + int ret = -1; + runner_t runner = {0, }; + char *path = NULL; + struct stat stat = {0, }; + + runinit (&runner); + runner_add_args (&runner, LVM_CREATE, NULL); + runner_add_args (&runner, "--thin", NULL); + runner_argprintf (&runner, "%s/%s", vg, pool); + runner_add_args (&runner, "--name", NULL); + runner_argprintf (&runner, "%s", lv); + runner_add_args (&runner, "--virtualsize", NULL); + runner_argprintf (&runner, "%ldB", extent); + runner_start (&runner); + runner_end (&runner); + + gf_asprintf (&path, "/dev/%s/%s", vg, lv); + if (!path) { + ret = ENOMEM; + goto out; + } + if (lstat (path, &stat) < 0) + ret = EAGAIN; + else + ret = 0; +out: + GF_FREE (path); + return ret; +} + +int +bd_create (uuid_t uuid, uint64_t size, char *type, bd_priv_t *priv) +{ + int ret = 0; + vg_t vg = NULL; + bd_gfid_t gfid = {0, }; + + uuid_utoa_r (uuid, gfid); + + if (!strcmp (type, BD_THIN)) + return create_thin_lv (priv->vg, priv->pool, gfid, + size); + + vg = lvm_vg_open (priv->handle, priv->vg, "w", 0); + if (!vg) { + gf_log (THIS->name, GF_LOG_WARNING, "opening VG %s failed", + priv->vg); + return ENOENT; + } + + if (!lvm_vg_create_lv_linear (vg, gfid, size)) { + gf_log (THIS->name, GF_LOG_WARNING, "lvm_vg_create_lv_linear " + "failed"); + ret = errno; + } + + lvm_vg_close (vg); + + return ret; +} + +int32_t +bd_resize (bd_priv_t *priv, uuid_t uuid, off_t size) +{ + uint64_t new_size = 0; + runner_t runner = {0, }; + bd_gfid_t gfid = {0, }; + int ret = 0; + vg_t vg = NULL; + lv_t lv = NULL; + + uuid_utoa_r (uuid, gfid); + + runinit (&runner); + + runner_add_args (&runner, LVM_RESIZE, NULL); + runner_argprintf (&runner, "%s/%s", priv->vg, gfid); + runner_argprintf (&runner, "-L%ldb", size); + runner_add_args (&runner, "-f", NULL); + + runner_start (&runner); + runner_end (&runner); + + vg = lvm_vg_open (priv->handle, priv->vg, "w", 0); + if (!vg) { + gf_log (THIS->name, GF_LOG_WARNING, "opening VG %s failed", + priv->vg); + return EAGAIN; + } + + lv = lvm_lv_from_name (vg, gfid); + if (!lv) { + gf_log (THIS->name, GF_LOG_WARNING, "LV %s not found", gfid); + ret = EIO; + goto out; + } + new_size = lvm_lv_get_size (lv); + + if (new_size != size) { + gf_log (THIS->name, GF_LOG_WARNING, "resized LV size %ld does " + "not match requested size %ld", new_size, size); + ret = EIO; + } + +out: + lvm_vg_close (vg); + return ret; +} + +uint64_t +bd_get_default_extent (bd_priv_t *priv) +{ + vg_t vg = NULL; + uint64_t size = 0; + + vg = lvm_vg_open (priv->handle, priv->vg, "w", 0); + if (!vg) { + gf_log (THIS->name, GF_LOG_WARNING, "opening VG %s failed", + priv->vg); + return 0; + } + + size = lvm_vg_get_extent_size (vg); + + lvm_vg_close (vg); + + return size; +} + +/* + * Adjusts the user specified size to VG specific extent size + */ +uint64_t +bd_adjust_size (bd_priv_t *priv, uint64_t size) +{ + uint64_t extent = 0; + uint64_t nr_ex = 0; + + extent = bd_get_default_extent (priv); + if (!extent) + return 0; + + nr_ex = size / extent; + if (size % extent) + nr_ex++; + + size = extent * nr_ex; + + return size; +} + +int +bd_delete_lv (bd_priv_t *priv, const char *lv_name, int *op_errno) +{ + vg_t vg = NULL; + lv_t lv = NULL; + int ret = -1; + + *op_errno = 0; + vg = lvm_vg_open (priv->handle, priv->vg, "w", 0); + if (!vg) { + gf_log (THIS->name, GF_LOG_WARNING, "opening VG %s failed", + priv->vg); + *op_errno = ENOENT; + return -1; + } + lv = lvm_lv_from_name (vg, lv_name); + if (!lv) { + gf_log (THIS->name, GF_LOG_WARNING, "No such LV %s", lv_name); + *op_errno = ENOENT; + goto out; + } + ret = lvm_vg_remove_lv (lv); + if (ret < 0) { + gf_log (THIS->name, GF_LOG_WARNING, "removing LV %s failed", + lv_name); + *op_errno = errno; + goto out; + } +out: + lvm_vg_close (vg); + + return ret; +} + +inline void +bd_update_amtime(struct iatt *iatt, int flag) +{ + struct timespec ts = {0, }; + + clock_gettime (CLOCK_REALTIME, &ts); + if (flag & GF_SET_ATTR_ATIME) { + iatt->ia_atime = ts.tv_sec; + iatt->ia_atime_nsec = ts.tv_nsec; + } + if (flag & GF_SET_ATTR_MTIME) { + iatt->ia_mtime = ts.tv_sec; + iatt->ia_mtime_nsec = ts.tv_nsec; + } +} + +int +bd_snapshot_create (bd_local_t *local, bd_priv_t *priv) +{ + char *path = NULL; + bd_gfid_t dest = {0, }; + bd_gfid_t origin = {0, }; + int ret = 0; + runner_t runner = {0, }; + struct stat stat = {0, }; + + uuid_utoa_r (local->dloc->gfid, dest); + uuid_utoa_r (local->loc.gfid, origin); + + gf_asprintf (&path, "/dev/%s/%s", priv->vg, dest); + if (!path) { + gf_log (THIS->name, GF_LOG_WARNING, + "Insufficient memory"); + return ENOMEM; + } + + runinit (&runner); + runner_add_args (&runner, LVM_CREATE, NULL); + runner_add_args (&runner, "--snapshot", NULL); + runner_argprintf (&runner, "/dev/%s/%s", priv->vg, origin); + runner_add_args (&runner, "--name", NULL); + runner_argprintf (&runner, "%s", dest); + if (strcmp (local->bdatt->type, BD_THIN)) + runner_argprintf (&runner, "-L%ldB", local->size); + runner_start (&runner); + runner_end (&runner); + + if (lstat (path, &stat) < 0) + ret = EIO; + + GF_FREE (path); + return ret; +} + +int +bd_clone (bd_local_t *local, bd_priv_t *priv) +{ + int ret = ENOMEM; + int fd1 = -1; + int fd2 = -1; + int i = 0; + char *buff = NULL; + ssize_t bytes = 0; + char *spath = NULL; + char *dpath = NULL; + struct iovec *vec = NULL; + bd_gfid_t source = {0, }; + bd_gfid_t dest = {0, }; + void *bufp[IOV_NR] = {0, }; + + vec = GF_CALLOC (IOV_NR, sizeof (struct iovec), gf_common_mt_iovec); + if (!vec) + return ENOMEM; + + for (i = 0; i < IOV_NR; i++) { + bufp[i] = page_aligned_alloc (IOV_SIZE, &buff); + if (!buff) + goto out; + vec[i].iov_base = buff; + vec[i].iov_len = IOV_SIZE; + } + + uuid_utoa_r (local->loc.gfid, source); + uuid_utoa_r (local->dloc->gfid, dest); + + gf_asprintf (&spath, "/dev/%s/%s", priv->vg, source); + gf_asprintf (&dpath, "/dev/%s/%s", priv->vg, dest); + if (!spath || !dpath) + goto out; + + ret = bd_create (local->dloc->gfid, local->size, + local->bdatt->type, priv); + if (ret) + goto out; + + fd1 = open (spath, O_RDONLY | O_DIRECT); + if (fd1 < 0) { + ret = errno; + goto out; + } + fd2 = open (dpath, O_WRONLY | O_DIRECT); + if (fd2 < 0) { + ret = errno; + goto out; + } + + while (1) { + bytes = readv (fd1, vec, IOV_NR); + if (bytes < 0) { + ret = errno; + gf_log (THIS->name, GF_LOG_WARNING, "read failed: %s", + strerror (ret)); + goto out; + } + if (!bytes) + break; + bytes = writev (fd2, vec, IOV_NR); + if (bytes < 0) { + ret = errno; + gf_log (THIS->name, GF_LOG_WARNING, + "write failed: %s", strerror (ret)); + goto out; + } + } + ret = 0; + +out: + for (i = 0; i < IOV_NR; i++) + GF_FREE (bufp[i]); + GF_FREE (vec); + + if (fd1 != -1) + close (fd1); + if (fd2 != -1) + close (fd2); + + FREE (spath); + FREE (dpath); + + return ret; +} + +/* + * Merges snapshot LV to origin LV and returns status + */ +int +bd_merge (bd_priv_t *priv, uuid_t gfid) +{ + bd_gfid_t dest = {0, }; + char *path = NULL; + struct stat stat = {0, }; + runner_t runner = {0, }; + int ret = 0; + + uuid_utoa_r (gfid, dest); + gf_asprintf (&path, "/dev/%s/%s", priv->vg, dest); + + runinit (&runner); + runner_add_args (&runner, LVM_CONVERT, NULL); + runner_add_args (&runner, "--merge", NULL); + runner_argprintf (&runner, "%s", path); + runner_start (&runner); + runner_end (&runner); + + if (!lstat (path, &stat)) + ret = EIO; + + GF_FREE (path); + + return ret; +} + +int +bd_get_origin (bd_priv_t *priv, loc_t *loc, fd_t *fd, dict_t *dict) +{ + vg_t brick = NULL; + lvm_property_value_t prop = {0, }; + lv_t lv = NULL; + int ret = -1; + bd_gfid_t gfid = {0, }; + inode_t *inode = NULL; + char *origin = NULL; + + brick = lvm_vg_open (priv->handle, priv->vg, "w", 0); + if (!brick) { + gf_log (THIS->name, GF_LOG_CRITICAL, "VG %s is not found", + priv->vg); + return ENOENT; + } + + if (fd) + inode = fd->inode; + else + inode = loc->inode; + + uuid_utoa_r (inode->gfid, gfid); + lv = lvm_lv_from_name (brick, gfid); + if (!lv) { + gf_log (THIS->name, GF_LOG_CRITICAL, "LV %s not found", gfid); + ret = ENOENT; + goto out; + } + + prop = lvm_lv_get_property (lv, "origin"); + if (!prop.is_valid || !prop.value.string) { + ret = ENODATA; + goto out; + } + + origin = gf_strdup (prop.value.string); + ret = dict_set_dynstr (dict, BD_ORIGIN, origin); + +out: + lvm_vg_close (brick); + return ret; +} + diff --git a/xlators/storage/bd/src/bd.c b/xlators/storage/bd/src/bd.c new file mode 100644 index 000000000..405474c58 --- /dev/null +++ b/xlators/storage/bd/src/bd.c @@ -0,0 +1,2404 @@ +/* + BD translator V2 - Exports Block devices on server side as regular + files to client + + Now only exporting Logical volumes supported. + + Copyright IBM, Corp. 2013 + + This file is part of GlusterFS. + + Author: + M. Mohan Kumar <mohan@in.ibm.com> + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#ifndef _CONFIG_H +#define _CONFIG_H +#include "config.h" +#endif +#include <lvm2app.h> +#include <openssl/md5.h> +#include <time.h> +#include <linux/fs.h> +#include <sys/ioctl.h> +#ifdef HAVE_LIBAIO +#include <libaio.h> +#endif + +#include "bd.h" +#include "bd-aio.h" +#include "defaults.h" +#include "glusterfs3-xdr.h" +#include "run.h" +#include "protocol-common.h" +#include "checksum.h" + +/* + * Call back function for setxattr and removexattr. + * does not do anything. FIXME: How to handle remove/setxattr failure + */ +int +bd_null_rmsetxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, dict_t *xdata) +{ + STACK_DESTROY (frame->root); + return 0; +} + +/* + * returns 0 if a file is mapped to BD or not. + */ +int +bd_get_bd_info (call_frame_t *frame, xlator_t *this, dict_t *xattr, uuid_t gfid, + char **type, uint64_t *size) +{ + char *bd_xattr = NULL; + char *bd = NULL; + int ret = -1; + loc_t loc = {0, }; + dict_t *dict = NULL; + char *p = NULL; + call_frame_t *bd_frame = NULL; + + if (!xattr) + return 1; + + if (dict_get_str (xattr, BD_XATTR, &p)) + return 1; + + bd_xattr = gf_strdup (p); + + memcpy (loc.gfid, gfid, sizeof (uuid_t)); + + bd_frame = copy_frame (frame); + BD_VALIDATE_MEM_ALLOC (bd_frame, ret, out); + + ret = bd_validate_bd_xattr (this, bd_xattr, type, size, gfid); + if (ret < 0) {/* LV does not exist */ + STACK_WIND (bd_frame, bd_null_rmsetxattr_cbk, FIRST_CHILD (this), + FIRST_CHILD (this)->fops->removexattr, &loc, + BD_XATTR, NULL); + + gf_log (this->name, GF_LOG_WARNING, + "Mapped LV not available for posix file <gfid:%s>, " + "deleting mapping", uuid_utoa (gfid)); + } else if (ret == 1) { + /* BD_XATTR size and LV size mismatch. Update BD_XATTR */ + gf_asprintf (&bd, "%s:%ld", *type, *size); + + dict = dict_new (); + BD_VALIDATE_MEM_ALLOC (dict, ret, out); + + ret = dict_set_dynstr (dict, BD_XATTR, bd); + if (ret) + goto out; + + STACK_WIND (bd_frame, bd_null_rmsetxattr_cbk, FIRST_CHILD (this), + FIRST_CHILD (this)->fops->setxattr, &loc, dict, 0, + NULL); + } + +out: + dict_del (xattr, BD_XATTR); + GF_FREE (bd_xattr); + GF_FREE (bd); + return ret; +} + +/* + * bd_lookup_cbk: Call back from posix_lookup. + */ +int32_t +bd_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, + int op_errno, inode_t *inode, struct iatt *buf, dict_t *xattr, + struct iatt *postparent) +{ + int ret = -1; + bd_attr_t *bdatt = NULL; + uint64_t size = 0; + char *type = BD_TYPE_NONE; + + /* only regular files are part of BD object */ + if (op_ret < 0 || buf->ia_type != IA_IFREG) + goto out; + + /* iatt already cached */ + if (!bd_inode_ctx_get (inode, this, &bdatt)) + goto next; + + if (bd_get_bd_info (frame, this, xattr, buf->ia_gfid, &type, &size)) + goto out; + + /* BD file, update buf */ + bdatt = GF_CALLOC (1, sizeof (bd_attr_t), gf_bd_attr); + if (!bdatt) { + op_errno = ENOMEM; + goto out; + } + memcpy (&bdatt->iatt, buf, sizeof (struct iatt)); + bdatt->type = type; + + /* Cache LV size in inode_ctx */ + ret = bd_inode_ctx_set (inode, this, bdatt); + if (ret < 0) { + GF_FREE (bdatt); + op_errno = EINVAL; + goto out; + } + + bdatt->iatt.ia_size = size; + bdatt->iatt.ia_blocks = size / 512; + +next: + dict_del (xattr, GF_CONTENT_KEY); + memcpy (buf, &bdatt->iatt, sizeof (struct iatt)); + +out: + BD_STACK_UNWIND (lookup, frame, op_ret, op_errno, inode, buf, + xattr, postparent); + return 0; +} + +/* + * bd_lookup: Issues posix_lookup to find out if file is mapped to BD + * bd_lookup -> posix_lookup -> bd_lookup_cbk +*/ +int32_t +bd_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req) +{ + dict_t *bd_xattr = NULL; + bd_attr_t *bdatt = NULL; + int op_errno = EINVAL; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (loc, out); + VALIDATE_OR_GOTO (loc->path, out); + VALIDATE_OR_GOTO (this->private, out); + + if (bd_inode_ctx_get (loc->inode, this, &bdatt) < 0) { + if (!xattr_req) { + bd_xattr = dict_new (); + BD_VALIDATE_MEM_ALLOC (bd_xattr, op_errno, out); + xattr_req = bd_xattr; + } + if (dict_set_int8 (xattr_req, BD_XATTR, 1) < 0) + goto out; + } + + STACK_WIND (frame, bd_lookup_cbk, FIRST_CHILD (this), + FIRST_CHILD (this)->fops->lookup, loc, xattr_req); + + if (bd_xattr) + dict_unref (bd_xattr); + return 0; +out: + BD_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL); + + return 0; +} + +int +bd_forget (xlator_t *this, inode_t *inode) +{ + int ret = -1; + uint64_t ctx = 0; + bd_attr_t *bdatt = NULL; + + ret = bd_inode_ctx_get (inode, this, &bdatt); + if (!ret) { + inode_ctx_del (inode, this, &ctx); + FREE (bdatt); + } + return 0; +} + +int +bd_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, + int op_errno, gf_dirent_t *entries, dict_t *xdata) +{ + gf_dirent_t *entry = NULL; + uint64_t size = 0; + char *type = NULL; + + if (op_ret < 0) + goto out; + + list_for_each_entry (entry, &entries->list, list) { + if (entry->d_type != DT_REG) + continue; + if (!bd_get_bd_info (frame, this, entry->dict, + entry->d_stat.ia_gfid, &type, &size)) { + entry->d_stat.ia_size = size; + entry->d_stat.ia_blocks = size / 512; + FREE (type); + } + } + +out: + BD_STACK_UNWIND (readdirp, frame, op_ret, op_errno, entries, xdata); + return 0; +} + +/* + * bd_readdirp: In bd_readdirp_cbk if the file and BD_XATTR_SIZE is set + * ia_size is updated with the LV(BD_XATTR_SIZE) size + */ +int32_t +bd_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size, + off_t off, dict_t *dict) +{ + int op_errno = EINVAL; + bd_local_t *local = NULL; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (fd, out); + VALIDATE_OR_GOTO (this->private, out); + + if (!dict) { + local = bd_local_init (frame, this); + BD_VALIDATE_MEM_ALLOC (local, op_errno, out); + local->dict = dict_new (); + BD_VALIDATE_MEM_ALLOC (local->dict, op_errno, out); + dict = local->dict; + } + + if (dict_set_int8 (dict, BD_XATTR, 0)) { + gf_log (this->name, GF_LOG_WARNING, + "failed to set key %s", BD_XATTR); + goto out; + } + + STACK_WIND (frame, bd_readdirp_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->readdirp, fd, size, off, dict); + + return 0; +out: + BD_STACK_UNWIND (readdirp, frame, -1, op_errno, NULL, dict); + return 0; +} + +int +bd_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, + int op_errno, struct iatt *buf, dict_t *xdata) +{ + bd_local_t *local = frame->local; + bd_attr_t *bdatt = NULL; + + /* only regular files are part of BD object */ + if (op_ret < 0 || buf->ia_type != IA_IFREG) + goto out; + + BD_VALIDATE_LOCAL_OR_GOTO (local, op_errno, out); + + /* update buf with LV size */ + if (!bd_inode_ctx_get (local->inode, this, &bdatt)) + memcpy (buf, bdatt, sizeof (struct iatt)); + +out: + BD_STACK_UNWIND (stat, frame, op_ret, op_errno, buf, xdata); + return 0; +} + +int +bd_stat (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata) +{ + int op_errno = EINVAL; + bd_local_t *local = NULL; + bd_attr_t *bdatt = NULL; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (loc, out); + VALIDATE_OR_GOTO (loc->path, out); + VALIDATE_OR_GOTO (this->private, out); + + if (!bd_inode_ctx_get (loc->inode, this, &bdatt)) { + BD_STACK_UNWIND (stat, frame, 0, 0, &bdatt->iatt, xdata); + return 0; + } + + local = bd_local_init (frame, this); + BD_VALIDATE_MEM_ALLOC (local, op_errno, out); + local->inode = inode_ref (loc->inode); + + STACK_WIND(frame, bd_stat_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->stat, loc, xdata); + return 0; +out: + BD_STACK_UNWIND (stat, frame, -1, op_errno, NULL, xdata); + return 0; +} + +int +bd_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, + int op_errno, struct statvfs *buff, dict_t *xdata) +{ + uint64_t size = 0; + uint64_t fr_size = 0; + bd_priv_t *priv = NULL; + vg_t vg = NULL; + + if (op_ret < 0) + goto out; + + priv = this->private; + + vg = lvm_vg_open (priv->handle, priv->vg, "r", 0); + if (!vg) { + gf_log (this->name, GF_LOG_WARNING, "opening VG %s failed", + priv->vg); + op_ret = -1; + op_errno = EAGAIN; + goto out; + } + size = lvm_vg_get_size (vg); + fr_size = lvm_vg_get_free_size (vg); + lvm_vg_close (vg); + + buff->f_blocks += size / buff->f_frsize; + buff->f_bfree += fr_size / buff->f_frsize; + buff->f_bavail += fr_size / buff->f_frsize; + +out: + BD_STACK_UNWIND (statfs, frame, op_ret, op_errno, buff, xdata); + return 0; +} + +/* + * bd_statfs: Mimics statfs by returning used/free extents in the VG + */ +int +bd_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata) +{ + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (this->private, out); + VALIDATE_OR_GOTO (loc, out); + + STACK_WIND (frame, bd_statfs_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->statfs, loc, xdata); + return 0; +out: + BD_STACK_UNWIND (statfs, frame, -1, EINVAL, NULL, NULL); + return 0; +} + +int +bd_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, + int op_errno, struct iatt *buf, dict_t *xdata) +{ + bd_attr_t *bdatt = NULL; + bd_local_t *local = frame->local; + + /* only regular files are part of BD object */ + if (op_ret < 0 || buf->ia_type != IA_IFREG) + goto out; + + BD_VALIDATE_LOCAL_OR_GOTO (local, op_errno, out); + + /* update buf with LV size */ + if (!bd_inode_ctx_get (local->inode, this, &bdatt)) + memcpy (buf, &bdatt->iatt, sizeof (struct iatt)); + +out: + BD_STACK_UNWIND (fstat, frame, op_ret, op_errno, buf, xdata); + return 0; +} + +int +bd_fstat (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata) +{ + int op_errno = EINVAL; + bd_local_t *local = NULL; + bd_attr_t *bdatt = NULL; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (fd, out); + VALIDATE_OR_GOTO (this->private, out); + + /* if its already cached return it */ + if (!bd_inode_ctx_get (fd->inode, this, &bdatt)) { + BD_STACK_UNWIND (fstat, frame, 0, 0, &bdatt->iatt, xdata); + return 0; + } + + local = bd_local_init (frame, this); + BD_VALIDATE_MEM_ALLOC (local, op_errno, out); + + local->inode = inode_ref (fd->inode); + + STACK_WIND (frame, bd_fstat_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->fstat, fd, xdata); + + return 0; +out: + BD_STACK_UNWIND (fstat, frame, -1, op_errno, NULL, xdata); + return 0; +} + +/* + * bd_readv: If posix file, invokes posix_readv otherwise reads from the BD + * file + */ +int +bd_readv (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size, + off_t offset, uint32_t flags, dict_t *xdata) +{ + int ret = -1; + int _fd = -1; + int32_t op_ret = -1; + int32_t op_errno = 0; + bd_fd_t *bd_fd = NULL; + struct iovec vec = {0, }; + struct iobuf *iobuf = NULL; + struct iobref *iobref = NULL; + uint64_t bd_size = 0; + bd_attr_t *bdatt = NULL; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (fd, out); + VALIDATE_OR_GOTO (this->private, out); + + ret = bd_fd_ctx_get (this, fd, &bd_fd); + if (ret < 0 || !bd_fd) { + STACK_WIND (frame, default_readv_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->readv, + fd, size, offset, flags, xdata); + return 0; + } + if (!size) { + op_errno = EINVAL; + gf_log (this->name, GF_LOG_WARNING, "size=%"GF_PRI_SIZET, size); + goto out; + } + iobuf = iobuf_get2 (this->ctx->iobuf_pool, size); + if (!iobuf) { + op_errno = ENOMEM; + goto out; + } + _fd = bd_fd->fd; + op_ret = pread (_fd, iobuf->ptr, size, offset); + if (op_ret == -1) { + op_errno = errno; + gf_log (this->name, GF_LOG_ERROR, + "read failed on fd=%p: %s", fd, + strerror (op_errno)); + goto out; + } + + vec.iov_base = iobuf->ptr; + vec.iov_len = op_ret; + + iobref = iobref_new (); + iobref_add (iobref, iobuf); + + if (bd_inode_ctx_get (fd->inode, this, &bdatt)) { + op_errno = EINVAL; + op_ret = -1; + goto out; + } + bd_size = bdatt->iatt.ia_size; + if (!bd_size || (offset + vec.iov_len) >= bd_size) + op_errno = ENOENT; + + op_ret = vec.iov_len; + bd_update_amtime (&bdatt->iatt, GF_SET_ATTR_ATIME); + +out: + BD_STACK_UNWIND (readv, frame, op_ret, op_errno, + &vec, 1, &bdatt->iatt, iobref, NULL); + + if (iobref) + iobref_unref (iobref); + if (iobuf) + iobuf_unref (iobuf); + + return 0; +} + +#ifdef BLKDISCARD +/* + * bd_discard: Sends BLKDISCARD ioctl to the block device + */ +int +bd_discard (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, + size_t len, dict_t *xdata) +{ + int ret = -1; + int op_errno = EINVAL; + bd_fd_t *bd_fd = NULL; + uint64_t param[2] = {0, }; + bd_attr_t *bdatt = NULL; + struct iatt prebuf = {0, }; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (this->private, out); + VALIDATE_OR_GOTO (fd, out); + + /* posix */ + if (bd_inode_ctx_get (fd->inode, this, &bdatt)) { + STACK_WIND (frame, default_discard_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->discard, + fd, offset, len, xdata); + return 0; + } + + ret = bd_fd_ctx_get (this, fd, &bd_fd); + if (ret < 0 || !bd_fd) { + op_errno = EINVAL; + goto out; + } + + param[0] = offset; + param[1] = len; + ret = ioctl (bd_fd->fd, BLKDISCARD, param); + if (ret < 0) { + if (errno == ENOTTY) + op_errno = ENOSYS; + else + op_errno = errno; + goto out; + } + memcpy (&prebuf, &bdatt->iatt, sizeof (prebuf)); + bd_update_amtime (&bdatt->iatt, GF_SET_ATTR_MTIME); + + BD_STACK_UNWIND (discard, frame, ret, op_errno, &prebuf, + &bdatt->iatt, xdata); + return 0; + +out: + BD_STACK_UNWIND (discard, frame, -1, op_errno, NULL, NULL, NULL); + return 0; +} +#else + +int +bd_discard (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, + size_t len, dict_t *xdata) +{ + BD_STACK_UNWIND (discard, frame, -1, ENOSYS, NULL, NULL, NULL); + return 0; +} +#endif + +/* + * Call back from posix_open for opening the backing posix file + * If it failed, close BD fd + */ +int +bd_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, fd_t *fd, dict_t *xdata) +{ + bd_fd_t *bd_fd = NULL; + bd_attr_t *bdatt = NULL; + + if (!op_ret) + goto out; + + bd_inode_ctx_get (fd->inode, this, &bdatt); + if (!bdatt) /* posix file */ + goto out; + + /* posix open failed */ + if (bd_fd_ctx_get (this, fd, &bd_fd) < 0) { + gf_log (this->name, GF_LOG_WARNING, + "bd_fd is NULL from fd=%p", fd); + goto out; + } + close (bd_fd->fd); + GF_FREE (bd_fd); + +out: + BD_STACK_UNWIND (open, frame, op_ret, op_errno, fd, NULL); + + return 0; +} + +/* + * bd_open: Opens BD file if given posix file is mapped to BD. Also opens + * posix file. + * fd contains both posix and BD fd + */ +int32_t +bd_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, + fd_t *fd, dict_t *xdata) +{ + int32_t ret = EINVAL; + bd_fd_t *bd_fd = NULL; + bd_attr_t *bdatt = NULL; + bd_gfid_t gfid = {0, }; + char *devpath = NULL; + bd_priv_t *priv = this->private; + int _fd = -1; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (this->private, out); + VALIDATE_OR_GOTO (loc, out); + VALIDATE_OR_GOTO (fd, out); + + /* not bd file */ + if (fd->inode->ia_type != IA_IFREG || + bd_inode_ctx_get (fd->inode, this, &bdatt)) + goto posix; + + uuid_utoa_r (fd->inode->gfid, gfid); + asprintf (&devpath, "/dev/%s/%s", priv->vg, gfid); + BD_VALIDATE_MEM_ALLOC (devpath, ret, out); + + _fd = open (devpath, flags | O_LARGEFILE, 0); + if (_fd < 0) { + ret = errno; + gf_log (this->name, GF_LOG_ERROR, "open on %s: %s", devpath, + strerror (ret)); + goto out; + } + bd_fd = GF_CALLOC (1, sizeof(bd_fd_t), gf_bd_fd); + BD_VALIDATE_MEM_ALLOC (bd_fd, ret, out); + + bd_fd->fd = _fd; + bd_fd->flag = flags | O_LARGEFILE; + + if (fd_ctx_set (fd, this, (uint64_t)(long)bd_fd) < 0) { + gf_log (this->name, GF_LOG_WARNING, + "failed to set the fd context fd=%p", fd); + goto out; + } + + ret = 0; + +posix: + + /* open posix equivalant of this file, fd needed for fd related + operations like fsetxattr, ftruncate etc */ + STACK_WIND (frame, bd_open_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->open, loc, flags, fd, xdata); + + return 0; +out: + BD_STACK_UNWIND (open, frame, -1, ret, fd, NULL); + + FREE (devpath); + if (ret) { + close (_fd); + GF_FREE (bd_fd); + } + + return 0; +} + +/* + * call back from posix_setattr after updating iatt to posix file. + */ +int +bd_fsync_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, struct iatt *pre, + struct iatt *post, dict_t *xdata) +{ + bd_local_t *local = frame->local; + bd_attr_t *bdatt = local->bdatt; + + BD_STACK_UNWIND (fsync, frame, op_ret, op_errno, &bdatt->iatt, + &bdatt->iatt, NULL); + return 0; +} + +int +bd_do_fsync (int fd, int datasync) +{ + int op_errno = 0; + +#ifdef HAVE_FDATASYNC + if (datasync) { + if (fdatasync (fd)) { + op_errno = errno; + gf_log (THIS->name, GF_LOG_ERROR, + "fdatasync on fd=%d failed: %s", + fd, strerror (errno)); + } + + } else +#endif + { + if (fsync (fd)) { + op_errno = errno; + gf_log (THIS->name, GF_LOG_ERROR, + "fsync on fd=%d failed: %s", + fd, strerror (op_errno)); + } + } + + return op_errno; +} + +/* + * bd_fsync: Syncs if BD fd, forwards the request to posix + * fsync -> posix_setattr -> posix_fsync +*/ +int32_t +bd_fsync (call_frame_t *frame, xlator_t *this, + fd_t *fd, int32_t datasync, dict_t *xdata) +{ + int ret = -1; + int32_t op_ret = -1; + int32_t op_errno = 0; + bd_fd_t *bd_fd = NULL; + bd_priv_t *priv = NULL; + bd_attr_t *bdatt = NULL; + bd_local_t *local = NULL; + int valid = GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME; + struct iatt prebuf = {0, }; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (fd, out); + VALIDATE_OR_GOTO (this->private, out); + + priv = this->private; + + ret = bd_inode_ctx_get (fd->inode, this, &bdatt); + ret = bd_fd_ctx_get (this, fd, &bd_fd); + if (ret < 0 || !bd_fd || !bdatt) { + STACK_WIND (frame, default_fsync_cbk, FIRST_CHILD (this), + FIRST_CHILD (this)->fops->fsync, fd, datasync, + xdata); + return 0; + } + + memcpy (&prebuf, &bdatt->iatt, sizeof (struct iatt)); + + op_errno = bd_do_fsync (bd_fd->fd, datasync); + if (op_errno) + goto out; + + /* For BD, Update the a|mtime during full fsync only */ + if (!datasync) { + local = bd_local_init (frame, this); + /* In case of mem failure, should posix flush called ? */ + BD_VALIDATE_MEM_ALLOC (local, op_errno, out); + + local->bdatt = GF_CALLOC (1, sizeof (bd_attr_t), gf_bd_attr); + BD_VALIDATE_MEM_ALLOC (local->bdatt, op_errno, out); + + local->bdatt->type = gf_strdup (bdatt->type); + memcpy (&local->bdatt->iatt, &bdatt->iatt, sizeof (struct iatt)); + bd_update_amtime (&local->bdatt->iatt, valid); + uuid_copy (local->loc.gfid, fd->inode->gfid); + STACK_WIND (frame, bd_fsync_setattr_cbk, FIRST_CHILD (this), + FIRST_CHILD (this)->fops->setattr, &local->loc, + &local->bdatt->iatt, + valid, NULL); + return 0; + } + +out: + BD_STACK_UNWIND (fsync, frame, op_ret, op_errno, &prebuf, + &bdatt->iatt, NULL); + return 0; +} + +int +bd_flush_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, struct iatt *pre, + struct iatt *post, dict_t *xdata) +{ + BD_STACK_UNWIND (flush, frame, op_ret, op_errno, xdata); + return 0; +} + +int +bd_flush (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata) +{ + int ret = -1; + bd_fd_t *bd_fd = NULL; + bd_priv_t *priv = NULL; + bd_attr_t *bdatt = NULL; + int valid = GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME; + bd_local_t *local = NULL; + int op_errno = EINVAL; + loc_t loc = {0, }; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (fd, out); + VALIDATE_OR_GOTO (this->private, out); + + priv = this->private; + + ret = bd_inode_ctx_get (fd->inode, this, &bdatt); + if (!bdatt) + goto out; + + ret = bd_fd_ctx_get (this, fd, &bd_fd); + if (ret < 0 || !bd_fd || !bdatt) { + gf_log (this->name, GF_LOG_WARNING, + "bdfd/bdatt is NULL from fd=%p", fd); + goto out; + } + + local = bd_local_init (frame, this); + BD_VALIDATE_MEM_ALLOC (local, op_errno, out); + + local->fd = fd_ref (fd); + uuid_copy (loc.gfid, bdatt->iatt.ia_gfid); + + /* Update the a|mtime during flush */ + STACK_WIND (frame, bd_flush_setattr_cbk, FIRST_CHILD (this), + FIRST_CHILD (this)->fops->setattr, &loc, &bdatt->iatt, + valid, NULL); + + return 0; + +out: + STACK_WIND (frame, default_flush_cbk, FIRST_CHILD (this), + FIRST_CHILD (this)->fops->flush, fd, xdata); + + return 0; +} + +int32_t +bd_release (xlator_t *this, fd_t *fd) +{ + int ret = -1; + bd_fd_t *bd_fd = NULL; + uint64_t tmp_bfd = 0; + bd_attr_t *bdatt = NULL; + bd_priv_t *priv = this->private; + + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (fd, out); + VALIDATE_OR_GOTO (priv, out); + + ret = bd_inode_ctx_get (fd->inode, this, &bdatt); + if (ret || !bdatt) /* posix file */ + goto out; + + /* FIXME: Update amtime during release */ + + ret = fd_ctx_del (fd, this, &tmp_bfd); + if (ret < 0) { + gf_log (this->name, GF_LOG_WARNING, + "bfd is NULL from fd=%p", fd); + goto out; + } + bd_fd = (bd_fd_t *)(long)tmp_bfd; + + close (bd_fd->fd); + GF_FREE (bd_fd); +out: + return 0; +} + +/* + * Call back for removexattr after removing BD_XATTR incase of + * bd create failure + */ +int +bd_setx_rm_xattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, dict_t *xdata) +{ + bd_local_t *local = frame->local; + + if (local->fd) + BD_STACK_UNWIND (setxattr, frame, -1, EIO, xdata); + else + BD_STACK_UNWIND (setxattr, frame, -1, EIO, xdata); + return 0; + +} + +/* + * Call back after setting BD_XATTR. Creates BD. If BD creation is a failure + * invokes posix_removexattr to remove created BD_XATTR + */ +int +bd_setx_setx_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, dict_t *xdata) +{ + bd_local_t *local = frame->local; + bd_attr_t *bdatt = NULL; + + if (op_ret < 0) + goto next; + + /* Create LV */ + op_errno = bd_create (local->inode->gfid, local->bdatt->iatt.ia_size, + local->bdatt->type, this->private); + if (!op_errno) + goto out; + + /* LV creation failed, remove BD_XATTR */ + if (local->fd) + STACK_WIND (frame, bd_setx_rm_xattr_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->fremovexattr, + local->fd, BD_XATTR, NULL); + else + STACK_WIND (frame, bd_setx_rm_xattr_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->removexattr, + &local->loc, BD_XATTR, NULL); + + return 0; +out: + + bdatt = GF_CALLOC (1, sizeof (bd_attr_t), gf_bd_attr); + if (!bdatt) { + op_ret = -1; + op_errno = ENOMEM; + goto next; + } + + memcpy (&bdatt->iatt, &local->bdatt->iatt, sizeof (struct iatt)); + bdatt->type = gf_strdup (local->bdatt->type); + + bd_inode_ctx_set (local->inode, THIS, bdatt); + +next: + if (local->fd) + BD_STACK_UNWIND (setxattr, frame, op_ret, op_errno, xdata); + else + BD_STACK_UNWIND (setxattr, frame, op_ret, op_errno, xdata); + return 0; + +} + +/* + * Call back from posix_stat + */ +int +bd_setx_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, struct iatt *iatt, + dict_t *xdata) +{ + char *param = NULL; + char *type = NULL; + char *s_size = NULL; + char *p = NULL; + char *copy = NULL; + bd_local_t *local = frame->local; + bd_priv_t *priv = this->private; + char *bd = NULL; + uint64_t size = 0; + + if (op_ret < 0) + goto out; + + if (!IA_ISREG (iatt->ia_type)) { + op_errno = EOPNOTSUPP; + goto out; + } + + param = copy = GF_CALLOC (1, local->data->len + 1, gf_common_mt_char); + BD_VALIDATE_MEM_ALLOC (param, op_errno, out); + + strncpy (param, local->data->data, local->data->len); + + type = strtok_r (param, ":", &p); + if (!type) { + op_errno = EINVAL; + goto out; + } + + if (strcmp (type, BD_LV) && strcmp (type, BD_THIN)) { + gf_log (this->name, GF_LOG_WARNING, "Invalid bd type %s given", + type); + op_errno = EINVAL; + goto out; + } + + s_size = strtok_r (NULL, ":", &p); + + /* If size not specified get default size */ + if (!s_size) + size = bd_get_default_extent (priv); + else + gf_string2bytesize (s_size, &size); + + gf_asprintf (&bd, "%s:%ld", type, size); + BD_VALIDATE_MEM_ALLOC (bd, op_errno, out); + + local->dict = dict_new (); + BD_VALIDATE_MEM_ALLOC (local->dict, op_errno, out); + + local->bdatt = GF_CALLOC (1, sizeof (bd_attr_t), gf_bd_attr); + BD_VALIDATE_MEM_ALLOC (local->bdatt, op_errno, out); + + if (dict_set_dynstr (local->dict, BD_XATTR, bd) < 0) { + op_errno = EINVAL; + goto out; + } + + local->bdatt->type = gf_strdup (type); + memcpy (&local->bdatt->iatt, iatt, sizeof (struct iatt)); + local->bdatt->iatt.ia_size = size; + + if (local->fd) + STACK_WIND (frame, bd_setx_setx_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->fsetxattr, + local->fd, local->dict, 0, NULL); + else + STACK_WIND (frame, bd_setx_setx_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->setxattr, + &local->loc, local->dict, 0, NULL); + + return 0; + +out: + if (local->fd) + BD_STACK_UNWIND (fsetxattr, frame, -1, op_errno, xdata); + else + BD_STACK_UNWIND (setxattr, frame, -1, op_errno, xdata); + + GF_FREE (bd); + GF_FREE (copy); + return 0; +} + +int +bd_offload_rm_xattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, dict_t *xdata) +{ + bd_local_t *local = frame->local; + + if (local->fd) + BD_STACK_UNWIND (fsetxattr, frame, -1, EIO, NULL); + else + BD_STACK_UNWIND (setxattr, frame, -1, EIO, NULL); + + return 0; +} + +int +bd_offload_setx_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, dict_t *xdata) +{ + bd_local_t *local = frame->local; + + if (op_ret < 0) + goto out; + + if (local->offload == BD_OF_SNAPSHOT) + op_ret = bd_snapshot_create (frame->local, this->private); + else + op_ret = bd_clone (frame->local, this->private); + + if (op_ret) { + STACK_WIND (frame, bd_offload_rm_xattr_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->removexattr, + local->dloc, BD_XATTR, NULL); + return 0; + } + +out: + if (local->fd) + BD_STACK_UNWIND (fsetxattr, frame, op_ret, op_errno, NULL); + else + BD_STACK_UNWIND (setxattr, frame, op_errno, op_errno, NULL); + + return 0; +} + +int +bd_offload_getx_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, dict_t *xattr, dict_t *xdata) +{ + char *bd = NULL; + bd_local_t *local = frame->local; + char *type = NULL; + char *p = NULL; + + if (op_ret < 0) + goto out; + + if (dict_get_str (xattr, BD_XATTR, &p)) { + op_errno = EINVAL; + goto out; + } + + type = gf_strdup (p); + BD_VALIDATE_MEM_ALLOC (type, op_errno, out); + + p = strrchr (type, ':'); + if (!p) { + op_errno = EINVAL; + gf_log (this->name, GF_LOG_WARNING, + "source file xattr %s corrupted?", type); + goto out; + } + + *p='\0'; + + /* For clone size is taken from source LV */ + if (!local->size) { + p++; + gf_string2bytesize (p, &local->size); + } + gf_asprintf (&bd, "%s:%ld", type, local->size); + local->bdatt->type = gf_strdup (type); + dict_del (local->dict, BD_XATTR); + dict_del (local->dict, LINKTO); + if (dict_set_dynstr (local->dict, BD_XATTR, bd)) { + op_errno = EINVAL; + goto out; + } + + STACK_WIND (frame, bd_offload_setx_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->setxattr, + local->dloc, local->dict, 0, NULL); + + return 0; + +out: + if (local->fd) + BD_STACK_UNWIND (fsetxattr, frame, -1, op_errno, NULL); + else + BD_STACK_UNWIND (setxattr, frame, -1, op_errno, NULL); + + GF_FREE (type); + GF_FREE (bd); + + return 0; +} + +int +bd_offload_dest_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, + inode_t *inode, struct iatt *iatt, + dict_t *xattr, struct iatt *postparent) +{ + bd_local_t *local = frame->local; + char *bd = NULL; + int ret = -1; + char *linkto = NULL; + + if (op_ret < 0 && op_errno != ENODATA) { + op_errno = EINVAL; + goto out; + } + + if (!IA_ISREG (iatt->ia_type)) { + op_errno = EINVAL; + gf_log (this->name, GF_LOG_WARNING, "destination gfid is not a " + "regular file"); + goto out; + } + + ret = dict_get_str (xattr, LINKTO, &linkto); + if (linkto) { + op_errno = EINVAL; + gf_log (this->name, GF_LOG_WARNING, "destination file not " + "present in same brick"); + goto out; + } + + ret = dict_get_str (xattr, BD_XATTR, &bd); + if (bd) { + op_errno = EEXIST; + goto out; + } + + local->bdatt = CALLOC (1, sizeof (bd_attr_t)); + BD_VALIDATE_MEM_ALLOC (local->bdatt, op_errno, out); + + STACK_WIND (frame, bd_offload_getx_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->getxattr, + &local->loc, BD_XATTR, NULL); + + return 0; +out: + if (local->fd) + BD_STACK_UNWIND (fsetxattr, frame, -1, op_errno, NULL); + else + BD_STACK_UNWIND (setxattr, frame, -1, op_errno, NULL); + + return 0; +} + +int +bd_merge_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, struct iatt *preparent, + struct iatt *postparent, dict_t *xdata) +{ + /* FIXME: if delete failed, remove xattr */ + + BD_STACK_UNWIND (setxattr, frame, op_ret, op_errno, NULL); + return 0; +} + +int +bd_do_merge(call_frame_t *frame, xlator_t *this) +{ + bd_local_t *local = frame->local; + inode_t *parent = NULL; + char *p = NULL; + int op_errno = 0; + + op_errno = bd_merge (this->private, local->inode->gfid); + if (op_errno) + goto out; + + /* + * posix_unlink needs loc->pargfid to be valid, but setxattr FOP does + * not have loc->pargfid set. Get parent's gfid by getting parents inode + */ + parent = inode_parent (local->inode, NULL, NULL); + if (!parent) { + /* + * FIXME: Snapshot LV already deleted. + * remove xattr, instead of returning failure + */ + op_errno = EINVAL; + goto out; + } + uuid_copy (local->loc.pargfid, parent->gfid); + + p = strrchr (local->loc.path, '/'); + if (p) + p++; + local->loc.name = p; + + STACK_WIND (frame, bd_merge_unlink_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->unlink, + &local->loc, 0, NULL); + + return 0; +out: + BD_STACK_UNWIND (fsetxattr, frame, -1, op_errno, NULL); + + return op_errno; +} + +int +bd_offload (call_frame_t *frame, xlator_t *this, loc_t *loc, + fd_t *fd, bd_offload_t offload) +{ + char *param = NULL; + char *param_copy = NULL; + char *p = NULL; + char *size = NULL; + char *gfid = NULL; + int op_errno = 0; + bd_local_t *local = frame->local; + + param = GF_CALLOC (1, local->data->len + 1, gf_common_mt_char); + BD_VALIDATE_MEM_ALLOC (param, op_errno, out); + param_copy = param; + + local->dict = dict_new (); + BD_VALIDATE_MEM_ALLOC (local->dict, op_errno, out); + + local->dloc = CALLOC (1, sizeof (loc_t)); + BD_VALIDATE_MEM_ALLOC (local->dloc, op_errno, out); + + strncpy (param, local->data->data, local->data->len); + + gfid = strtok_r (param, ":", &p); + size = strtok_r (NULL, ":", &p); + if (size) + gf_string2bytesize (size, &local->size); + else if (offload != BD_OF_CLONE) + local->size = bd_get_default_extent (this->private); + + if (dict_set_int8 (local->dict, BD_XATTR, 1) < 0) { + op_errno = EINVAL; + goto out; + } + if (dict_set_int8 (local->dict, LINKTO, 1) < 0) { + op_errno = EINVAL; + goto out; + } + + uuid_parse (gfid, local->dloc->gfid); + local->offload = offload; + + STACK_WIND (frame, bd_offload_dest_lookup_cbk, FIRST_CHILD (this), + FIRST_CHILD (this)->fops->lookup, local->dloc, + local->dict); + + return 0; + +out: + if (fd) + BD_STACK_UNWIND (fsetxattr, frame, -1, op_errno, NULL); + else + BD_STACK_UNWIND (setxattr, frame, -1, op_errno, NULL); + + GF_FREE (param_copy); + return 0; +} + +/* + * bd_setxattr: Used to create & map an LV to a posix file using + * BD_XATTR xattr + * bd_setxattr -> posix_stat -> bd_setx_stat_cbk -> posix_setxattr -> + * bd_setx_setx_cbk -> create_lv + * if create_lv failed, posix_removexattr -> bd_setx_rm_xattr_cbk + */ +int +bd_setxattr (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict, + int flags, dict_t *xdata) +{ + int op_errno = 0; + data_t *data = NULL; + bd_local_t *local = NULL; + bd_attr_t *bdatt = NULL; + bd_offload_t cl_type = BD_OF_NONE; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + + if ((data = dict_get (dict, BD_XATTR))) + cl_type = BD_OF_NONE; + else if ((data = dict_get (dict, BD_CLONE))) + cl_type = BD_OF_CLONE; + else if ((data = dict_get (dict, BD_SNAPSHOT))) + cl_type = BD_OF_SNAPSHOT; + else if ((data = dict_get (dict, BD_MERGE))) + cl_type = BD_OF_MERGE; + + bd_inode_ctx_get (loc->inode, this, &bdatt); + if (!cl_type && !data) { + STACK_WIND (frame, default_setxattr_cbk, FIRST_CHILD (this), + FIRST_CHILD (this)->fops->setxattr, loc, dict, + flags, xdata); + return 0; + } + + local = bd_local_init (frame, this); + BD_VALIDATE_MEM_ALLOC (local, op_errno, out); + + local->data = data; + loc_copy (&local->loc, loc); + local->inode = inode_ref (loc->inode); + + if (cl_type) { + /* For cloning/snapshot, source file must be mapped to LV */ + if (!bdatt) { + gf_log (this->name, GF_LOG_WARNING, + "%s not mapped to BD", loc->path); + op_errno = EINVAL; + goto out; + } + if (cl_type == BD_OF_MERGE) + bd_do_merge (frame, this); + else + bd_offload (frame, this, loc, NULL, cl_type); + } else if (data) { + if (bdatt) { + gf_log (this->name, GF_LOG_WARNING, + "%s already mapped to BD", loc->path); + op_errno = EEXIST; + goto out; + } + STACK_WIND (frame, bd_setx_stat_cbk, FIRST_CHILD (this), + FIRST_CHILD (this)->fops->stat, loc, xdata); + } + + return 0; +out: + if (op_errno) + STACK_UNWIND_STRICT (setxattr, frame, -1, op_errno, xdata); + + return 0; +} + +/* + * bd_fsetxattr: Used to create/map an LV to a posix file using + * BD_XATTR xattr + * bd_fsetxattr -> posix_fstat -> bd_setx_stat_cbk -> posix_fsetxattr -> + * bd_setx_setx_cbk -> create_lv + * if create_lv failed, posix_removexattr -> bd_setx_rm_xattr_cbk + * -> bd_fsetxattr_cbk + */ +int32_t +bd_fsetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict, + int flags, dict_t *xdata) +{ + int op_errno = 0; + data_t *data = NULL; + bd_attr_t *bdatt = NULL; + bd_local_t *local = NULL; + bd_offload_t cl_type = BD_OF_NONE; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (this->private, out); + VALIDATE_OR_GOTO (fd, out); + + bd_inode_ctx_get (fd->inode, this, &bdatt); + + data = dict_get (dict, BD_XATTR); + if ((data = dict_get (dict, BD_XATTR))) + cl_type = BD_OF_NONE; + else if ((data = dict_get (dict, BD_CLONE))) + cl_type = BD_OF_CLONE; + else if ((data = dict_get (dict, BD_SNAPSHOT))) + cl_type = BD_OF_SNAPSHOT; + else if ((data = dict_get (dict, BD_MERGE))) { + /* + * bd_merge is not supported for fsetxattr, because snapshot LV + * is opened and it causes problem in snapshot merge + */ + op_errno = EOPNOTSUPP; + goto out; + } + + bd_inode_ctx_get (fd->inode, this, &bdatt); + + if (!cl_type && !data) { + /* non bd file object */ + STACK_WIND (frame, default_fsetxattr_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->fsetxattr, + fd, dict, flags, xdata); + return 0; + } + + local = bd_local_init (frame, this); + BD_VALIDATE_MEM_ALLOC (local, op_errno, out); + + local->inode = inode_ref (fd->inode); + local->fd = fd_ref (fd); + local->data = data; + + if (cl_type) { + /* For cloning/snapshot, source file must be mapped to LV */ + if (!bdatt) { + gf_log (this->name, GF_LOG_WARNING, + "fd %p not mapped to BD", fd); + op_errno = EINVAL; + goto out; + + } + bd_offload (frame, this, NULL, fd, cl_type); + } else if (data) { + if (bdatt) { + gf_log (this->name, GF_LOG_WARNING, + "fd %p already mapped to BD", fd); + op_errno = EEXIST; + goto out; + } + STACK_WIND(frame, bd_setx_stat_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->fstat, fd, xdata); + } + + return 0; +out: + + BD_STACK_UNWIND (setxattr, frame, -1, op_errno, NULL); + + return 0; +} + +int32_t +bd_removexattr (call_frame_t *frame, xlator_t *this, + loc_t *loc, const char *name, dict_t *xdata) +{ + if (!strcmp (name, BD_XATTR)) + goto out; + + STACK_WIND (frame, default_removexattr_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->removexattr, loc, name, xdata); + return 0; +out: + BD_STACK_UNWIND (removexattr, frame, -1, ENODATA, NULL); + return 0; +} + +int32_t +bd_fremovexattr (call_frame_t *frame, xlator_t *this, + fd_t *fd, const char *name, dict_t *xdata) +{ + if (!strcmp (name, BD_XATTR)) + goto out; + + STACK_WIND (frame, default_removexattr_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->fremovexattr, fd, name, xdata); + + return 0; +out: + BD_STACK_UNWIND (fremovexattr, frame, -1, ENODATA, NULL); + return 0; +} + +int +bd_trunc_setxattr_setx_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, dict_t *xdata) +{ + bd_local_t *local = frame->local; + + if (local->fd) + BD_STACK_UNWIND (ftruncate, frame, -1, EIO, NULL, NULL, NULL); + else + BD_STACK_UNWIND (truncate, frame, -1, EIO, NULL, NULL, NULL); + + return 0; +} + +/* + * Call back for setxattr after setting BD_XATTR_SIZE. + */ +int +bd_trunc_setxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, dict_t *xdata) +{ + bd_local_t *local = frame->local; + bd_attr_t *bdatt = NULL; + struct iatt prebuf = {0, }; + char *bd = NULL; + + if (op_ret < 0) + goto out; + + bd_inode_ctx_get (local->inode, this, &bdatt); + if (!bdatt) + goto revert_xattr; + + op_errno = bd_resize (this->private, local->inode->gfid, + local->bdatt->iatt.ia_size); + if (op_errno) + goto revert_xattr; + + memcpy (&prebuf, &bdatt->iatt, sizeof (struct iatt)); + /* LV resized, update new size in the cache */ + bdatt->iatt.ia_size = local->bdatt->iatt.ia_size; + + if (local->fd) + BD_STACK_UNWIND (ftruncate, frame, 0, 0, &prebuf, &bdatt->iatt, + NULL); + else + BD_STACK_UNWIND (truncate, frame, 0, 0, &prebuf, &bdatt->iatt, + NULL); + + return 0; + +revert_xattr: + /* revert setxattr */ + op_ret = dict_get_str (local->dict, BD_XATTR, &bd); + GF_FREE (bd); + gf_asprintf (&bd, "%s:%ld", bdatt->type, bdatt->iatt.ia_size); + + if (local->fd) + STACK_WIND (frame, bd_trunc_setxattr_setx_cbk, + FIRST_CHILD(this), + FIRST_CHILD(this)->fops->fsetxattr, + local->fd, local->dict, 0, NULL); + else + STACK_WIND (frame, bd_trunc_setxattr_setx_cbk, + FIRST_CHILD(this), + FIRST_CHILD(this)->fops->setxattr, + &local->loc, local->dict, 0, NULL); + + return 0; +out: + if (local->fd) + BD_STACK_UNWIND (ftruncate, frame, -1, EIO, NULL, NULL, NULL); + else + BD_STACK_UNWIND (truncate, frame, -1, EIO, NULL, NULL, NULL); + + return 0; +} + +/* + * call back from posix_[f]truncate_stat + * If offset > LV size, it resizes the LV and calls posix_setxattr + * to update new LV size in xattr else calls posix_setattr for updating + * the posix file so that truncate fop behaves properly + */ +int +bd_trunc_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, struct iatt *buf, dict_t *xdata) +{ + char *bd = NULL; + bd_local_t *local = frame->local; + bd_attr_t *bdatt = NULL; + + if (op_ret < 0) + goto out; + + local->dict = dict_new (); + BD_VALIDATE_MEM_ALLOC (local->dict, op_errno, out); + + bd_inode_ctx_get (local->inode, this, &bdatt); + if (!bdatt) { + op_errno = EINVAL; + goto out; + } + + gf_asprintf (&bd, "%s:%ld", bdatt->type, local->bdatt->iatt.ia_size); + if (dict_set_dynstr (local->dict, BD_XATTR, bd)) { + op_errno = EINVAL; + goto out; + } + + if (local->fd) + STACK_WIND (frame, bd_trunc_setxattr_cbk, + FIRST_CHILD(this), + FIRST_CHILD(this)->fops->fsetxattr, + local->fd, local->dict, 0, NULL); + else + STACK_WIND (frame, bd_trunc_setxattr_cbk, + FIRST_CHILD(this), + FIRST_CHILD(this)->fops->setxattr, + &local->loc, local->dict, 0, NULL); + + return 0; +out: + if (local->fd) + BD_STACK_UNWIND (ftruncate, frame, -1, op_errno, NULL, NULL, + NULL); + else + BD_STACK_UNWIND (truncate, frame, -1, op_errno, NULL, NULL, + NULL); + GF_FREE (bd); + return 0; +} + +void +bd_do_trunc (call_frame_t *frame, xlator_t *this, fd_t *fd, loc_t *loc, + off_t offset, bd_attr_t *bdatt) +{ + bd_local_t *local = NULL; + struct iatt prebuf = {0, }; + int op_errno = 0; + int op_ret = -1; + + /* If requested size is less than LV size, return success */ + if (offset <= bdatt->iatt.ia_size) { + memcpy (&prebuf, &bdatt->iatt, sizeof (struct iatt)); + bd_update_amtime (&bdatt->iatt, GF_SET_ATTR_MTIME); + op_ret = 0; + goto out; + } + + local = bd_local_init (frame, this); + BD_VALIDATE_MEM_ALLOC (local, op_errno, out); + + local->bdatt = GF_CALLOC (1, sizeof (bd_attr_t), gf_bd_attr); + BD_VALIDATE_MEM_ALLOC (local->bdatt, op_errno, out); + + if (fd) { + local->inode = inode_ref (fd->inode); + local->fd = fd_ref (fd); + } else { + local->inode = inode_ref (loc->inode); + loc_copy (&local->loc, loc); + } + + local->bdatt->iatt.ia_size = + bd_adjust_size (this->private, offset); + + STACK_WIND (frame, bd_trunc_stat_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->fstat, fd, NULL); + + return; + +out: + if (fd) + BD_STACK_UNWIND (ftruncate, frame, op_ret, op_errno, + &prebuf, &bdatt->iatt, NULL); + else + BD_STACK_UNWIND (truncate, frame, op_ret, op_errno, + &prebuf, &bdatt->iatt, NULL); + return; +} + +/* + * bd_ftruncate: Resizes a LV if fd belongs to BD. + */ +int32_t +bd_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, + dict_t *xdata) +{ + int op_errno = 0; + bd_attr_t *bdatt = NULL; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (fd, out); + + if (bd_inode_ctx_get (fd->inode, this, &bdatt)) { + STACK_WIND (frame, default_ftruncate_cbk, + FIRST_CHILD(this), + FIRST_CHILD(this)->fops->ftruncate, fd, + offset, xdata); + return 0; + } + + bd_do_trunc (frame, this, fd, NULL, offset, bdatt); + return 0; +out: + BD_STACK_UNWIND (ftruncate, frame, -1, op_errno, NULL, NULL, NULL); + return 0; +} + +/* + * bd_truncate: Resizes a LV if file maps to LV. + */ +int32_t +bd_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset, + dict_t *xdata) +{ + int op_errno = 0; + bd_attr_t *bdatt = NULL; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (loc, out); + + if (bd_inode_ctx_get (loc->inode, this, &bdatt)) { + STACK_WIND (frame, default_truncate_cbk, + FIRST_CHILD(this), + FIRST_CHILD(this)->fops->truncate, loc, + offset, xdata); + return 0; + } + + bd_do_trunc (frame, this, NULL, loc, offset, bdatt); + return 0; + +out: + BD_STACK_UNWIND (truncate, frame, -1, op_errno, NULL, NULL, NULL); + return 0; +} + +int32_t +__bd_pwritev (int fd, struct iovec *vector, int count, off_t offset, + uint64_t bd_size) +{ + int index = 0; + int retval = 0; + off_t internal_offset = 0; + + if (!vector) + return -EFAULT; + + retval = pwritev (fd, vector, count, offset); + if (retval == -1) { + gf_log (THIS->name, GF_LOG_WARNING, + "base %p, length %ld, offset %ld, message %s", + vector[index].iov_base, vector[index].iov_len, + internal_offset, strerror (errno)); + retval = -errno; + goto err; + } +/* + + + internal_offset = offset; + for (index = 0; index < count; index++) { + if (internal_offset > bd_size) { + op_ret = -ENOSPC; + goto err; + } + if (internal_offset + vector[index].iov_len > bd_size) { + vector[index].iov_len = bd_size - internal_offset; + no_space = 1; + } + retval = pwritev (fd, vector[index].iov_base, + vector[index].iov_len, internal_offset); + if (retval == -1) { + gf_log (THIS->name, GF_LOG_WARNING, + "base %p, length %ld, offset %ld, message %s", + vector[index].iov_base, vector[index].iov_len, + internal_offset, strerror (errno)); + op_ret = -errno; + goto err; + } + op_ret += retval; + internal_offset += retval; + if (no_space) + break; + } +*/ +err: + return retval; +} + +/* + * bd_writev: Writes to LV if its BD file or forwards the request to posix_write + * bd_writev -> posix_writev -> bd_writev_cbk + */ +int +bd_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector, + int32_t count, off_t offset, uint32_t flags, struct iobref *iobref, + dict_t *xdict) +{ + int32_t op_ret = -1; + int32_t op_errno = 0; + int _fd = -1; + bd_fd_t *bd_fd = NULL; + int ret = -1; + uint64_t size = 0; + struct iatt prebuf = {0, }; + bd_attr_t *bdatt = NULL; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (fd, out); + VALIDATE_OR_GOTO (vector, out); + + ret = bd_fd_ctx_get (this, fd, &bd_fd); + if (ret < 0 || !bd_fd) { /* posix fd */ + STACK_WIND (frame, default_writev_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->writev, fd, vector, count, + offset, flags, iobref, xdict); + return 0; + } + + _fd = bd_fd->fd; + + if (bd_inode_ctx_get (fd->inode, this, &bdatt)) { + op_ret = -1; + op_errno = EINVAL; + goto out; + } + size = bdatt->iatt.ia_size; + + op_ret = __bd_pwritev (_fd, vector, count, offset, size); + if (op_ret < 0) { + op_errno = -op_ret; + op_ret = -1; + gf_log (this->name, GF_LOG_ERROR, "write failed: offset %"PRIu64 + ", %s", offset, strerror (op_errno)); + goto out; + } + + memcpy (&prebuf, &bdatt->iatt, sizeof (struct iatt)); + bd_update_amtime (&bdatt->iatt, GF_SET_ATTR_MTIME); +out: + + BD_STACK_UNWIND (writev, frame, op_ret, op_errno, &prebuf, + &bdatt->iatt, NULL); + return 0; +} + +int +bd_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, + int op_errno, struct iatt *prebuf, struct iatt *postbuf, + dict_t *xdata) +{ + bd_attr_t *bdatt = NULL; + int *valid = cookie; + bd_local_t *local = frame->local; + + if (op_ret < 0 || !valid || !local) + goto out; + + if (bd_inode_ctx_get (local->inode, this, &bdatt)) + goto out; + + if (*valid & GF_SET_ATTR_UID) + bdatt->iatt.ia_uid = postbuf->ia_uid; + else if (*valid & GF_SET_ATTR_GID) + bdatt->iatt.ia_gid = postbuf->ia_gid; + else if (*valid & GF_SET_ATTR_MODE) { + bdatt->iatt.ia_type = postbuf->ia_type; + bdatt->iatt.ia_prot = postbuf->ia_prot; + } else if (*valid & GF_SET_ATTR_ATIME) { + bdatt->iatt.ia_atime = postbuf->ia_atime; + bdatt->iatt.ia_atime_nsec = postbuf->ia_atime_nsec; + } else if (*valid & GF_SET_ATTR_MTIME) { + bdatt->iatt.ia_mtime = postbuf->ia_mtime; + bdatt->iatt.ia_mtime_nsec = postbuf->ia_mtime_nsec; + } + + bdatt->iatt.ia_ctime = postbuf->ia_ctime; + bdatt->iatt.ia_ctime_nsec = postbuf->ia_ctime_nsec; + + memcpy (postbuf, &bdatt->iatt, sizeof (struct iatt)); +out: + FREE (valid); + BD_STACK_UNWIND (setattr, frame, op_ret, op_errno, prebuf, + postbuf, xdata); + return 0; +} + +int +bd_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf, + int32_t valid, dict_t *xdata) +{ + bd_local_t *local = NULL; + bd_attr_t *bdatt = NULL; + int *ck_valid = NULL; + int op_errno = 0; + + if (bd_inode_ctx_get (loc->inode, this, &bdatt)) { + STACK_WIND(frame, default_setattr_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->setattr, + loc, stbuf, valid, xdata); + return 0; + } + + local = bd_local_init (frame, this); + BD_VALIDATE_MEM_ALLOC (local, op_errno, out); + + ck_valid = CALLOC (1, sizeof (valid)); + BD_VALIDATE_MEM_ALLOC (ck_valid, op_errno, out); + + local->inode = inode_ref (loc->inode); + *ck_valid = valid; + + STACK_WIND_COOKIE (frame, bd_setattr_cbk, ck_valid, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->setattr, + loc, stbuf, valid, xdata); + + return 0; +out: + BD_STACK_UNWIND (setattr, frame, -1, ENOMEM, NULL, NULL, xdata); + return 0; +} + +int +bd_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, inode_t *inode, struct iatt *buf, + struct iatt *preparent, struct iatt *postparent, dict_t *xdata) +{ + bd_attr_t *bdatt = NULL; + + if (op_ret < 0) + goto out; + + if (bd_inode_ctx_get (inode, this, &bdatt)) + goto out; + + bdatt->iatt.ia_ctime = buf->ia_ctime; + bdatt->iatt.ia_ctime_nsec = buf->ia_ctime_nsec; + bdatt->iatt.ia_nlink = buf->ia_nlink; + memcpy (buf, &bdatt->iatt, sizeof (struct iatt)); + +out: + BD_STACK_UNWIND (link, frame, op_ret, op_errno, inode, buf, + preparent, postparent, NULL); + return 0; +} + +int +bd_link (call_frame_t *frame, xlator_t *this, + loc_t *oldloc, loc_t *newloc, dict_t *xdata) +{ + STACK_WIND (frame, bd_link_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata); + return 0; +} + +int +bd_handle_special_xattrs (call_frame_t *frame, xlator_t *this, loc_t *loc, + fd_t *fd, const char *name, dict_t *xdata) +{ + dict_t *xattr = NULL; + int op_ret = -1; + int op_errno = ENOMEM;; + bd_priv_t *priv = this->private; + + xattr = dict_new (); + if (!xattr) + goto out; + + if (!strcmp (name, VOL_TYPE)) + op_ret = dict_set_int64 (xattr, (char *)name, 1); + else if (!strcmp (name, VOL_CAPS)) + op_ret = dict_set_int64 (xattr, (char *)name, priv->caps); + else + op_ret = bd_get_origin (this->private, loc, fd, xattr); + +out: + if (loc) + BD_STACK_UNWIND (getxattr, frame, op_ret, op_errno, xattr, + xdata); + else + BD_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, xattr, + xdata); + + op_ret = dict_reset (xattr); + dict_unref (xattr); + + return 0; +} + +int +bd_fgetxattr (call_frame_t *frame, xlator_t *this, + fd_t *fd, const char *name, dict_t *xdata) +{ + if (name && (!strcmp (name, VOL_TYPE) || !strcmp (name, VOL_CAPS) + || !strcmp (name, BD_ORIGIN))) + bd_handle_special_xattrs (frame, this, NULL, fd, name, xdata); + else + STACK_WIND (frame, default_fgetxattr_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->fgetxattr, + fd, name, xdata); + return 0; +} + +int +bd_getxattr (call_frame_t *frame, xlator_t *this, + loc_t *loc, const char *name, dict_t *xdata) +{ + if (name && (!strcmp (name, VOL_TYPE) || !strcmp (name, VOL_CAPS) + || !strcmp (name, BD_ORIGIN))) + bd_handle_special_xattrs (frame, this, loc, NULL, name, xdata); + else + STACK_WIND (frame, default_getxattr_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->getxattr, + loc, name, xdata); + + return 0; +} + +int +bd_unlink_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, inode_t *inode, + struct iatt *buf, dict_t *xattr, + struct iatt *postparent) +{ + bd_gfid_t gfid = {0, }; + bd_local_t *local = frame->local; + + if (buf->ia_nlink > 1) + goto posix; + + BD_VALIDATE_LOCAL_OR_GOTO (local, op_errno, out); + + uuid_utoa_r (inode->gfid, gfid); + if (bd_delete_lv (this->private, gfid, &op_errno) < 0) { + if (op_errno != ENOENT) + goto out; + } + +posix: + /* remove posix */ + STACK_WIND (frame, default_unlink_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->unlink, + &local->loc, 0, NULL); + + return 0; +out: + BD_STACK_UNWIND (unlink, frame, -1, op_errno, NULL, NULL, NULL); + return 0; +} + +int +bd_unlink (call_frame_t *frame, xlator_t *this, + loc_t *loc, int xflag, dict_t *xdata) +{ + int op_errno = 0; + bd_attr_t *bdatt = NULL; + bd_local_t *local = NULL; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (loc, out); + + if (bd_inode_ctx_get (loc->inode, this, &bdatt)) { + STACK_WIND (frame, default_unlink_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->unlink, + loc, xflag, xdata); + return 0; + } + + local = bd_local_init (frame, this); + BD_VALIDATE_MEM_ALLOC (local, op_errno, out); + + loc_copy (&local->loc, loc); + + STACK_WIND (frame, bd_unlink_lookup_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->lookup, loc, NULL); + return 0; +out: + BD_STACK_UNWIND (unlink, frame, -1, op_errno, NULL, NULL, NULL); + return 0; +} + +int32_t +bd_priv (xlator_t *this) +{ + return 0; +} + +int32_t +bd_inode (xlator_t *this) +{ + return 0; +} + +int32_t +bd_rchecksum (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, + int32_t len, dict_t *xdata) +{ + int op_ret = -1; + int op_errno = 0; + int ret = 0; + int _fd = -1; + char *alloc_buf = NULL; + char *buf = NULL; + int32_t weak_checksum = 0; + bd_fd_t *bd_fd = NULL; + unsigned char strong_checksum[MD5_DIGEST_LENGTH] = {0}; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (fd, out); + + ret = bd_fd_ctx_get (this, fd, &bd_fd); + if (ret < 0 || !bd_fd) { + STACK_WIND (frame, default_rchecksum_cbk, FIRST_CHILD (this), + FIRST_CHILD (this)->fops->rchecksum, fd, offset, + len, xdata); + return 0; + } + + memset (strong_checksum, 0, MD5_DIGEST_LENGTH); + + alloc_buf = page_aligned_alloc (len, &buf); + if (!alloc_buf) { + op_errno = ENOMEM; + goto out; + } + + _fd = bd_fd->fd; + + LOCK (&fd->lock); + { + ret = pread (_fd, buf, len, offset); + if (ret < 0) { + gf_log (this->name, GF_LOG_WARNING, + "pread of %d bytes returned %d (%s)", + len, ret, strerror (errno)); + op_errno = errno; + } + } + UNLOCK (&fd->lock); + + if (ret < 0) + goto out; + + weak_checksum = gf_rsync_weak_checksum ((unsigned char *) buf, + (size_t) len); + gf_rsync_strong_checksum ((unsigned char *) buf, (size_t) len, + (unsigned char *) strong_checksum); + + op_ret = 0; +out: + BD_STACK_UNWIND (rchecksum, frame, op_ret, op_errno, + weak_checksum, strong_checksum, NULL); + + GF_FREE (alloc_buf); + + return 0; +} + +/** + * notify - when parent sends PARENT_UP, send CHILD_UP event from here + */ +int32_t +notify (xlator_t *this, + int32_t event, + void *data, + ...) +{ + switch (event) + { + case GF_EVENT_PARENT_UP: + { + /* Tell the parent that bd xlator is up */ + default_notify (this, GF_EVENT_CHILD_UP, data); + } + break; + default: + break; + } + return 0; +} + +int32_t +mem_acct_init (xlator_t *this) +{ + int ret = -1; + + if (!this) + return ret; + + ret = xlator_mem_acct_init (this, gf_bd_mt_end + 1); + + if (ret != 0) + gf_log (this->name, GF_LOG_ERROR, "Memory accounting init" + "failed"); + return ret; +} + +int +reconfigure (xlator_t *this, dict_t *options) +{ + int ret = -1; + bd_priv_t *priv = this->private; + + GF_OPTION_RECONF ("bd-aio", priv->aio_configured, options, + bool, out); + + if (priv->aio_configured) + bd_aio_on (this); + else + bd_aio_off (this); + + ret = 0; +out: + return ret; +} + +/** + * bd xlator init - Validate configured VG + */ +int +init (xlator_t *this) +{ + int ret = 0; + char *vg_data = NULL; + char *device = NULL; + bd_priv_t *_private = NULL; + + if (!this->children) { + gf_log (this->name, GF_LOG_CRITICAL, + "FATAL: storage/bd needs posix as subvolume"); + return -1; + } + + if (!this->parents) { + gf_log (this->name, GF_LOG_WARNING, + "Volume is dangling. Please check the volume file."); + } + + GF_OPTION_INIT ("export", vg_data, str, error); + GF_OPTION_INIT ("device", device, str, error); + + /* Now we support only LV device */ + if (strcasecmp (device, BACKEND_VG)) { + gf_log (this->name, GF_LOG_CRITICAL, + "FATAL: unknown %s backend %s", BD_XLATOR, device); + return -1; + } + + this->local_pool = mem_pool_new (bd_local_t, 64); + if (!this->local_pool) { + gf_log (this->name, GF_LOG_CRITICAL, + "FATAL: Failed to create bd memory pool"); + return -1; + } + + ret = 0; + _private = GF_CALLOC (1, sizeof (*_private), gf_bd_private); + if (!_private) + goto error; + + this->private = _private; + _private->vg = gf_strdup (vg_data); + if (!_private->vg) + goto error; + + _private->handle = lvm_init (NULL); + if (!_private->handle) { + gf_log (this->name, GF_LOG_CRITICAL, "lvm_init failed"); + goto error; + } + _private->caps = BD_CAPS_BD; + if (bd_scan_vg (this, _private)) + goto error; + + _private->aio_init_done = _gf_false; + _private->aio_capable = _gf_false; + + GF_OPTION_INIT ("bd-aio", _private->aio_configured, bool, error); + if (_private->aio_configured) { + if (bd_aio_on (this)) { + gf_log (this->name, GF_LOG_ERROR, + "BD AIO init failed"); + ret = -1; + goto error; + } + } + + _private->caps |= BD_CAPS_OFFLOAD_COPY | BD_CAPS_OFFLOAD_SNAPSHOT; + + return 0; +error: + GF_FREE (_private->vg); + if (_private->handle) + lvm_quit (_private->handle); + mem_pool_destroy (this->local_pool); + GF_FREE (_private); + return -1; +} + +void +fini (xlator_t *this) +{ + bd_priv_t *priv = this->private; + mem_pool_destroy (this->local_pool); + this->local_pool = NULL; + if (!priv) + return; + lvm_quit (priv->handle); + GF_FREE (priv->vg); + this->private = NULL; + GF_FREE (priv); + return; +} + +struct xlator_dumpops dumpops = { + .priv = bd_priv, + .inode = bd_inode, +}; + +struct xlator_fops fops = { + .readdirp = bd_readdirp, + .lookup = bd_lookup, + .stat = bd_stat, + .statfs = bd_statfs, + .open = bd_open, + .fstat = bd_fstat, + .rchecksum = bd_rchecksum, + .readv = bd_readv, + .fsync = bd_fsync, + .setxattr = bd_setxattr, + .fsetxattr = bd_fsetxattr, + .removexattr = bd_removexattr, + .fremovexattr=bd_fremovexattr, + .truncate = bd_truncate, + .ftruncate = bd_ftruncate, + .writev = bd_writev, + .getxattr = bd_getxattr, + .fgetxattr = bd_fgetxattr, + .unlink = bd_unlink, + .link = bd_link, + .flush = bd_flush, + .setattr = bd_setattr, + .discard = bd_discard, +}; + +struct xlator_cbks cbks = { + .release = bd_release, + .forget = bd_forget, +}; + +struct volume_options options[] = { + { .key = {"export"}, + .type = GF_OPTION_TYPE_STR}, + { .key = {"device"}, + .type = GF_OPTION_TYPE_STR, + .default_value = BACKEND_VG}, + { + .key = {"bd-aio"}, + .type = GF_OPTION_TYPE_BOOL, + .default_value = "off", + .description = "Support for native Linux AIO" + }, + + { .key = {NULL} } +}; diff --git a/xlators/storage/bd/src/bd.h b/xlators/storage/bd/src/bd.h new file mode 100644 index 000000000..34b4c9e22 --- /dev/null +++ b/xlators/storage/bd/src/bd.h @@ -0,0 +1,178 @@ +/* + BD translator - Exports Block devices on server side as regular + files to client + + Copyright IBM, Corp. 2012 + + This file is part of GlusterFS. + + Author: + M. Mohan Kumar <mohan@in.ibm.com> + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#ifndef _BD_H +#define _BD_H + +#ifndef _CONFIG_H +#define _CONFIG_H +#include "config.h" +#endif + +#ifdef HAVE_LIBAIO +#include <libaio.h> +#endif + +#include "xlator.h" +#include "mem-types.h" + +#define BD_XLATOR "block device mapper xlator" +#define BACKEND_VG "vg" +#define GF_XATTR "user.glusterfs" +#define BD_XATTR GF_XATTR ".bd" + +#define BD_LV "lv" +#define BD_THIN "thin" + +#define LVM_RESIZE "/sbin/lvresize" +#define LVM_CREATE "/sbin/lvcreate" +#define LVM_CONVERT "/sbin/lvconvert" + +#define VOL_TYPE "volume.type" +#define VOL_CAPS "volume.caps" + +#define ALIGN_SIZE 4096 + +#define BD_CAPS_BD 0x01 +#define BD_CAPS_THIN 0x02 +#define BD_CAPS_OFFLOAD_COPY 0x04 +#define BD_CAPS_OFFLOAD_SNAPSHOT 0x08 + +#define BD_CLONE "clone" +#define BD_SNAPSHOT "snapshot" +#define BD_MERGE "merge" +#define BD_ORIGIN "list-origin" + +#define IOV_NR 4 +#define IOV_SIZE (64 * 1024) + +#define ALIGN_SIZE 4096 + +#define LINKTO "trusted.glusterfs.dht.linkto" + +#define BD_VALIDATE_MEM_ALLOC(buff, op_errno, label) \ + if (!buff) { \ + op_errno = ENOMEM; \ + gf_log (this->name, GF_LOG_ERROR, "out of memory"); \ + goto label; \ + } + +#define BD_VALIDATE_LOCAL_OR_GOTO(local, op_errno, label) \ + if (!local) { \ + op_errno = EINVAL; \ + goto label; \ + } + +#define BD_STACK_UNWIND(typ, frame, args ...) do { \ + bd_local_t *__local = frame->local; \ + xlator_t *__this = frame->this; \ + \ + frame->local = NULL; \ + STACK_UNWIND_STRICT (typ, frame, args); \ + if (__local) \ + bd_local_free (__this, __local); \ + } while (0) + +typedef char bd_gfid_t[GF_UUID_BUF_SIZE]; + +enum gf_bd_mem_types_ { + gf_bd_private = gf_common_mt_end + 1, + gf_bd_attr, + gf_bd_fd, + gf_bd_mt_end +}; + +/** + * bd_fd - internal structure + */ +typedef struct bd_fd { + int fd; + int32_t flag; + int odirect; +} bd_fd_t; + +typedef struct bd_priv { + lvm_t handle; + char *vg; + char *pool; + int caps; + gf_boolean_t aio_init_done; + gf_boolean_t aio_capable; + gf_boolean_t aio_configured; +#ifdef HAVE_LIBAIO + io_context_t ctxp; + pthread_t aiothread; +#endif +} bd_priv_t; + + +typedef enum bd_type { + BD_TYPE_NONE, + BD_TYPE_LV, +} bd_type_t; + +typedef struct { + struct iatt iatt; + char *type; +} bd_attr_t; + +typedef enum { + BD_OF_NONE, + BD_OF_CLONE, + BD_OF_SNAPSHOT, + BD_OF_MERGE, +} bd_offload_t; + +typedef struct { + dict_t *dict; + bd_attr_t *bdatt; + inode_t *inode; + loc_t loc; + fd_t *fd; + data_t *data; /* for setxattr */ + bd_offload_t offload; + uint64_t size; + loc_t *dloc; +} bd_local_t; + +/* Prototypes */ +int bd_inode_ctx_set (inode_t *inode, xlator_t *this, bd_attr_t *ctx); +int bd_inode_ctx_get (inode_t *inode, xlator_t *this, bd_attr_t **ctx); +int bd_scan_vg (xlator_t *this, bd_priv_t *priv); +bd_local_t *bd_local_init (call_frame_t *frame, xlator_t *this); +void bd_local_free (xlator_t *this, bd_local_t *local); +int bd_fd_ctx_get (xlator_t *this, fd_t *fd, bd_fd_t **bdfd); +char *page_aligned_alloc (size_t size, char **aligned_buf); +int bd_validate_bd_xattr (xlator_t *this, char *bd, char **type, + uint64_t *lv_size, uuid_t uuid); +uint64_t bd_get_default_extent (bd_priv_t *priv); +uint64_t bd_adjust_size (bd_priv_t *priv, uint64_t size); +int bd_create (uuid_t uuid, uint64_t size, char *type, bd_priv_t *priv); +int bd_resize (bd_priv_t *priv, uuid_t uuid, off_t size); +int bd_delete_lv (bd_priv_t *priv, const char *lv_name, int *op_errno); +int bd_snapshot_create (bd_local_t *local, bd_priv_t *priv); +int bd_clone (bd_local_t *local, bd_priv_t *priv); + +int bd_merge (bd_priv_t *priv, uuid_t gfid); +int bd_get_origin (bd_priv_t *priv, loc_t *loc, fd_t *fd, dict_t *dict); +inline void bd_update_amtime(struct iatt *iatt, int flag); +int bd_snapshot_create (bd_local_t *local, bd_priv_t *priv); +int bd_clone (bd_local_t *local, bd_priv_t *priv); +int bd_merge (bd_priv_t *priv, uuid_t gfid); +int bd_get_origin (bd_priv_t *priv, loc_t *loc, fd_t *fd, dict_t *dict); + +#endif |